apptest: rename integration test to apptest

This PR fixes broken links or anchors in the Anomaly Detection pages:
- changelog
- anomaly-detection/UI
- Find a heading substitute to "Sampling Period"
- Correct "vlogs-reader" with "victorialogs-reader"

.github/workflows/build.yml

@@ -22,8 +22,7 @@ on:
       - '!app/vmui/**'
       - '.github/workflows/build.yml'
 
-permissions:
-  contents: read
+permissions: {}
 
 concurrency:
   cancel-in-progress: true
@@ -32,6 +31,8 @@ concurrency:
 jobs:
   build:
     name: ${{ matrix.os }}-${{ matrix.arch }}
+    permissions:
+      contents: read
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false

.github/workflows/changelog-linter.yml

@@ -5,8 +5,12 @@ on:
     paths:
       - "docs/victoriametrics/changelog/CHANGELOG.md"
 
+permissions: {}
+
 jobs:
   tip-lint:
+    permissions:
+      contents: read
     runs-on: 'ubuntu-latest'
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2

.github/workflows/check-commit-signed.yml

@@ -3,8 +3,12 @@ name: check-commit-signed
 on:
   pull_request:
 
+permissions: {}
+
 jobs:
   check-commit-signed:
+    permissions:
+      contents: read
     runs-on: ubuntu-latest
     steps:
       - name: Checkout code

.github/workflows/check-licenses.yml

@@ -6,12 +6,14 @@ on:
   pull_request:
     paths:
       - 'vendor'
-permissions:
-  contents: read
+
+permissions: {}
 
 jobs:
   build:
     name: Build
+    permissions:
+      contents: read
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout

.github/workflows/codeql-analysis-go.yml

@@ -18,6 +18,8 @@ concurrency:
   cancel-in-progress: true
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
 
+permissions: {}
+
 jobs:
   analyze:
     name: Analyze

.github/workflows/docs.yaml

@@ -7,12 +7,14 @@ on:
       - 'docs/**'
       - '.github/workflows/docs.yaml'
   workflow_dispatch: {}
-permissions:
-  contents: read  # This is required for actions/checkout and to commit back image update
-  deployments: write
+
+permissions: {}
+
 jobs:
   build:
     name: Build
+    permissions:
+      contents: read
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout

.github/workflows/test.yml

@@ -18,8 +18,7 @@ on:
       - 'go.*'
       - '.github/workflows/main.yml'
 
-permissions:
-  contents: read
+permissions: {}
 
 concurrency:
   cancel-in-progress: true
@@ -29,6 +28,8 @@ concurrency:
 jobs:
   lint:
     name: lint
+    permissions:
+      contents: read
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout
@@ -61,6 +62,8 @@ jobs:
 
   unit:
     name: unit
+    permissions:
+      contents: read
     runs-on: ubuntu-latest
 
     strategy:
@@ -90,6 +93,8 @@ jobs:
 
   apptest:
     name: apptest
+    permissions:
+      contents: read
     runs-on: apptest
 
     steps:

.github/workflows/vmui.yml

@@ -16,11 +16,7 @@ on:
       - 'app/vmui/packages/vmui/**'
       - '.github/workflows/vmui.yml'
 
-permissions:
-  contents: read
-  packages: read
-  pull-requests: read
-  checks: write
+permissions: {}
 
 concurrency:
   cancel-in-progress: true
@@ -29,6 +25,10 @@ concurrency:
 jobs:
   vmui-checks:
     name: VMUI Checks (lint, test, typecheck)
+    permissions:
+      checks: write
+      contents: read
+      pull-requests: read
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout

Makefile

@@ -471,7 +471,7 @@ test-full-386:
 
 apptest:
 	$(MAKE) victoria-metrics-race vmagent-race vmalert-race vmauth-race vmctl-race vmbackup-race vmrestore-race
-	go test ./apptest/... -skip="^Test(Cluster|Legacy).*"
+	VM_APPTEST=1 go test ./apptest/... -skip="^Test(Cluster|Legacy).*"
 
 apptest-legacy: victoria-metrics-race vmbackup-race vmrestore-race
 	OS=$$(uname | tr '[:upper:]' '[:lower:]'); \

app/victoria-metrics/main.go

@@ -34,8 +34,21 @@ var (
 		"This can be changed with -promscrape.config.strictParse=false command-line flag")
 	maxIngestionRate = flag.Int("maxIngestionRate", 0, "The maximum number of samples vmsingle can receive per second. Data ingestion is paused when the limit is exceeded. "+
 		"By default there are no limits on samples ingestion rate.")
+	vmselectMaxConcurrentRequests = flag.Int("search.maxConcurrentRequests", getDefaultMaxConcurrentRequests(), "The maximum number of concurrent search requests. "+
+		"It shouldn't be high, since a single request can saturate all the CPU cores, while many concurrently executed requests may require high amounts of memory. "+
+		"See also -search.maxQueueDuration and -search.maxMemoryPerQuery")
+	vmselectMaxQueueDuration = flag.Duration("search.maxQueueDuration", 10*time.Second, "The maximum time the request waits for execution when -search.maxConcurrentRequests "+
+		"limit is reached; see also -search.maxQueryDuration")
 )
 
+func getDefaultMaxConcurrentRequests() int {
+	// A single request can saturate all the CPU cores, so there is no sense
+	// in allowing higher number of concurrent requests - they will just contend
+	// for unavailable CPU time.
+	n := min(cgroup.AvailableCPUs()*2, 16)
+	return n
+}
+
 func main() {
 	// VictoriaMetrics is optimized for reduced memory allocations,
 	// so it can run with the reduced GOGC in order to reduce the used memory,
@@ -76,8 +89,8 @@ func main() {
 	}
 	logger.Infof("starting VictoriaMetrics at %q...", listenAddrs)
 	startTime := time.Now()
-	vmstorage.Init(promql.ResetRollupResultCacheIfNeeded)
-	vmselect.Init()
+	vmstorage.Init(*vmselectMaxConcurrentRequests, promql.ResetRollupResultCacheIfNeeded)
+	vmselect.Init(*vmselectMaxConcurrentRequests, *vmselectMaxQueueDuration)
 	vminsertcommon.StartIngestionRateLimiter(*maxIngestionRate)
 	vminsert.Init()
 

app/victoria-metrics/self_scraper.go

@@ -93,7 +93,7 @@ func selfScraper(scrapeInterval time.Duration) {
 				mr.Value = r.Value
 			}
 		}
-		if err := vmstorage.AddRows(mrs); err != nil {
+		if err := vmstorage.VMInsertAPI.WriteRows(mrs); err != nil {
 			logger.Errorf("cannot store self-scraped metrics: %s", err)
 		}
 		if len(metadataRows.Rows) > 0 {
@@ -105,7 +105,7 @@ func selfScraper(scrapeInterval time.Duration) {
 					Type:             mm.Type,
 				})
 			}
-			if err := vmstorage.AddMetadataRows(mms); err != nil {
+			if err := vmstorage.VMInsertAPI.WriteMetadata(mms); err != nil {
 				logger.Errorf("cannot store self-scraped metrics metadata: %s", err)
 			}
 		}

app/vmagent/remotewrite/remotewrite.go

@@ -79,7 +79,8 @@ var (
 		"writing them to remote storage. "+
 		"Examples: -remoteWrite.roundDigits=2 would round 1.236 to 1.24, while -remoteWrite.roundDigits=-1 would round 126.78 to 130. "+
 		"By default, digits rounding is disabled. Set it to 100 for disabling it for a particular remote storage. "+
-		"This option may be used for improving data compression for the stored metrics")
+		"This option may be used for improving data compression for the stored metrics. "+
+		"See also -remoteWrite.significantFigures")
 	sortLabels = flag.Bool("sortLabels", false, `Whether to sort labels for incoming samples before writing them to all the configured remote storage systems. `+
 		`This may be needed for reducing memory usage at remote storage when the order of labels in incoming samples is random. `+
 		`For example, if m{k1="v1",k2="v2"} may be sent as m{k2="v2",k1="v1"}`+

app/vmalert-tool/unittest/input.go

@@ -52,7 +52,7 @@ func writeInputSeries(input []series, interval *promutil.Duration, startStamp ti
 	data := testutil.Compress(r)
 	// write input series to vm
 	httpWrite(dst, bytes.NewBuffer(data))
-	vmstorage.Storage.DebugFlush()
+	vmstorage.DebugFlush()
 	return nil
 }
 

app/vmalert-tool/unittest/unittest.go

@@ -108,7 +108,9 @@ func UnitTest(files []string, disableGroupLabel bool, externalLabels []string, e
 	storagePath = tmpFolder
 	processFlags()
 	vminsert.Init()
-	vmselect.Init()
+	const maxConcurrentRequests = 4
+	maxQueueDuration := 5 * time.Second
+	vmselect.Init(maxConcurrentRequests, maxQueueDuration)
 	// storagePath will be created again when closing vmselect, so remove it again.
 	defer fs.MustRemoveDir(storagePath)
 	defer vminsert.Stop()
@@ -279,7 +281,8 @@ func processFlags() {
 }
 
 func setUp() {
-	vmstorage.Init(promql.ResetRollupResultCacheIfNeeded)
+	const maxConcurrentRequests = 4
+	vmstorage.Init(maxConcurrentRequests, promql.ResetRollupResultCacheIfNeeded)
 	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 	defer cancel()
 	readyCheckFunc := func() bool {
@@ -384,7 +387,7 @@ func (tg *testGroup) test(evalInterval time.Duration, groupOrderMap map[string]i
 				}
 			}
 			// flush series after each group evaluation
-			vmstorage.Storage.DebugFlush()
+			vmstorage.DebugFlush()
 		}
 
 		// check alert_rule_test case at every eval time

app/vmalert/rule/group.go

@@ -95,6 +95,7 @@ type groupMetrics struct {
 	iterationTotal    *metrics.Counter
 	iterationDuration *metrics.Summary
 	iterationMissed   *metrics.Counter
+	iterationReset    *metrics.Counter
 	iterationInterval *metrics.Gauge
 }
 
@@ -330,6 +331,7 @@ func (g *Group) Init() {
 	g.metrics.iterationTotal = g.metrics.set.NewCounter(fmt.Sprintf(`vmalert_iteration_total{%s}`, labels))
 	g.metrics.iterationDuration = g.metrics.set.NewSummary(fmt.Sprintf(`vmalert_iteration_duration_seconds{%s}`, labels))
 	g.metrics.iterationMissed = g.metrics.set.NewCounter(fmt.Sprintf(`vmalert_iteration_missed_total{%s}`, labels))
+	g.metrics.iterationReset = g.metrics.set.NewCounter(fmt.Sprintf(`vmalert_iteration_reset_total{%s}`, labels))
 	g.metrics.iterationInterval = g.metrics.set.NewGauge(fmt.Sprintf(`vmalert_iteration_interval_seconds{%s}`, labels), func() float64 {
 		i := g.Interval.Seconds()
 		return i
@@ -474,14 +476,16 @@ func (g *Group) Start(ctx context.Context, rw remotewrite.RWClient, rr datasourc
 			if missed < 0 {
 				// missed can become < 0 due to irregular delays during evaluation
 				// which can result in time.Since(evalTS) < g.Interval;
-				// or the system wall clock was changed backward
-				missed = 0
+				// or the system wall clock was changed backward,
+				// Reset the evalTS to the current time.
 				evalTS = time.Now()
+				g.metrics.iterationReset.Inc()
+			} else {
+				evalTS = evalTS.Add((missed + 1) * g.Interval)
 			}
 			if missed > 0 {
 				g.metrics.iterationMissed.Inc()
 			}
-			evalTS = evalTS.Add((missed + 1) * g.Interval)
 
 			eval(evalCtx, evalTS)
 		}

app/vmalert/web.go

@@ -11,6 +11,8 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/VictoriaMetrics/metricsql"
+
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/rule"
@@ -160,12 +162,12 @@ func (rh *requestHandler) handler(w http.ResponseWriter, r *http.Request) bool {
 
 	case "/vmalert/api/v1/alerts", "/api/v1/alerts":
 		// path used by Grafana for ng alerting
-		gf, err := newGroupsFilter(r)
+		af, err := newAlertsFilter(r)
 		if err != nil {
 			errJson(w, r, err)
 			return true
 		}
-		data, err := rh.listAlerts(gf)
+		data, err := rh.listAlerts(af)
 		if err != nil {
 			errJson(w, r, err)
 			return true
@@ -325,6 +327,48 @@ func (gf *groupsFilter) matches(group *rule.Group) bool {
 	return true
 }
 
+type alertsFilter struct {
+	gf    *groupsFilter
+	match [][]metricsql.LabelFilter
+}
+
+func getMatchFilters(matches []string) ([][]metricsql.LabelFilter, *httpserver.ErrorWithStatusCode) {
+	if len(matches) == 0 {
+		return nil, nil
+	}
+	tfss := make([][]metricsql.LabelFilter, 0, len(matches))
+	for _, s := range matches {
+		expr, err := metricsql.Parse(s)
+		if err != nil {
+			return nil, errResponse(fmt.Errorf(`invalid parameter "match[]": failed to parse %q: %w`, s, err), http.StatusBadRequest)
+		}
+		me, ok := expr.(*metricsql.MetricExpr)
+		if !ok {
+			return nil, errResponse(fmt.Errorf(`invalid parameter "match[]": expecting metricSelector; got %q`, expr.AppendString(nil)), http.StatusBadRequest)
+		}
+		if len(me.LabelFilterss) == 0 {
+			return nil, errResponse(fmt.Errorf(`invalid parameter "match[]": labelFilterss cannot be empty`), http.StatusBadRequest)
+		}
+		tfss = append(tfss, me.LabelFilterss...)
+	}
+	return tfss, nil
+}
+
+func newAlertsFilter(r *http.Request) (*alertsFilter, *httpserver.ErrorWithStatusCode) {
+	gf, err := newGroupsFilter(r)
+	if err != nil {
+		return nil, err
+	}
+
+	var af alertsFilter
+	af.gf = gf
+	af.match, err = getMatchFilters(r.Form["match[]"])
+	if err != nil {
+		return nil, err
+	}
+	return &af, nil
+}
+
 // see https://prometheus.io/docs/prometheus/latest/querying/api/#rules
 type rulesFilter struct {
 	gf             *groupsFilter
@@ -335,6 +379,7 @@ type rulesFilter struct {
 	maxGroups      int
 	pageNum        int
 	search         string
+	match          [][]metricsql.LabelFilter
 	extendedStates bool
 }
 
@@ -355,7 +400,10 @@ func newRulesFilter(r *http.Request) (*rulesFilter, *httpserver.ErrorWithStatusC
 			return nil, errResponse(fmt.Errorf(`invalid parameter "type": not supported value %q`, ruleTypeParam), http.StatusBadRequest)
 		}
 	}
-
+	rf.match, err = getMatchFilters(r.Form["match[]"])
+	if err != nil {
+		return nil, err
+	}
 	states := vs["state"]
 	if len(states) == 0 {
 		states = vs["filter"]
@@ -416,12 +464,47 @@ func (rf *rulesFilter) matchesRule(r *rule.ApiRule) bool {
 	if len(rf.ruleNames) > 0 && !slices.Contains(rf.ruleNames, r.Name) {
 		return false
 	}
+	if !areLabelsMatch(r.Labels, rf.match) {
+		return false
+	}
 	if len(rf.states) == 0 {
 		return true
 	}
 	return slices.Contains(rf.states, r.State)
 }
 
+func areLabelsMatch(labels map[string]string, matches [][]metricsql.LabelFilter) bool {
+	if len(matches) == 0 {
+		return true
+	}
+	// labels need to match at least one of the provided match[] arg
+	return slices.ContainsFunc(matches, func(filters []metricsql.LabelFilter) bool {
+		for _, mf := range filters {
+			if !isLabelFilterMatch(labels[mf.Label], mf) {
+				return false
+			}
+		}
+		return true
+	})
+}
+
+func isLabelFilterMatch(s string, match metricsql.LabelFilter) bool {
+	if !match.IsRegexp {
+		if match.IsNegative {
+			return s != match.Value
+		}
+		return s == match.Value
+	}
+	re, err := metricsql.CompileRegexpAnchored(match.Value)
+	if err != nil {
+		return false
+	}
+	if match.IsNegative {
+		return !re.MatchString(s)
+	}
+	return re.MatchString(s)
+}
+
 func (rh *requestHandler) groups(rf *rulesFilter) *listGroupsResponse {
 	rh.m.groupsMu.RLock()
 	defer rh.m.groupsMu.RUnlock()
@@ -543,14 +626,14 @@ func (rh *requestHandler) groupAlerts() []rule.GroupAlerts {
 	return gAlerts
 }
 
-func (rh *requestHandler) listAlerts(gf *groupsFilter) ([]byte, *httpserver.ErrorWithStatusCode) {
+func (rh *requestHandler) listAlerts(af *alertsFilter) ([]byte, *httpserver.ErrorWithStatusCode) {
 	rh.m.groupsMu.RLock()
 	defer rh.m.groupsMu.RUnlock()
 
 	lr := listAlertsResponse{Status: "success"}
 	lr.Data.Alerts = make([]*rule.ApiAlert, 0)
 	for _, group := range rh.m.groups {
-		if !gf.matches(group) {
+		if !af.gf.matches(group) {
 			continue
 		}
 		g := group.ToAPI()
@@ -558,7 +641,11 @@ func (rh *requestHandler) listAlerts(gf *groupsFilter) ([]byte, *httpserver.Erro
 			if r.Type != rule.TypeAlerting {
 				continue
 			}
-			lr.Data.Alerts = append(lr.Data.Alerts, r.Alerts...)
+			for _, alert := range r.Alerts {
+				if areLabelsMatch(alert.Labels, af.match) {
+					lr.Data.Alerts = append(lr.Data.Alerts, alert)
+				}
+			}
 		}
 	}
 

app/vmalert/web.qtpl

@@ -348,7 +348,7 @@
                 typeK, ns := keys[i], targets[notifier.TargetType(keys[i])]
                 count := len(ns)
             %}
-            <div class="w-100 flex-column vm-group">
+            <div class="w-100 flex-column">
                 <span class="d-flex justify-content-between" id="group-{%s typeK %}">
                     <a href="#group-{%s typeK %}">{%s typeK %} ({%d count %})</a>
                     <span
@@ -361,7 +361,7 @@
                 <div id="item-{%s typeK %}" class="collapse show">
                     <table class="table table-striped table-hover table-sm">
                         <thead>
-                            <tr class="vm-item">
+                            <tr>
                                 <th scope="col">Labels</th>
                                 <th scope="col">Address</th>
                             </tr>

app/vmalert/web.qtpl.go

@@ -1115,7 +1115,7 @@ func StreamListTargets(qw422016 *qt422016.Writer, r *http.Request, targets map[n
 
 //line app/vmalert/web.qtpl:350
 			qw422016.N().S(`
-            <div class="w-100 flex-column vm-group">
+            <div class="w-100 flex-column">
                 <span class="d-flex justify-content-between" id="group-`)
 //line app/vmalert/web.qtpl:352
 			qw422016.E().S(typeK)
@@ -1152,7 +1152,7 @@ func StreamListTargets(qw422016 *qt422016.Writer, r *http.Request, targets map[n
 			qw422016.N().S(`" class="collapse show">
                     <table class="table table-striped table-hover table-sm">
                         <thead>
-                            <tr class="vm-item">
+                            <tr>
                                 <th scope="col">Labels</th>
                                 <th scope="col">Address</th>
                             </tr>

app/vmalert/web_test.go

@@ -10,6 +10,8 @@ import (
 	"testing"
 	"time"
 
+	"github.com/VictoriaMetrics/metricsql"
+
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
@@ -37,12 +39,14 @@ func TestHandler(t *testing.T) {
 			Concurrency: 1,
 			Rules: []config.Rule{
 				{
-					ID:    0,
-					Alert: "alert",
+					ID:     0,
+					Alert:  "alert",
+					Labels: map[string]string{"job": "foo"},
 				},
 				{
 					ID:     1,
 					Record: "record",
+					Labels: map[string]string{"job": "bar"},
 				},
 			},
 		}, fq, 1*time.Minute, nil)
@@ -128,6 +132,18 @@ func TestHandler(t *testing.T) {
 		if length := len(lr.Data.Alerts); length != 2 {
 			t.Fatalf("expected 2 alert got %d", length)
 		}
+
+		lr = listAlertsResponse{}
+		getResp(t, ts.URL+`/api/v1/alerts?match[]={job="foo"}`, &lr, 200)
+		if length := len(lr.Data.Alerts); length != 3 {
+			t.Fatalf("expected 3 alerts got %d", length)
+		}
+
+		lr = listAlertsResponse{}
+		getResp(t, ts.URL+`/api/v1/alerts?match[]={job="bar"}`, &lr, 200)
+		if length := len(lr.Data.Alerts); length != 0 {
+			t.Fatalf("expected 0 alerts got %d", length)
+		}
 	})
 	t.Run("/api/v1/alert?alertID&groupID", func(t *testing.T) {
 		expAlert := rule.NewAlertAPI(ar, ar.GetAlerts()[0])
@@ -242,6 +258,13 @@ func TestHandler(t *testing.T) {
 		check("/vmalert/api/v1/rules?datasource_type=graphite", 200, 1, 2)
 		check("/vmalert/api/v1/rules?datasource_type=graphiti", 400, 0, 0)
 
+		// invalid match[] params
+		check(`/vmalert/api/v1/rules?match[]={job=!"foo"}`, 400, 0, 0)
+		check(`/vmalert/api/v1/rules?match[]={job="foo"}`, 200, 3, 3)
+		check(`/vmalert/api/v1/rules?match[]={job="bar"}`, 200, 3, 3)
+		check(`/vmalert/api/v1/rules?match[]={job="bar"}&match[]={job="foo"}`, 200, 3, 6)
+		check(`/vmalert/api/v1/rules?match[]={job="barzz"}`, 200, 0, 0)
+
 		// no filtering expected due to bad params
 		check("/api/v1/rules?type=badParam", 400, 0, 0)
 		check("/api/v1/rules?foo=bar", 200, 3, 6)
@@ -367,3 +390,116 @@ func TestEmptyResponse(t *testing.T) {
 		}
 	})
 }
+
+func TestMatchesRule(t *testing.T) {
+	parseMatch := func(t *testing.T, selectors []string) [][]metricsql.LabelFilter {
+		t.Helper()
+		var match [][]metricsql.LabelFilter
+		for _, s := range selectors {
+			expr, err := metricsql.Parse(s)
+			if err != nil {
+				t.Fatalf("failed to parse selector %q: %v", s, err)
+			}
+			me, ok := expr.(*metricsql.MetricExpr)
+			if !ok {
+				t.Fatalf("expected MetricExpr for %q, got %T", s, expr)
+			}
+			match = append(match, me.LabelFilterss...)
+		}
+		return match
+	}
+
+	f := func(t *testing.T, selectors []string, labels map[string]string, wantMatch bool) {
+		t.Helper()
+		rf := &rulesFilter{
+			gf:    &groupsFilter{},
+			match: parseMatch(t, selectors),
+		}
+		r := &rule.ApiRule{Labels: labels}
+		got := rf.matchesRule(r)
+		if got != wantMatch {
+			t.Fatalf("matchesRule(%v) with selectors %v: got %v, want %v",
+				labels, selectors, got, wantMatch)
+		}
+	}
+
+	f(t, nil, map[string]string{"foo": "bar"}, true)
+
+	f(t, []string{`{foo="bar"}`}, map[string]string{"foo": "bar"}, true)
+	f(t, []string{`{foo="bar"}`}, map[string]string{"foo": "baz"}, false)
+
+	f(t, []string{`{foo="bar"}`}, map[string]string{"bar": "baz"}, false)
+	f(t, []string{`{foo=""}`}, map[string]string{"bar": "baz"}, true)
+
+	f(t, []string{`{foo!="bar"}`}, map[string]string{"foo": "baz"}, true)
+	f(t, []string{`{foo!="bar"}`}, map[string]string{"foo": "bar"}, false)
+
+	f(t, []string{`{foo=~"bar.*"}`}, map[string]string{"foo": "bar"}, true)
+	f(t, []string{`{foo=~"bar.*"}`}, map[string]string{"foo": "baz"}, false)
+	f(t, []string{`{bar=~"baz|bar"}`}, map[string]string{"bar": "baz"}, true)
+	f(t, []string{`{bar=~"baz|bar"}`}, map[string]string{"bar": "bar"}, true)
+	f(t, []string{`{bar=~"baz|bar"}`}, map[string]string{"bar": "foo"}, false)
+
+	f(t, []string{`{foo!~"bar.*"}`}, map[string]string{"foo": "baz"}, true)
+	f(t, []string{`{foo!~"bar.*"}`}, map[string]string{"foo": "bar"}, false)
+
+	// single match[] with multiple filters
+	f(t,
+		[]string{`{job="foo",instance="bar"}`},
+		map[string]string{"job": "foo", "instance": "bar"},
+		true,
+	)
+	f(t,
+		[]string{`{job="foo",instance="bar"}`},
+		map[string]string{"job": "other", "instance": "bar"},
+		false,
+	)
+
+	f(t,
+		[]string{`{foo="bar",baz=~"b.*"}`},
+		map[string]string{"foo": "bar", "baz": "bazinga"},
+		true,
+	)
+	f(t,
+		[]string{`{foo="bar",baz=~"b.*"}`},
+		map[string]string{"foo": "other", "baz": "bazinga"},
+		false,
+	)
+
+	// multiple matches[]
+	f(t,
+		[]string{`{foo="bar"}`, `{foo="baz"}`},
+		map[string]string{"foo": "baz"},
+		true,
+	)
+	f(t,
+		[]string{`{foo="bar"}`, `{foo="baz"}`},
+		map[string]string{"foo": "unknown"},
+		false,
+	)
+	f(t,
+		[]string{`{foo=~"bar.*"}`, `{bar=~"baz.*"}`},
+		map[string]string{"bar": "bazinga"},
+		true,
+	)
+	f(t,
+		[]string{`{foo=~"bar.*"}`, `{bar=~"baz.*"}`},
+		map[string]string{"foo": "bartender"},
+		true,
+	)
+	f(t,
+		[]string{`{foo=~"bar.*"}`, `{bar=~"baz.*"}`},
+		map[string]string{"foo": "other", "bar": "other"},
+		false,
+	)
+	f(t,
+		[]string{`{job="foo",instance="bar"}`, `{foo="bar"}`},
+		map[string]string{"foo": "bar"},
+		true,
+	)
+	f(t,
+		[]string{`{job="foo", instance="bar"}`, `{foo="bar"}`},
+		map[string]string{"instance": "barr", "job": "foo"},
+		false,
+	)
+}

app/vmauth/auth_config.go

@@ -889,7 +889,8 @@ func reloadAuthConfig() (bool, error) {
 	}
 
 	mp := authUsers.Load()
-	logger.Infof("loaded information about %d users from -auth.config=%q", len(*mp), *authConfigPath)
+	jwtc := jwtAuthCache.Load()
+	logger.Infof("loaded information about %d users from -auth.config=%q", len(*mp)+len(jwtc.users), *authConfigPath)
 	return true, nil
 }
 

app/vmauth/main.go

@@ -317,7 +317,7 @@ func processUserRequest(w http.ResponseWriter, r *http.Request, ui *UserInfo, tk
 	defer ui.endConcurrencyLimit()
 
 	// Process the request.
-	processRequest(w, r, ui, tkn)
+	processRequest(w, r, ui, tkn, userName)
 }
 
 func beginConcurrencyLimit(ctx context.Context) error {
@@ -391,7 +391,7 @@ func bufferRequestBody(ctx context.Context, r io.ReadCloser, userName string) (i
 	return bb, nil
 }
 
-func processRequest(w http.ResponseWriter, r *http.Request, ui *UserInfo, tkn *jwt.Token) {
+func processRequest(w http.ResponseWriter, r *http.Request, ui *UserInfo, tkn *jwt.Token, userName string) {
 	u := normalizeURL(r.URL)
 	up, hc := ui.getURLPrefixAndHeaders(u, r.Host, r.Header)
 	isDefault := false
@@ -409,7 +409,7 @@ func processRequest(w http.ResponseWriter, r *http.Request, ui *UserInfo, tkn *j
 			if ui.DumpRequestOnErrors {
 				di = debugInfo(u, r)
 			}
-			httpserver.Errorf(w, r, "missing route for %q%s", u.String(), di)
+			httpserver.Errorf(w, r, "user %s missing route for %q%s", userName, u.String(), di)
 			return
 		}
 		up, hc = ui.DefaultURL, ui.HeadersConf
@@ -455,7 +455,7 @@ func processRequest(w http.ResponseWriter, r *http.Request, ui *UserInfo, tkn *j
 		ui.backendErrors.Inc()
 	}
 	err := &httpserver.ErrorWithStatusCode{
-		Err:        fmt.Errorf("all the %d backends for the user %q are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend", up.getBackendsCount(), ui.name()),
+		Err:        fmt.Errorf("all the %d backends for the user %q are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend", up.getBackendsCount(), userName),
 		StatusCode: http.StatusBadGateway,
 	}
 	httpserver.Errorf(w, r, "%s", err)

app/vmauth/main_test.go

@@ -307,6 +307,24 @@ statusCode=200
 requested_url={BACKEND}/bar/a/b`
 	f(cfgStr, requestURL, backendHandler, responseExpected)
 
+	// correct authorization but unexisted path, hence missing route error.
+	cfgStr = `
+users:
+- username: foo
+  password: secret
+  url_map:
+  - src_paths:
+    - "/api/v1/write"
+    url_prefix: "{BACKEND}/bar"`
+	requestURL = "http://foo:secret@some-host.com/a/b"
+	backendHandler = func(w http.ResponseWriter, r *http.Request) {
+		fmt.Fprintf(w, "requested_url=http://%s%s", r.Host, r.URL)
+	}
+	responseExpected = `
+statusCode=400
+user foo missing route for "http://foo:secret@some-host.com/a/b"`
+	f(cfgStr, requestURL, backendHandler, responseExpected)
+
 	// verify how path cleanup works
 	cfgStr = `
 unauthorized_user:
@@ -403,7 +421,7 @@ unauthorized_user:
 	}
 	responseExpected = `
 statusCode=400
-missing route for "http://some-host.com/abc?de=fg"`
+user unauthorized missing route for "http://some-host.com/abc?de=fg"`
 	f(cfgStr, requestURL, backendHandler, responseExpected)
 
 	// missing default_url and default url_prefix for unauthorized user with dump_request_on_errors enabled
@@ -419,7 +437,7 @@ unauthorized_user:
 	}
 	responseExpected = `
 statusCode=400
-missing route for "http://some-host.com/abc?de=fg" (host: "some-host.com"; path: "/abc"; args: "de=fg"; headers:Connection: Some-Header,Other-Header
+user unauthorized missing route for "http://some-host.com/abc?de=fg" (host: "some-host.com"; path: "/abc"; args: "de=fg"; headers:Connection: Some-Header,Other-Header
 Pass-Header: abc
 Some-Header: foobar
 X-Forwarded-For: 12.34.56.78
@@ -461,7 +479,7 @@ unauthorized_user:
 	}
 	responseExpected = `
 statusCode=502
-all the 2 backends for the user "" are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend`
+all the 2 backends for the user "unauthorized" are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend`
 	f(cfgStr, requestURL, backendHandler, responseExpected)
 
 	// all the backend_urls are unavailable for authorized user
@@ -501,7 +519,7 @@ unauthorized_user:
 	}
 	responseExpected = `
 statusCode=502
-all the 0 backends for the user "" are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend`
+all the 0 backends for the user "unauthorized" are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend`
 	f(cfgStr, requestURL, backendHandler, responseExpected)
 	netutil.Resolver = origResolver
 
@@ -518,7 +536,7 @@ unauthorized_user:
 	}
 	responseExpected = `
 statusCode=502
-all the 2 backends for the user "" are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend`
+all the 2 backends for the user "unauthorized" are unavailable for proxying the request - check previous WARN logs to see the exact error for each failed backend`
 	f(cfgStr, requestURL, backendHandler, responseExpected)
 	if n := retries.Load(); n != 2 {
 		t.Fatalf("unexpected number of retries; got %d; want 2", n)
@@ -545,6 +563,31 @@ requested_url={BACKEND}/path2/foo/?de=fg`
 	if n := retries.Load(); n != 2 {
 		t.Fatalf("unexpected number of retries; got %d; want 2", n)
 	}
+
+	// make sure that empty config value erases client extra filters and extra labels
+	cfgStr = `
+unauthorized_user:
+  url_prefix: {BACKEND}/foo?bar=baz&extra_filters[]=&extra_label=&extra_filters=`
+	requestURL = "http://some-host.com/abc/def?some_arg=some_value&extra_filters[]=baz&extra_label=tenant=admin&extra_filters=bar"
+	backendHandler = func(w http.ResponseWriter, r *http.Request) {
+		h := w.Header()
+		h.Set("Connection", "close")
+		h.Set("Foo", "bar")
+
+		var bb bytes.Buffer
+		if err := r.Header.Write(&bb); err != nil {
+			panic(fmt.Errorf("unexpected error when marshaling headers: %w", err))
+		}
+		fmt.Fprintf(w, "requested_url=http://%s%s\n%s", r.Host, r.URL, bb.String())
+	}
+	responseExpected = `
+statusCode=200
+Foo: bar
+requested_url={BACKEND}/foo/abc/def?bar=baz&extra_filters=&extra_filters%5B%5D=&extra_label=&some_arg=some_value
+Pass-Header: abc
+User-Agent: vmauth
+X-Forwarded-For: 12.34.56.78, 42.2.3.84`
+	f(cfgStr, requestURL, backendHandler, responseExpected)
 }
 
 func TestJWTRequestHandler(t *testing.T) {

app/vmctl/flags.go

@@ -146,7 +146,8 @@ var (
 			Name:  vmRoundDigits,
 			Value: 100,
 			Usage: "Round metric values to the given number of decimal digits after the point. " +
-				"This option may be used for increasing on-disk compression level for the stored metrics",
+				"This option may be used for increasing on-disk compression level for the stored metrics. " +
+				"See also --vm-significant-figures option",
 		},
 		&cli.StringSliceFlag{
 			Name:  vmExtraLabel,
@@ -500,6 +501,96 @@ var (
 	}
 )
 
+const (
+	mimirPath             = "mimir-path"
+	mimirTenantID         = "mimir-tenant-id"
+	mimirConcurrency      = "mimir-concurrency"
+	mimirFilterTimeStart  = "mimir-filter-time-start"
+	mimirFilterTimeEnd    = "mimir-filter-time-end"
+	mimirFilterLabel      = "mimir-filter-label"
+	mimirFilterLabelValue = "mimir-filter-label-value"
+
+	mimirCredsFilePath           = "mimir-creds-file-path"
+	mimirConfigFilePath          = "mimir-config-file-path"
+	mimirConfigProfile           = "mimir-config-profile"
+	mimirCustomS3Endpoint        = "mimir-custom-s3-endpoint"
+	mimirS3ForcePathStyle        = "mimir-s3-force-path-style"
+	mimirS3TLSInsecureSkipVerify = "mimir-s3-tls-insecure-skip-verify"
+	mimirSSEKMSKeyID             = "mimir-s3-sse-kms-key-id"
+	mimirSSEAlgorithm            = "mimir-s3-sse-algorithm"
+)
+
+var (
+	mimirFlags = []cli.Flag{
+		&cli.StringFlag{
+			Name:     mimirPath,
+			Usage:    "Path to Mimir storage bucket or local folder.",
+			Required: true,
+		},
+		&cli.StringFlag{
+			Name:  mimirTenantID,
+			Usage: "Tenant ID for Mimir storage",
+		},
+		&cli.IntFlag{
+			Name:  mimirConcurrency,
+			Usage: "Number of concurrently running block readers",
+			Value: 1,
+		},
+		&cli.StringFlag{
+			Name:     mimirFilterTimeStart,
+			Usage:    "The time filter in RFC3339 format to select timeseries with timestamp equal or higher than provided value. E.g. '2020-01-01T20:07:00Z'",
+			Required: true,
+		},
+		&cli.StringFlag{
+			Name:     mimirFilterTimeEnd,
+			Usage:    "The time filter in RFC3339 format to select timeseries with timestamp equal or lower than provided value. E.g. '2020-01-01T20:07:00Z'",
+			Required: true,
+		},
+		&cli.StringFlag{
+			Name:  mimirFilterLabel,
+			Usage: "Mimir label name to filter timeseries by. E.g. '__name__' will filter timeseries by name.",
+		},
+		&cli.StringFlag{
+			Name:  mimirFilterLabelValue,
+			Usage: fmt.Sprintf("Regular expression to filter label from %q flag.", mimirFilterLabel),
+			Value: ".*",
+		},
+		&cli.StringFlag{
+			Name:  mimirCredsFilePath,
+			Usage: "Path to file with GCS or S3 credentials. Credentials are loaded from default locations if not set. See https://cloud.google.com/iam/docs/creating-managing-service-account-keys and https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html",
+		},
+		&cli.StringFlag{
+			Name:  mimirConfigFilePath,
+			Usage: "Path to file with S3 configs. Configs are loaded from default location if not set. See https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html",
+		},
+		&cli.StringFlag{
+			Name:  mimirConfigProfile,
+			Usage: "Profile name for S3 configs. If no set, the value of the environment variable will be loaded (AWS_PROFILE or AWS_DEFAULT_PROFILE), or if both not set, DefaultSharedConfigProfile is used",
+		},
+		&cli.StringFlag{
+			Name:  mimirCustomS3Endpoint,
+			Usage: "Custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set",
+		},
+		&cli.BoolFlag{
+			Name:  mimirS3ForcePathStyle,
+			Usage: "Prefixing endpoint with bucket name when set false, true by default.",
+			Value: true,
+		},
+		&cli.BoolFlag{
+			Name:  mimirS3TLSInsecureSkipVerify,
+			Usage: "Whether to skip TLS verification when connecting to the S3 endpoint.",
+		},
+		&cli.StringFlag{
+			Name:  mimirSSEKMSKeyID,
+			Usage: "SSE KMS Key ID for use with S3-compatible storages.",
+		},
+		&cli.StringFlag{
+			Name:  mimirSSEAlgorithm,
+			Usage: "SSE algorithm for use with S3-compatible storages.",
+		},
+	}
+)
+
 const (
 	vmNativeFilterMatch       = "vm-native-filter-match"
 	vmNativeFilterTimeStart   = "vm-native-filter-time-start"

app/vmctl/main.go

@@ -18,6 +18,7 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/auth"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/backoff"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/barpool"
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/mimir"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/native"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/remoteread"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
@@ -297,12 +298,12 @@ func main() {
 				},
 			},
 			{
-				Name:   "thanos",
-				Usage:  "Migrate time series from Thanos blocks (supports raw and downsampled data)",
-				Flags:  mergeFlags(globalFlags, thanosFlags, vmFlags),
+				Name:   "mimir",
+				Usage:  "Migrate time series from Mimir object storage or local filesystem",
+				Flags:  mergeFlags(globalFlags, mimirFlags, vmFlags),
 				Before: beforeFn,
 				Action: func(c *cli.Context) error {
-					fmt.Println("Thanos import mode")
+					fmt.Println("Mimir import mode")
 
 					vmCfg, err := initConfigVM(c)
 					if err != nil {
@@ -314,6 +315,54 @@ func main() {
 						return fmt.Errorf("failed to create VM importer: %s", err)
 					}
 
+					mCfg := mimir.Config{
+						Filter: mimir.Filter{
+							TimeMin:    c.String(mimirFilterTimeStart),
+							TimeMax:    c.String(mimirFilterTimeEnd),
+							Label:      c.String(mimirFilterLabel),
+							LabelValue: c.String(mimirFilterLabelValue),
+						},
+						Path:                    c.String(mimirPath),
+						TenantID:                c.String(mimirTenantID),
+						CredsFilePath:           c.String(mimirCredsFilePath),
+						ConfigFilePath:          c.String(mimirConfigFilePath),
+						ConfigProfile:           c.String(mimirConfigProfile),
+						CustomS3Endpoint:        c.String(mimirCustomS3Endpoint),
+						S3ForcePathStyle:        c.Bool(mimirS3ForcePathStyle),
+						S3TLSInsecureSkipVerify: c.Bool(mimirS3TLSInsecureSkipVerify),
+						SSEKMSKeyID:             c.String(mimirSSEKMSKeyID),
+						SSEAlgorithm:            c.String(mimirSSEAlgorithm),
+					}
+					cl, err := mimir.NewClient(ctx, mCfg)
+					if err != nil {
+						return fmt.Errorf("failed to create mimir client: %s", err)
+					}
+
+					pp := prometheusProcessor{
+						cl:        cl,
+						im:        importer,
+						cc:        c.Int(mimirConcurrency),
+						isVerbose: c.Bool(globalVerbose),
+					}
+					return pp.run(ctx)
+				},
+			},
+			{
+				Name:   "thanos",
+				Usage:  "Migrate time series from Thanos blocks (supports raw and downsampled data)",
+				Flags:  mergeFlags(globalFlags, thanosFlags, vmFlags),
+				Before: beforeFn,
+				Action: func(c *cli.Context) error {
+					fmt.Println("Thanos import mode")
+					vmCfg, err := initConfigVM(c)
+					if err != nil {
+						return fmt.Errorf("failed to init VM configuration: %s", err)
+					}
+
+					importer, err = vm.NewImporter(ctx, vmCfg)
+					if err != nil {
+						return fmt.Errorf("failed to create VM importer: %s", err)
+					}
 					thanosCfg := thanos.Config{
 						Snapshot: c.String(thanosSnapshot),
 						Filter: thanos.Filter{

app/vmctl/mimir/lazyreader.go

@@ -0,0 +1,195 @@
+package mimir
+
+import (
+	"fmt"
+	"log"
+	"os"
+	"path/filepath"
+	"sync"
+
+	"github.com/oklog/ulid/v2"
+	"github.com/prometheus/prometheus/tsdb"
+	"github.com/prometheus/prometheus/tsdb/tombstones"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+)
+
+var _ tsdb.BlockReader = (*lazyBlockReader)(nil)
+
+// lazyBlockReader is stores block id and segment num information.
+// It is used to lazily fetch and parse block data.
+// It implements tsdb.BlockReader interface.
+type lazyBlockReader struct {
+	// Block ID.
+	ID ulid.ULID
+	// SegmentsNum stores the number of chunks segments in the block.
+	SegmentsNum int
+
+	mu          sync.Mutex
+	reader      *tsdb.Block
+	tempDirPath string
+	fs          common.RemoteFS
+	err         error
+}
+
+// newLazyBlockReader returns a new LazyBlockReader for the given block.
+func newLazyBlockReader(block *Block, fs common.RemoteFS) (*lazyBlockReader, error) {
+	if block.SegmentsFormat != "1b6d" {
+		return nil, fmt.Errorf("unsupported segments format: %s", block.SegmentsFormat)
+	}
+
+	return &lazyBlockReader{
+		ID:          block.ID,
+		SegmentsNum: block.SegmentsNum,
+		fs:          fs,
+	}, nil
+}
+
+func (lbr *lazyBlockReader) initialize() error {
+	lbr.mu.Lock()
+	defer lbr.mu.Unlock()
+	if lbr.reader != nil {
+		return nil
+	}
+	// fetching block and parse it and store it in lbr.reader
+	temp, err := lbr.mkTempDir()
+	if err != nil {
+		return fmt.Errorf("failed to create temp dir: %s", err)
+	}
+
+	lbr.tempDirPath = temp
+
+	// TODO: replace fetchFile and writeFile with buffered IO if needed
+	meta, err := lbr.fetchFile(metaFilename)
+	if err != nil {
+		return err
+	}
+	if err := lbr.writeFile(temp, metaFilename, meta); err != nil {
+		return fmt.Errorf("failed to write meta file: %w", err)
+	}
+	idx, err := lbr.fetchFile(indexFilename)
+	if err != nil {
+		return fmt.Errorf("failed to fetch index file %q: %w", indexFilename, err)
+	}
+	if err := lbr.writeFile(temp, indexFilename, idx); err != nil {
+		return err
+	}
+
+	for i := 1; i <= lbr.SegmentsNum; i++ {
+		// segments formats has format 1b06d
+		// https://github.com/grafana/mimir/blob/main/pkg/storage/tsdb/bucketindex/index.go#L32
+		chunkName := fmt.Sprintf("%06d", i)
+		blockChunkPath := filepath.Join("chunks", chunkName)
+		chunk, err := lbr.fetchFile(blockChunkPath)
+		if err != nil {
+			return fmt.Errorf("failed to fetch chunk file: %q: %w", chunkName, err)
+		}
+		if err := lbr.writeFile(temp, blockChunkPath, chunk); err != nil {
+			return fmt.Errorf("failed to write chunk file: %q: %s", chunkName, err)
+		}
+	}
+
+	// Set postingDecoder to nil because
+	// If it is nil then a default decoder is used, compatible with Prometheus v2.
+	pb, err := tsdb.OpenBlock(nil, temp, nil, nil)
+	if err != nil {
+		return fmt.Errorf("failed to open block %q: %w", lbr.ID, err)
+	}
+	lbr.reader = pb
+	return nil
+}
+
+// Index returns an IndexReader over the block's data.
+func (lbr *lazyBlockReader) Index() (tsdb.IndexReader, error) {
+	if err := lbr.initialize(); err != nil {
+		return nil, err
+	}
+	return lbr.reader.Index()
+}
+
+// Chunks returns a ChunkReader over the block's data.
+func (lbr *lazyBlockReader) Chunks() (tsdb.ChunkReader, error) {
+	if err := lbr.initialize(); err != nil {
+		return nil, err
+	}
+	return lbr.reader.Chunks()
+}
+
+// Tombstones returns a tombstones.Reader over the block's deleted data.
+func (lbr *lazyBlockReader) Tombstones() (tombstones.Reader, error) {
+	if err := lbr.initialize(); err != nil {
+		return nil, err
+	}
+	return lbr.reader.Tombstones()
+}
+
+// Meta provides meta information about the block reader.
+func (lbr *lazyBlockReader) Meta() tsdb.BlockMeta {
+	if err := lbr.initialize(); err != nil {
+		lbr.err = fmt.Errorf("cannot get BlockMeta: %w", err)
+		return tsdb.BlockMeta{}
+	}
+	return lbr.reader.Meta()
+}
+
+// Size returns the number of bytes that the block takes up on disk.
+func (lbr *lazyBlockReader) Size() int64 {
+	if err := lbr.initialize(); err != nil {
+		lbr.err = fmt.Errorf("error get Size of the block: %s, return zero size", err)
+		return 0
+	}
+	return lbr.reader.Size()
+}
+
+// Err returns the last error that occurred on the block reader.
+func (lbr *lazyBlockReader) Err() error {
+	return lbr.err
+}
+
+// Close closes block and releases all resources
+func (lbr *lazyBlockReader) Close() error {
+	lbr.mu.Lock()
+	defer lbr.mu.Unlock()
+	if lbr.reader == nil {
+		return nil
+	}
+
+	err := lbr.reader.Close()
+	if err := os.RemoveAll(lbr.tempDirPath); err != nil {
+		log.Printf("failed to remove temp dir: %s", err)
+	}
+	lbr.reader = nil
+	lbr.tempDirPath = ""
+
+	return err
+}
+
+func (lbr *lazyBlockReader) mkTempDir() (string, error) {
+	temp, err := os.MkdirTemp("", lbr.ID.String())
+	if err != nil {
+		return "", fmt.Errorf("failed to create temp dir: %s", err)
+	}
+	err = os.Mkdir(filepath.Join(temp, "chunks"), os.ModePerm)
+	if err != nil {
+		return "", fmt.Errorf("failed to create temp dir: %s", err)
+	}
+	return temp, nil
+}
+
+func (lbr *lazyBlockReader) fetchFile(filePath string) ([]byte, error) {
+	blockID := lbr.ID.String()
+	blockPath := filepath.Join(blockID, filePath)
+	has, err := lbr.fs.HasFile(blockPath)
+	if err != nil {
+		return nil, err
+	}
+	if !has {
+		return nil, fmt.Errorf("block meta %s not found", blockID)
+	}
+	return lbr.fs.ReadFile(blockPath)
+}
+
+func (lbr *lazyBlockReader) writeFile(folder string, filename string, file []byte) error {
+	fileName := filepath.Join(folder, filename)
+	return os.WriteFile(fileName, file, os.ModePerm)
+}

app/vmctl/mimir/mimir.go

@@ -0,0 +1,238 @@
+package mimir
+
+import (
+	"bytes"
+	"compress/gzip"
+	"context"
+	"encoding/json"
+	"fmt"
+	"log"
+
+	"github.com/oklog/ulid/v2"
+	"github.com/prometheus/prometheus/model/labels"
+	"github.com/prometheus/prometheus/tsdb"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/prometheus"
+	utils "github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/vmctlutil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+)
+
+const (
+	bucketIndex                   = "bucket-index.json"
+	bucketIndexCompressedFilename = bucketIndex + ".gz"
+	metaFilename                  = "meta.json"
+	indexFilename                 = "index"
+)
+
+// BlockDeletionMark holds the information about a block's deletion mark in the index.
+// This type was copied from the mimir repository https://github.com/grafana/mimir/blob/main/pkg/storage/tsdb/bucketindex/index.go#L234.
+type BlockDeletionMark struct {
+	// Block ID.
+	ID ulid.ULID `json:"block_id"`
+
+	// DeletionTime is a unix timestamp (seconds precision) of when the block was marked to be deleted.
+	DeletionTime int64 `json:"deletion_time"`
+}
+
+// Block holds the information about a block in the index.
+// This is a partial implementation of the https://github.com/grafana/mimir/blob/main/pkg/storage/tsdb/bucketindex/index.go#L73
+type Block struct {
+	// Block ID.
+	ID ulid.ULID `json:"block_id"`
+
+	// MinTime and MaxTime specify the time range all samples in the block are in (millis precision).
+	MinTime int64 `json:"min_time"`
+	MaxTime int64 `json:"max_time"`
+
+	// SegmentsFormat and SegmentsNum stores the format and number of chunks segments
+	// in the block.
+	SegmentsFormat string `json:"segments_format,omitempty"`
+	SegmentsNum    int    `json:"segments_num,omitempty"`
+}
+
+// Index contains all known blocks and markers of a tenant.
+// This is a partial implementation pof the https://github.com/grafana/mimir/blob/main/pkg/storage/tsdb/bucketindex/index.go#L36
+type Index struct {
+	// Version of the index format.
+	Version int `json:"version"`
+
+	// List of complete blocks (partial blocks are excluded from the index).
+	Blocks []*Block `json:"blocks"`
+}
+
+// Config contains a list of params needed
+// for reading mimir snapshots
+type Config struct {
+	// Path to remote storage bucket
+	Path string
+	// TenantID is the tenant id for the storage
+	TenantID string
+
+	Filter Filter
+
+	CredsFilePath           string
+	ConfigFilePath          string
+	ConfigProfile           string
+	CustomS3Endpoint        string
+	S3ForcePathStyle        bool
+	S3TLSInsecureSkipVerify bool
+
+	SSEKMSKeyID  string
+	SSEAlgorithm string
+}
+
+// Filter contains configuration for filtering
+// the timeseries
+type Filter struct {
+	TimeMin    string
+	TimeMax    string
+	Label      string
+	LabelValue string
+}
+
+// Client is a wrapper over Prometheus tsdb.DBReader
+type Client struct {
+	common.RemoteFS
+	filter filter
+}
+
+type filter struct {
+	min, max   int64
+	label      string
+	labelValue string
+}
+
+func (f filter) inRange(minTime, maxTime int64) bool {
+	fmin, fmax := f.min, f.max
+	if minTime == 0 {
+		fmin = minTime
+	}
+	if fmax == 0 {
+		fmax = maxTime
+	}
+	return minTime <= fmax && fmin <= maxTime
+}
+
+// NewClient creates and validates new Client
+// with given Config
+func NewClient(ctx context.Context, cfg Config) (*Client, error) {
+	if cfg.Path == "" {
+		return nil, fmt.Errorf("path cannot be empty")
+	}
+
+	if cfg.TenantID != "" {
+		cfg.Path = fmt.Sprintf("%s/%s", cfg.Path, cfg.TenantID)
+	}
+
+	var c Client
+	rfs, err := newRemoteFS(ctx, cfg)
+	if err != nil {
+		return nil, fmt.Errorf("cannot parse `-src`=%q: %w", cfg.Path, err)
+	}
+
+	c.RemoteFS = rfs
+	timeMin, err := utils.ParseTime(cfg.Filter.TimeMin)
+	if err != nil {
+		return nil, fmt.Errorf("failed to parse min time in filter: %s", err)
+	}
+	timeMax, err := utils.ParseTime(cfg.Filter.TimeMax)
+	if err != nil {
+		return nil, fmt.Errorf("failed to parse max time in filter: %s", err)
+	}
+	c.filter = filter{
+		min:        timeMin.UnixMilli(),
+		max:        timeMax.UnixMilli(),
+		label:      cfg.Filter.Label,
+		labelValue: cfg.Filter.LabelValue,
+	}
+	return &c, nil
+}
+
+// Explore a fetches bucket-index.json file from a remote storage or local filesystem
+// and filter blocks via the defined time range, but does not take into account label filters.
+func (c *Client) Explore() ([]tsdb.BlockReader, error) {
+
+	log.Printf("Fetching blocks from remote storage")
+
+	indexFile, err := c.fetchIndexFile()
+	if err != nil {
+		return nil, fmt.Errorf("failed to fetch index file: %s", err)
+	}
+
+	var blocksToImport []tsdb.BlockReader
+	for _, block := range indexFile.Blocks {
+		if !c.filter.inRange(block.MinTime, block.MaxTime) {
+			// Skipping block outside of time range
+			continue
+		}
+
+		if block.ID.String() == "" {
+			continue
+		}
+
+		lazyBlockReader, err := newLazyBlockReader(block, c.RemoteFS)
+		if err != nil {
+			return nil, fmt.Errorf("failed to create lazy block reader: %s", err)
+		}
+		blocksToImport = append(blocksToImport, lazyBlockReader)
+	}
+
+	return blocksToImport, nil
+}
+
+// Read reads the given BlockReader according to configured
+// time and label filters.
+func (c *Client) Read(ctx context.Context, block tsdb.BlockReader) (*prometheus.CloseableSeriesSet, error) {
+	meta := block.Meta()
+	if b, ok := block.(*lazyBlockReader); ok && b.Err() != nil {
+		return nil, fmt.Errorf("failed to read block: %s", b.Err())
+	}
+
+	if meta.ULID.String() == "" {
+		return nil, fmt.Errorf("unexpected block without id")
+	}
+
+	minTime, maxTime := meta.MinTime, meta.MaxTime
+	if c.filter.min != 0 {
+		minTime = c.filter.min
+	}
+	if c.filter.max != 0 {
+		maxTime = c.filter.max
+	}
+	q, err := tsdb.NewBlockQuerier(block, minTime, maxTime)
+	if err != nil {
+		return nil, err
+	}
+	ss := q.Select(ctx, false, nil, labels.MustNewMatcher(labels.MatchRegexp, c.filter.label, c.filter.labelValue))
+	return &prometheus.CloseableSeriesSet{SeriesSet: ss, Close: q.Close}, nil
+}
+
+func (c *Client) fetchIndexFile() (*Index, error) {
+	has, err := c.HasFile(bucketIndexCompressedFilename)
+	if err != nil {
+		return nil, err
+	}
+	if !has {
+		return nil, fmt.Errorf("bucket-index.json.gz not found")
+	}
+
+	file, err := c.ReadFile(bucketIndexCompressedFilename)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read bucket index: %s", err)
+	}
+
+	r := bytes.NewReader(file)
+	// Read all the content.
+	gzipReader, err := gzip.NewReader(r)
+	if err != nil {
+		return nil, fmt.Errorf("failed to create gzip reader: %s", err)
+	}
+
+	var indexFile Index
+	err = json.NewDecoder(gzipReader).Decode(&indexFile)
+	if err != nil {
+		return nil, fmt.Errorf("failed to decode bucket index: %s", err)
+	}
+
+	return &indexFile, nil
+}

app/vmctl/mimir/remotefs.go

@@ -0,0 +1,93 @@
+package mimir
+
+import (
+	"context"
+	"fmt"
+	"path/filepath"
+	"strings"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/azremote"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fsremote"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/gcsremote"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/s3remote"
+)
+
+// newRemoteFS returns new remote fs from the given Config.
+func newRemoteFS(ctx context.Context, cfg Config) (common.RemoteFS, error) {
+	if len(cfg.Path) == 0 {
+		return nil, fmt.Errorf("path cannot be empty")
+	}
+	n := strings.Index(cfg.Path, "://")
+	if n < 0 {
+		return nil, fmt.Errorf("missing scheme in path %q. Supported schemes: `gs://`, `s3://`, `azblob://`, `fs://`", cfg.Path)
+	}
+	scheme := cfg.Path[:n]
+	dir := cfg.Path[n+len("://"):]
+	switch scheme {
+	case "fs":
+		if !filepath.IsAbs(dir) {
+			return nil, fmt.Errorf("dir must be absolute; got %q", dir)
+		}
+		fsr := &fsremote.FS{
+			Dir: filepath.Clean(dir),
+		}
+		return fsr, nil
+	case "gcs", "gs":
+		n := strings.Index(dir, "/")
+		if n < 0 {
+			return nil, fmt.Errorf("missing directory on the gcs bucket %q", dir)
+		}
+		bucket := dir[:n]
+		dir = dir[n:]
+		fsr := &gcsremote.FS{
+			CredsFilePath: cfg.CredsFilePath,
+			Bucket:        bucket,
+			Dir:           dir,
+		}
+		if err := fsr.Init(ctx); err != nil {
+			return nil, fmt.Errorf("cannot initialize connection to gcs: %w", err)
+		}
+		return fsr, nil
+	case "azblob":
+		n := strings.Index(dir, "/")
+		if n < 0 {
+			return nil, fmt.Errorf("missing directory on the AZBlob container %q", dir)
+		}
+		bucket := dir[:n]
+		dir = dir[n:]
+		fsr := &azremote.FS{
+			Container: bucket,
+			Dir:       dir,
+		}
+		if err := fsr.Init(ctx); err != nil {
+			return nil, fmt.Errorf("cannot initialize connection to AZBlob: %w", err)
+		}
+		return fsr, nil
+	case "s3":
+		n := strings.Index(dir, "/")
+		if n < 0 {
+			return nil, fmt.Errorf("missing directory on the s3 bucket %q", dir)
+		}
+		bucket := dir[:n]
+		dir = dir[n:]
+		fsr := &s3remote.FS{
+			CredsFilePath:         cfg.CredsFilePath,
+			ConfigFilePath:        cfg.ConfigFilePath,
+			CustomEndpoint:        cfg.CustomS3Endpoint,
+			TLSInsecureSkipVerify: cfg.S3TLSInsecureSkipVerify,
+			S3ForcePathStyle:      cfg.S3ForcePathStyle,
+			ProfileName:           cfg.ConfigProfile,
+			Bucket:                bucket,
+			Dir:                   dir,
+			SSEKMSKeyId:           cfg.SSEKMSKeyID,
+			SSEAlgorithm:          s3remote.StringToEncryptionAlgorithm(cfg.SSEAlgorithm),
+		}
+		if err := fsr.Init(ctx); err != nil {
+			return nil, fmt.Errorf("cannot initialize connection to s3: %w", err)
+		}
+		return fsr, nil
+	default:
+		return nil, fmt.Errorf("unsupported scheme %q", scheme)
+	}
+}

app/vmctl/prometheus.go

@@ -3,6 +3,7 @@ package main
 import (
 	"context"
 	"fmt"
+	"io"
 	"log"
 	"strings"
 	"sync"
@@ -18,10 +19,17 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/vm"
 )
 
+// Runner is an interface for fetching and reading
+// snapshot blocks
+type Runner interface {
+	Explore() ([]tsdb.BlockReader, error)
+	Read(context.Context, tsdb.BlockReader) (*prometheus.CloseableSeriesSet, error)
+}
+
 type prometheusProcessor struct {
-	// prometheus client fetches and reads
+	// Runner fetches and reads
 	// snapshot blocks
-	cl *prometheus.Client
+	cl Runner
 	// importer performs import requests
 	// for timeseries data returned from
 	// snapshot blocks
@@ -48,7 +56,7 @@ func (pp *prometheusProcessor) run(ctx context.Context) error {
 		return nil
 	}
 
-	if err := pp.processBlocks(blocks); err != nil {
+	if err := pp.processBlocks(ctx, blocks); err != nil {
 		return fmt.Errorf("migration failed: %s", err)
 	}
 
@@ -57,11 +65,17 @@ func (pp *prometheusProcessor) run(ctx context.Context) error {
 	return nil
 }
 
-func (pp *prometheusProcessor) do(b tsdb.BlockReader) error {
-	ss, err := pp.cl.Read(b)
+func (pp *prometheusProcessor) do(ctx context.Context, b tsdb.BlockReader) error {
+	css, err := pp.cl.Read(ctx, b)
 	if err != nil {
 		return fmt.Errorf("failed to read block: %s", err)
 	}
+	defer func() {
+		if err := css.Close(); err != nil {
+			log.Printf("cannot close SeriesSet for block: %q : %s\n", b.Meta().ULID, err)
+		}
+	}()
+	ss := css.SeriesSet
 	var it chunkenc.Iterator
 	for ss.Next() {
 		var name string
@@ -114,7 +128,7 @@ func (pp *prometheusProcessor) do(b tsdb.BlockReader) error {
 	return ss.Err()
 }
 
-func (pp *prometheusProcessor) processBlocks(blocks []tsdb.BlockReader) error {
+func (pp *prometheusProcessor) processBlocks(ctx context.Context, blocks []tsdb.BlockReader) error {
 	promBlocksTotal.Add(len(blocks))
 	bar := barpool.AddWithTemplate(fmt.Sprintf(barTpl, "Processing blocks"), len(blocks))
 	if err := barpool.Start(); err != nil {
@@ -130,11 +144,16 @@ func (pp *prometheusProcessor) processBlocks(blocks []tsdb.BlockReader) error {
 	for range pp.cc {
 		wg.Go(func() {
 			for br := range blockReadersCh {
-				if err := pp.do(br); err != nil {
+				if err := pp.do(ctx, br); err != nil {
 					promErrorsTotal.Inc()
-					errCh <- fmt.Errorf("read failed for block %q: %s", br.Meta().ULID, err)
+					errCh <- fmt.Errorf("cannot read block %q: %s", br.Meta().ULID, err)
 					return
 				}
+				if cb, ok := br.(io.Closer); ok {
+					if err := cb.Close(); err != nil {
+						errCh <- fmt.Errorf("cannot close block: %q: %w", br.Meta().ULID, err)
+					}
+				}
 				promBlocksProcessed.Inc()
 				bar.Increment()
 			}

app/vmctl/prometheus/prometheus.go

@@ -8,6 +8,8 @@ import (
 	"github.com/prometheus/prometheus/model/labels"
 	"github.com/prometheus/prometheus/storage"
 	"github.com/prometheus/prometheus/tsdb"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmctl/vmctlutil"
 )
 
 // Config contains a list of params needed
@@ -60,13 +62,13 @@ func NewClient(cfg Config) (*Client, error) {
 		return nil, fmt.Errorf("failed to open snapshot %q: %s", cfg.Snapshot, err)
 	}
 	c := &Client{DBReadOnly: db}
-	minTime, maxTime, err := parseTime(cfg.Filter.TimeMin, cfg.Filter.TimeMax)
+	timeMin, timeMax, err := parseTime(cfg.Filter.TimeMin, cfg.Filter.TimeMax)
 	if err != nil {
 		return nil, fmt.Errorf("failed to parse time in filter: %s", err)
 	}
 	c.filter = filter{
-		min:        minTime,
-		max:        maxTime,
+		min:        timeMin,
+		max:        timeMax,
 		label:      cfg.Filter.Label,
 		labelValue: cfg.Filter.LabelValue,
 	}
@@ -83,7 +85,7 @@ func (c *Client) Explore() ([]tsdb.BlockReader, error) {
 	if err != nil {
 		return nil, fmt.Errorf("failed to fetch blocks: %s", err)
 	}
-	s := &Stats{
+	s := &vmctlutil.Stats{
 		Filtered: c.filter.min != 0 || c.filter.max != 0 || c.filter.label != "",
 		Blocks:   len(blocks),
 	}
@@ -108,9 +110,15 @@ func (c *Client) Explore() ([]tsdb.BlockReader, error) {
 	return blocksToImport, nil
 }
 
+// CloseableSeriesSet defines a SeriesSet with Close method
+type CloseableSeriesSet struct {
+	SeriesSet storage.SeriesSet
+	Close     func() error
+}
+
 // Read reads the given BlockReader according to configured
 // time and label filters.
-func (c *Client) Read(block tsdb.BlockReader) (storage.SeriesSet, error) {
+func (c *Client) Read(ctx context.Context, block tsdb.BlockReader) (*CloseableSeriesSet, error) {
 	minTime, maxTime := block.Meta().MinTime, block.Meta().MaxTime
 	if c.filter.min != 0 {
 		minTime = c.filter.min
@@ -122,8 +130,8 @@ func (c *Client) Read(block tsdb.BlockReader) (storage.SeriesSet, error) {
 	if err != nil {
 		return nil, err
 	}
-	ss := q.Select(context.Background(), false, nil, labels.MustNewMatcher(labels.MatchRegexp, c.filter.label, c.filter.labelValue))
-	return ss, nil
+	ss := q.Select(ctx, false, nil, labels.MustNewMatcher(labels.MatchRegexp, c.filter.label, c.filter.labelValue))
+	return &CloseableSeriesSet{ss, q.Close}, nil
 }
 
 func parseTime(start, end string) (int64, int64, error) {

app/vmctl/vmctlutil/stats.go

@@ -1,4 +1,4 @@
-package prometheus
+package vmctlutil
 
 import (
 	"fmt"
@@ -18,7 +18,7 @@ type Stats struct {
 
 // String returns string representation for s.
 func (s Stats) String() string {
-	str := fmt.Sprintf("Prometheus snapshot stats:\n"+
+	str := fmt.Sprintf("Snapshot stats:\n"+
 		"  blocks found: %d;\n"+
 		"  blocks skipped by time filter: %d;\n"+
 		"  min time: %d (%v);\n"+

app/vminsert/common/insert_ctx.go

@@ -184,7 +184,7 @@ func (ctx *InsertCtx) WriteMetadata(mmpbs []prompb.MetricMetadata) error {
 	}
 	ctx.mms = mms
 
-	err := vmstorage.AddMetadataRows(mms)
+	err := vmstorage.VMInsertAPI.WriteMetadata(mms)
 	if err != nil {
 		return &httpserver.ErrorWithStatusCode{
 			Err:        fmt.Errorf("cannot store metrics metadata: %w", err),
@@ -209,7 +209,7 @@ func (ctx *InsertCtx) WritePromMetadata(mmps []prometheus.Metadata) error {
 	}
 	ctx.mms = mms
 
-	err := vmstorage.AddMetadataRows(mms)
+	err := vmstorage.VMInsertAPI.WriteMetadata(mms)
 	if err != nil {
 		return &httpserver.ErrorWithStatusCode{
 			Err:        fmt.Errorf("cannot store prometheus metrics metadata: %w", err),
@@ -278,7 +278,7 @@ func (ctx *InsertCtx) FlushBufs() error {
 	// since the number of concurrent FlushBufs() calls should be already limited via writeconcurrencylimiter
 	// used at every stream.Parse() call under lib/protoparser/*
 
-	err := vmstorage.AddRows(ctx.mrs)
+	err := vmstorage.VMInsertAPI.WriteRows(ctx.mrs)
 	ctx.Reset(0)
 	if err == nil {
 		return nil

app/vminsert/common/streamaggr.go

@@ -283,7 +283,7 @@ func pushAggregateSeries(tss []prompb.TimeSeries) {
 	}
 	// There is no need in limiting the number of concurrent calls to vmstorage.AddRows() here,
 	// since the number of concurrent pushAggregateSeries() calls should be already limited by lib/streamaggr.
-	if err := vmstorage.AddRows(ctx.mrs); err != nil {
+	if err := vmstorage.VMInsertAPI.WriteRows(ctx.mrs); err != nil {
 		logger.Errorf("cannot flush aggregate series: %s", err)
 	}
 }

app/vmselect/graphite/metrics_api.go

@@ -1,7 +1,6 @@
 package graphite
 
 import (
-	"flag"
 	"fmt"
 	"math"
 	"net/http"
@@ -21,8 +20,6 @@ import (
 	"github.com/VictoriaMetrics/metricsql"
 )
 
-var maxTagValueSuffixes = flag.Int("search.maxTagValueSuffixesPerSearch", 100e3, "The maximum number of tag value suffixes returned from /metrics/find")
-
 // MetricsFindHandler implements /metrics/find handler.
 //
 // See https://graphite-api.readthedocs.io/en/latest/api.html#metrics-find
@@ -222,10 +219,11 @@ func MetricsIndexHandler(startTime time.Time, w http.ResponseWriter, r *http.Req
 
 // metricsFind searches for label values that match the given qHead and qTail.
 func metricsFind(tr storage.TimeRange, label, qHead, qTail string, delimiter byte, isExpand bool, deadline searchutil.Deadline) ([]string, error) {
+	maxSuffixes := 0 // let vmstorage use its maxTagValueSuffixesPerSearch limit
 	n := strings.IndexAny(qTail, "*{[")
 	if n < 0 {
 		query := qHead + qTail
-		suffixes, err := netstorage.TagValueSuffixes(nil, tr, label, query, delimiter, *maxTagValueSuffixes, deadline)
+		suffixes, err := netstorage.TagValueSuffixes(nil, tr, label, query, delimiter, maxSuffixes, deadline)
 		if err != nil {
 			return nil, err
 		}
@@ -245,7 +243,7 @@ func metricsFind(tr storage.TimeRange, label, qHead, qTail string, delimiter byt
 	}
 	if n == len(qTail)-1 && strings.HasSuffix(qTail, "*") {
 		query := qHead + qTail[:len(qTail)-1]
-		suffixes, err := netstorage.TagValueSuffixes(nil, tr, label, query, delimiter, *maxTagValueSuffixes, deadline)
+		suffixes, err := netstorage.TagValueSuffixes(nil, tr, label, query, delimiter, maxSuffixes, deadline)
 		if err != nil {
 			return nil, err
 		}

app/vmselect/graphite/tags_api.go

@@ -138,7 +138,9 @@ func registerMetrics(startTime time.Time, w http.ResponseWriter, r *http.Request
 		mr.MetricNameRaw = storage.MarshalMetricNameRaw(mr.MetricNameRaw[:0], labels)
 		mr.Timestamp = ct
 	}
-	vmstorage.RegisterMetricNames(nil, mrs)
+	if err := vmstorage.VMSelectAPI.RegisterMetricNames(nil, mrs, 0); err != nil {
+		return err
+	}
 
 	// Return response
 	contentType := "text/plain; charset=utf-8"

app/vmselect/main.go

@@ -21,7 +21,6 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect/stats"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmstorage"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/buildinfo"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/cgroup"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/httpserver"
@@ -36,12 +35,6 @@ var (
 	deleteAuthKey                = flagutil.NewPassword("deleteAuthKey", "authKey for metrics' deletion via /api/v1/admin/tsdb/delete_series and /tags/delSeries. It could be passed via authKey query arg. It overrides -httpAuth.*")
 	metricNamesStatsResetAuthKey = flagutil.NewPassword("metricNamesStatsResetAuthKey", "authKey for resetting metric names usage cache via /api/v1/admin/status/metric_names_stats/reset. It overrides -httpAuth.*. "+
 		"See https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#track-ingested-metrics-usage")
-
-	maxConcurrentRequests = flag.Int("search.maxConcurrentRequests", getDefaultMaxConcurrentRequests(), "The maximum number of concurrent search requests. "+
-		"It shouldn't be high, since a single request can saturate all the CPU cores, while many concurrently executed requests may require high amounts of memory. "+
-		"See also -search.maxQueueDuration and -search.maxMemoryPerQuery")
-	maxQueueDuration = flag.Duration("search.maxQueueDuration", 10*time.Second, "The maximum time the request waits for execution when -search.maxConcurrentRequests "+
-		"limit is reached; see also -search.maxQueryDuration")
 	resetCacheAuthKey    = flagutil.NewPassword("search.resetCacheAuthKey", "Optional authKey for resetting rollup cache via /internal/resetRollupResultCache call. It could be passed via authKey query arg. It overrides -httpAuth.*")
 	logSlowQueryDuration = flag.Duration("search.logSlowQueryDuration", 5*time.Second, "Log queries with execution time exceeding this value. Zero disables slow query logging. "+
 		"See also -search.logQueryMemoryUsage")
@@ -50,23 +43,17 @@ var (
 
 var slowQueries = metrics.NewCounter(`vm_slow_queries_total`)
 
-func getDefaultMaxConcurrentRequests() int {
-	// A single request can saturate all the CPU cores, so there is no sense
-	// in allowing higher number of concurrent requests - they will just contend
-	// for unavailable CPU time.
-	n := min(cgroup.AvailableCPUs()*2, 16)
-	return n
-}
-
 // Init initializes vmselect
-func Init() {
+func Init(vmselectMaxConcurrentRequests int, vmselectMaxQueueDuration time.Duration) {
 	tmpDirPath := vmstorage.DataPath() + "/tmp"
 	fs.MustRemoveDirContents(tmpDirPath)
 	netstorage.InitTmpBlocksDir(tmpDirPath)
 	promql.InitRollupResultCache(vmstorage.DataPath() + "/cache/rollupResult")
-	prometheus.InitMaxUniqueTimeseries(*maxConcurrentRequests)
 
-	concurrencyLimitCh = make(chan struct{}, *maxConcurrentRequests)
+	maxConcurrentRequests = vmselectMaxConcurrentRequests
+	maxQueueDuration = vmselectMaxQueueDuration
+	concurrencyLimitCh = make(chan struct{}, maxConcurrentRequests)
+
 	initVMUIConfig()
 	initVMAlertProxy()
 
@@ -78,7 +65,11 @@ func Stop() {
 	promql.StopRollupResultCache()
 }
 
-var concurrencyLimitCh chan struct{}
+var (
+	maxConcurrentRequests int
+	maxQueueDuration      time.Duration
+	concurrencyLimitCh    chan struct{}
+)
 
 var (
 	concurrencyLimitReached = metrics.NewCounter(`vm_concurrent_select_limit_reached_total`)
@@ -90,9 +81,6 @@ var (
 	_ = metrics.NewGauge(`vm_concurrent_select_current`, func() float64 {
 		return float64(len(concurrencyLimitCh))
 	})
-	_ = metrics.NewGauge(`vm_search_max_unique_timeseries`, func() float64 {
-		return float64(prometheus.GetMaxUniqueTimeSeries())
-	})
 )
 
 //go:embed vmui
@@ -131,12 +119,12 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	default:
 		// Sleep for a while until giving up. This should resolve short bursts in requests.
 		concurrencyLimitReached.Inc()
-		d := min(searchutil.GetMaxQueryDuration(r), *maxQueueDuration)
+		d := min(searchutil.GetMaxQueryDuration(r), maxQueueDuration)
 		t := timerpool.Get(d)
 		select {
 		case concurrencyLimitCh <- struct{}{}:
 			timerpool.Put(t)
-			qt.Printf("wait in queue because -search.maxConcurrentRequests=%d concurrent requests are executed", *maxConcurrentRequests)
+			qt.Printf("wait in queue because -search.maxConcurrentRequests=%d concurrent requests are executed", maxConcurrentRequests)
 			defer func() { <-concurrencyLimitCh }()
 		case <-r.Context().Done():
 			timerpool.Put(t)
@@ -152,7 +140,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 				Err: fmt.Errorf("couldn't start executing the request in %.3f seconds, since -search.maxConcurrentRequests=%d concurrent requests "+
 					"are executed. Possible solutions: to reduce query load; to add more compute resources to the server; "+
 					"to increase -search.maxQueueDuration=%s; to increase -search.maxQueryDuration; to increase -search.maxConcurrentRequests",
-					d.Seconds(), *maxConcurrentRequests, maxQueueDuration),
+					d.Seconds(), maxConcurrentRequests, maxQueueDuration),
 				StatusCode: http.StatusTooManyRequests,
 			}
 			w.Header().Add("Retry-After", "10")

app/vmselect/netstorage/netstorage.go

@@ -27,10 +27,6 @@ import (
 )
 
 var (
-	maxTagKeysPerSearch = flag.Int("search.maxTagKeys", 100e3, "The maximum number of tag keys returned from /api/v1/labels . "+
-		"See also -search.maxLabelsAPISeries and -search.maxLabelsAPIDuration")
-	maxTagValuesPerSearch = flag.Int("search.maxTagValues", 100e3, "The maximum number of tag values returned from /api/v1/label/<label_name>/values . "+
-		"See also -search.maxLabelsAPISeries and -search.maxLabelsAPIDuration")
 	maxSamplesPerSeries = flag.Int("search.maxSamplesPerSeries", 30e6, "The maximum number of raw samples a single query can scan per each time series. This option allows limiting memory usage")
 	maxSamplesPerQuery  = flag.Int("search.maxSamplesPerQuery", 1e9, "The maximum number of raw samples a single query can process across all time series. "+
 		"This protects from heavy queries, which select unexpectedly high number of raw samples. See also -search.maxSamplesPerSeries")
@@ -80,7 +76,7 @@ func (rss *Results) Cancel() {
 }
 
 func (rss *Results) mustClose() {
-	putStorageSearch(rss.sr)
+	vmstorage.PutSearch(rss.sr)
 	rss.sr = nil
 	putTmpBlocksFile(rss.tbf)
 	rss.tbf = nil
@@ -758,12 +754,7 @@ var sbhPool sync.Pool
 func DeleteSeries(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline searchutil.Deadline) (int, error) {
 	qt = qt.NewChild("delete series: %s", sq)
 	defer qt.Done()
-	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
-	if err != nil {
-		return 0, err
-	}
-	return vmstorage.DeleteSeries(qt, tfss, sq.MaxMetrics)
+	return vmstorage.VMSelectAPI.DeleteSeries(qt, sq, deadline.Deadline())
 }
 
 // LabelNames returns label names matching the given sq until the given deadline.
@@ -773,15 +764,7 @@ func LabelNames(qt *querytracer.Tracer, sq *storage.SearchQuery, maxLabelNames i
 	if deadline.Exceeded() {
 		return nil, fmt.Errorf("timeout exceeded before starting the query processing: %s", deadline.String())
 	}
-	if maxLabelNames > *maxTagKeysPerSearch || maxLabelNames <= 0 {
-		maxLabelNames = *maxTagKeysPerSearch
-	}
-	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
-	if err != nil {
-		return nil, err
-	}
-	labels, err := vmstorage.SearchLabelNames(qt, tfss, tr, maxLabelNames, sq.MaxMetrics, deadline.Deadline())
+	labels, err := vmstorage.VMSelectAPI.LabelNames(qt, sq, maxLabelNames, deadline.Deadline())
 	if err != nil {
 		return nil, fmt.Errorf("error during labels search on time range: %w", err)
 	}
@@ -841,15 +824,7 @@ func LabelValues(qt *querytracer.Tracer, labelName string, sq *storage.SearchQue
 	if deadline.Exceeded() {
 		return nil, fmt.Errorf("timeout exceeded before starting the query processing: %s", deadline.String())
 	}
-	if maxLabelValues > *maxTagValuesPerSearch || maxLabelValues <= 0 {
-		maxLabelValues = *maxTagValuesPerSearch
-	}
-	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
-	if err != nil {
-		return nil, err
-	}
-	labelValues, err := vmstorage.SearchLabelValues(qt, labelName, tfss, tr, maxLabelValues, sq.MaxMetrics, deadline.Deadline())
+	labelValues, err := vmstorage.VMSelectAPI.LabelValues(qt, sq, labelName, maxLabelValues, deadline.Deadline())
 	if err != nil {
 		return nil, fmt.Errorf("error during label values search on time range for labelName=%q: %w", labelName, err)
 	}
@@ -864,7 +839,10 @@ func GetMetricsMetadata(qt *querytracer.Tracer, limit int, metricName string) ([
 	qt = qt.NewChild("get metrics metadata: limit=%d, metric_name=%q", limit, metricName)
 	defer qt.Done()
 
-	metadata := vmstorage.Storage.GetMetadataRows(qt, limit, metricName)
+	metadata, err := vmstorage.VMSelectAPI.GetMetadataRecords(qt, nil, limit, metricName, 0)
+	if err != nil {
+		return nil, err
+	}
 
 	sort.Slice(metadata, func(i, j int) bool {
 		return string(metadata[i].MetricFamilyName) < string(metadata[j].MetricFamilyName)
@@ -912,16 +890,11 @@ func TagValueSuffixes(qt *querytracer.Tracer, tr storage.TimeRange, tagKey, tagV
 	if deadline.Exceeded() {
 		return nil, fmt.Errorf("timeout exceeded before starting the query processing: %s", deadline.String())
 	}
-	suffixes, err := vmstorage.SearchTagValueSuffixes(qt, tr, tagKey, tagValuePrefix, delimiter, maxSuffixes, deadline.Deadline())
+	suffixes, err := vmstorage.VMSelectAPI.TagValueSuffixes(qt, 0, 0, tr, tagKey, tagValuePrefix, delimiter, maxSuffixes, deadline.Deadline())
 	if err != nil {
 		return nil, fmt.Errorf("error during search for suffixes for tagKey=%q, tagValuePrefix=%q, delimiter=%c on time range %s: %w",
 			tagKey, tagValuePrefix, delimiter, tr.String(), err)
 	}
-	if len(suffixes) >= maxSuffixes {
-		return nil, fmt.Errorf("more than -search.maxTagValueSuffixesPerSearch=%d tag value suffixes found for tagKey=%q, tagValuePrefix=%q, delimiter=%c on time range %s; "+
-			"either narrow down the query or increase -search.maxTagValueSuffixesPerSearch command-line flag value",
-			maxSuffixes, tagKey, tagValuePrefix, delimiter, tr.String())
-	}
 	return suffixes, nil
 }
 
@@ -934,13 +907,7 @@ func TSDBStatus(qt *querytracer.Tracer, sq *storage.SearchQuery, focusLabel stri
 	if deadline.Exceeded() {
 		return nil, fmt.Errorf("timeout exceeded before starting the query processing: %s", deadline.String())
 	}
-	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
-	if err != nil {
-		return nil, err
-	}
-	date := uint64(tr.MinTimestamp) / (3600 * 24 * 1000)
-	status, err := vmstorage.GetTSDBStatus(qt, tfss, date, focusLabel, topN, sq.MaxMetrics, deadline.Deadline())
+	status, err := vmstorage.VMSelectAPI.TSDBStatus(qt, sq, focusLabel, topN, deadline.Deadline())
 	if err != nil {
 		return nil, fmt.Errorf("error during tsdb status request: %w", err)
 	}
@@ -954,28 +921,13 @@ func SeriesCount(qt *querytracer.Tracer, deadline searchutil.Deadline) (uint64,
 	if deadline.Exceeded() {
 		return 0, fmt.Errorf("timeout exceeded before starting the query processing: %s", deadline.String())
 	}
-	n, err := vmstorage.GetSeriesCount(deadline.Deadline())
+	n, err := vmstorage.VMSelectAPI.SeriesCount(qt, 0, 0, deadline.Deadline())
 	if err != nil {
 		return 0, fmt.Errorf("error during series count request: %w", err)
 	}
 	return n, nil
 }
 
-func getStorageSearch() *storage.Search {
-	v := ssPool.Get()
-	if v == nil {
-		return &storage.Search{}
-	}
-	return v.(*storage.Search)
-}
-
-func putStorageSearch(sr *storage.Search) {
-	sr.MustClose()
-	ssPool.Put(sr)
-}
-
-var ssPool sync.Pool
-
 // ExportBlocks searches for time series matching sq and calls f for each found block.
 //
 // f is called in parallel from multiple goroutines.
@@ -989,18 +941,13 @@ func ExportBlocks(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline sear
 	if deadline.Exceeded() {
 		return fmt.Errorf("timeout exceeded before starting data export: %s", deadline.String())
 	}
+
 	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
+	sr, _, err := vmstorage.GetSearch(qt, sq, deadline.Deadline())
 	if err != nil {
 		return err
 	}
-
-	vmstorage.WG.Add(1)
-	defer vmstorage.WG.Done()
-
-	sr := getStorageSearch()
-	defer putStorageSearch(sr)
-	sr.Init(qt, vmstorage.Storage, tfss, tr, sq.MaxMetrics, deadline.Deadline())
+	defer vmstorage.PutSearch(sr)
 
 	// Start workers that call f in parallel on available CPU cores.
 	workCh := make(chan *exportWork, gomaxprocs*8)
@@ -1093,14 +1040,7 @@ func SearchMetricNames(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline
 		return nil, fmt.Errorf("timeout exceeded before starting to search metric names: %s", deadline.String())
 	}
 
-	// Setup search.
-	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
-	if err != nil {
-		return nil, err
-	}
-
-	metricNames, err := vmstorage.SearchMetricNames(qt, tfss, tr, sq.MaxMetrics, deadline.Deadline())
+	metricNames, err := vmstorage.VMSelectAPI.SearchMetricNames(qt, sq, deadline.Deadline())
 	if err != nil {
 		return nil, fmt.Errorf("cannot find metric names: %w", err)
 	}
@@ -1119,18 +1059,11 @@ func ProcessSearchQuery(qt *querytracer.Tracer, sq *storage.SearchQuery, deadlin
 		return nil, fmt.Errorf("timeout exceeded before starting the query processing: %s", deadline.String())
 	}
 
-	// Setup search.
-	tr := sq.GetTimeRange()
-	tfss, err := setupTfss(qt, tr, sq.TagFilterss, sq.MaxMetrics, deadline)
+	sr, maxSeriesCount, err := vmstorage.GetSearch(qt, sq, deadline.Deadline())
 	if err != nil {
 		return nil, err
 	}
 
-	vmstorage.WG.Add(1)
-	defer vmstorage.WG.Done()
-
-	sr := getStorageSearch()
-	maxSeriesCount := sr.Init(qt, vmstorage.Storage, tfss, tr, sq.MaxMetrics, deadline.Deadline())
 	type blockRefs struct {
 		brs []blockRef
 	}
@@ -1168,7 +1101,7 @@ func ProcessSearchQuery(qt *querytracer.Tracer, sq *storage.SearchQuery, deadlin
 		blocksRead++
 		if deadline.Exceeded() {
 			putTmpBlocksFile(tbf)
-			putStorageSearch(sr)
+			vmstorage.PutSearch(sr)
 			return nil, fmt.Errorf("timeout exceeded while fetching data block #%d from storage: %s", blocksRead, deadline.String())
 		}
 		br := sr.MetricBlockRef.BlockRef
@@ -1180,7 +1113,7 @@ func ProcessSearchQuery(qt *querytracer.Tracer, sq *storage.SearchQuery, deadlin
 		samples += br.RowsCount()
 		if *maxSamplesPerQuery > 0 && samples > *maxSamplesPerQuery {
 			putTmpBlocksFile(tbf)
-			putStorageSearch(sr)
+			vmstorage.PutSearch(sr)
 			return nil, fmt.Errorf("cannot select more than -search.maxSamplesPerQuery=%d samples; possible solutions: increase the -search.maxSamplesPerQuery; "+
 				"reduce time range for the query; use more specific label filters in order to select fewer series", *maxSamplesPerQuery)
 		}
@@ -1189,7 +1122,7 @@ func ProcessSearchQuery(qt *querytracer.Tracer, sq *storage.SearchQuery, deadlin
 		addr, err := tbf.WriteBlockRefData(buf)
 		if err != nil {
 			putTmpBlocksFile(tbf)
-			putStorageSearch(sr)
+			vmstorage.PutSearch(sr)
 			return nil, fmt.Errorf("cannot write %d bytes to temporary file: %w", len(buf), err)
 		}
 
@@ -1247,7 +1180,7 @@ func ProcessSearchQuery(qt *querytracer.Tracer, sq *storage.SearchQuery, deadlin
 
 	if err := sr.Error(); err != nil {
 		putTmpBlocksFile(tbf)
-		putStorageSearch(sr)
+		vmstorage.PutSearch(sr)
 		if errors.Is(err, storage.ErrDeadlineExceeded) {
 			return nil, fmt.Errorf("timeout exceeded during the query: %s", deadline.String())
 		}
@@ -1255,13 +1188,13 @@ func ProcessSearchQuery(qt *querytracer.Tracer, sq *storage.SearchQuery, deadlin
 	}
 	if err := tbf.Finalize(); err != nil {
 		putTmpBlocksFile(tbf)
-		putStorageSearch(sr)
+		vmstorage.PutSearch(sr)
 		return nil, fmt.Errorf("cannot finalize temporary file: %w", err)
 	}
 	qt.Printf("fetch unique series=%d, blocks=%d, samples=%d, bytes=%d", len(m), blocksRead, samples, tbf.Len())
 
 	var rss Results
-	rss.tr = tr
+	rss.tr = sq.GetTimeRange()
 	rss.deadline = deadline
 	pts := make([]packedTimeseries, len(orderedMetricNames))
 	for i, metricName := range orderedMetricNames {
@@ -1302,35 +1235,6 @@ func getBlockRefsEnd(a []blockRef) uintptr {
 	return uintptr(unsafe.Pointer(unsafe.SliceData(a))) + uintptr(len(a))*unsafe.Sizeof(blockRef{})
 }
 
-func setupTfss(qt *querytracer.Tracer, tr storage.TimeRange, tagFilterss [][]storage.TagFilter, maxMetrics int, deadline searchutil.Deadline) ([]*storage.TagFilters, error) {
-	tfss := make([]*storage.TagFilters, 0, len(tagFilterss))
-	for _, tagFilters := range tagFilterss {
-		tfs := storage.NewTagFilters()
-		for i := range tagFilters {
-			tf := &tagFilters[i]
-			if string(tf.Key) == "__graphite__" {
-				query := tf.Value
-				paths, err := vmstorage.SearchGraphitePaths(qt, tr, query, maxMetrics, deadline.Deadline())
-				if err != nil {
-					return nil, fmt.Errorf("error when searching for Graphite paths for query %q: %w", query, err)
-				}
-				if len(paths) >= maxMetrics {
-					return nil, fmt.Errorf("more than %d time series match Graphite query %q; "+
-						"either narrow down the query or increase the corresponding -search.max* command-line flag value; "+
-						"see https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#resource-usage-limits", maxMetrics, query)
-				}
-				tfs.AddGraphiteQuery(query, paths, tf.IsNegative)
-				continue
-			}
-			if err := tfs.Add(tf.Key, tf.Value, tf.IsNegative, tf.IsRegexp); err != nil {
-				return nil, fmt.Errorf("cannot parse tag filter %s: %w", tf, err)
-			}
-		}
-		tfss = append(tfss, tfs)
-	}
-	return tfss, nil
-}
-
 func applyGraphiteRegexpFilter(filter string, ss []string) ([]string, error) {
 	// Anchor filter regexp to the beginning of the string as Graphite does.
 	// See https://github.com/graphite-project/graphite-web/blob/3ad279df5cb90b211953e39161df416e54a84948/webapp/graphite/tags/localdatabase.py#L157
@@ -1357,13 +1261,12 @@ const maxFastAllocBlockSize = 32 * 1024
 func GetMetricNamesStats(qt *querytracer.Tracer, limit, le int, matchPattern string) (metricnamestats.StatsResult, error) {
 	qt = qt.NewChild("get metric names usage statistics with limit: %d, less or equal to: %d, match pattern=%q", limit, le, matchPattern)
 	defer qt.Done()
-	return vmstorage.GetMetricNamesStats(qt, limit, le, matchPattern)
+	return vmstorage.VMSelectAPI.GetMetricNamesUsageStats(qt, nil, limit, le, matchPattern, 0)
 }
 
 // ResetMetricNamesStats resets state of metric names usage
 func ResetMetricNamesStats(qt *querytracer.Tracer) error {
 	qt = qt.NewChild("reset metric names usage stats")
 	defer qt.Done()
-	vmstorage.ResetMetricNamesStats(qt)
-	return nil
+	return vmstorage.VMSelectAPI.ResetMetricNamesUsageStats(qt, 0)
 }

app/vmselect/prometheus/federate.qtpl

@@ -2,13 +2,16 @@
 	"math"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect/netstorage"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
+    "github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
+    "github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
 ) %}
 
 {% stripspace %}
 
 // Federate writes rs in /federate format.
 // See https://prometheus.io/docs/prometheus/latest/federation/
-{% func Federate(rs *netstorage.Result) %}
+{% func Federate(rs *netstorage.Result, escapeScheme string) %}
 	{% code
 		values := rs.Values
 		timestamps := rs.Timestamps
@@ -24,10 +27,54 @@
 			See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3185
 		{% endcomment %}
 		{% return %}
-	{% endif %}
-	{%= prometheusMetricName(&rs.MetricName) %}{% space %}
+    {% endif %}
+
+    {% switch escapeScheme %}
+       {% case federateEscapeSchemeUTF8 %}
+          {%= prometheusFederateMetricNameUTF8(&rs.MetricName) %}{% space %}
+
+       {% case federateEscapeSchemeUnderscore %}
+         {%= prometheusFederateMetricNameEscapeUnderscore(&rs.MetricName) %}{% space %}
+
+       {% case "" %}
+         {%= prometheusMetricName(&rs.MetricName) %}{% space %}
+    {% endswitch %}
+
 	{%f= lastValue %}{% space %}
 	{%dl= timestamps[len(timestamps)-1] %}{% newline %}
 {% endfunc %}
 
+{% func prometheusFederateMetricNameEscapeUnderscore(mn *storage.MetricName) %}
+	{%s= promrelabel.SanitizeMetricName(bytesutil.ToUnsafeString(mn.MetricGroup)) %}
+	{% if len(mn.Tags) > 0 %}
+	{
+		{% code tags := mn.Tags %}
+		{%s= promrelabel.SanitizeLabelName(bytesutil.ToUnsafeString(tags[0].Key)) %}={%= escapePrometheusLabel(tags[0].Value) %}
+		{% code tags = tags[1:] %}
+		{% for i := range tags %}
+			{% code tag := &tags[i] %}
+			,{%s= promrelabel.SanitizeLabelName(bytesutil.ToUnsafeString(tag.Key)) %}={%= escapePrometheusLabel(tag.Value) %}
+		{% endfor %}
+	}
+	{% endif %}
+{% endfunc %}
+
+{% func prometheusFederateMetricNameUTF8(mn *storage.MetricName) %}
+	{
+        {%= escapePrometheusLabel(mn.MetricGroup) %}
+        {% if len(mn.Tags) > 0 %}
+        ,
+		{% code tags := mn.Tags %}
+		{%= escapePrometheusLabel(tags[0].Key) %}={%= escapePrometheusLabel(tags[0].Value) %}
+		{% code tags = tags[1:] %}
+		{% for i := range tags %}
+			{% code tag := &tags[i] %}
+			,{%= escapePrometheusLabel(tag.Key) %}={%= escapePrometheusLabel(tag.Value) %}
+		{% endfor %}
+        {% endif %}
+	}
+{% endfunc %}
+
+
+
 {% endstripspace %}

app/vmselect/prometheus/federate.qtpl.go

@@ -9,82 +9,241 @@ import (
 	"math"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect/netstorage"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
 )
 
 // Federate writes rs in /federate format.// See https://prometheus.io/docs/prometheus/latest/federation/
 
-//line app/vmselect/prometheus/federate.qtpl:11
+//line app/vmselect/prometheus/federate.qtpl:14
 import (
 	qtio422016 "io"
 
 	qt422016 "github.com/valyala/quicktemplate"
 )
 
-//line app/vmselect/prometheus/federate.qtpl:11
+//line app/vmselect/prometheus/federate.qtpl:14
 var (
 	_ = qtio422016.Copy
 	_ = qt422016.AcquireByteBuffer
 )
 
-//line app/vmselect/prometheus/federate.qtpl:11
-func StreamFederate(qw422016 *qt422016.Writer, rs *netstorage.Result) {
-//line app/vmselect/prometheus/federate.qtpl:13
+//line app/vmselect/prometheus/federate.qtpl:14
+func StreamFederate(qw422016 *qt422016.Writer, rs *netstorage.Result, escapeScheme string) {
+//line app/vmselect/prometheus/federate.qtpl:16
 	values := rs.Values
 	timestamps := rs.Timestamps
 
-//line app/vmselect/prometheus/federate.qtpl:16
+//line app/vmselect/prometheus/federate.qtpl:19
 	if len(timestamps) == 0 || len(values) == 0 {
-//line app/vmselect/prometheus/federate.qtpl:16
+//line app/vmselect/prometheus/federate.qtpl:19
 		return
-//line app/vmselect/prometheus/federate.qtpl:16
+//line app/vmselect/prometheus/federate.qtpl:19
 	}
-//line app/vmselect/prometheus/federate.qtpl:18
+//line app/vmselect/prometheus/federate.qtpl:21
 	lastValue := values[len(values)-1]
 
-//line app/vmselect/prometheus/federate.qtpl:20
+//line app/vmselect/prometheus/federate.qtpl:23
 	if math.IsNaN(lastValue) {
-//line app/vmselect/prometheus/federate.qtpl:26
+//line app/vmselect/prometheus/federate.qtpl:29
 		return
-//line app/vmselect/prometheus/federate.qtpl:27
+//line app/vmselect/prometheus/federate.qtpl:30
 	}
-//line app/vmselect/prometheus/federate.qtpl:28
-	streamprometheusMetricName(qw422016, &rs.MetricName)
-//line app/vmselect/prometheus/federate.qtpl:28
-	qw422016.N().S(` `)
-//line app/vmselect/prometheus/federate.qtpl:29
+//line app/vmselect/prometheus/federate.qtpl:32
+	switch escapeScheme {
+//line app/vmselect/prometheus/federate.qtpl:33
+	case federateEscapeSchemeUTF8:
+//line app/vmselect/prometheus/federate.qtpl:34
+		streamprometheusFederateMetricNameUTF8(qw422016, &rs.MetricName)
+//line app/vmselect/prometheus/federate.qtpl:34
+		qw422016.N().S(` `)
+//line app/vmselect/prometheus/federate.qtpl:36
+	case federateEscapeSchemeUnderscore:
+//line app/vmselect/prometheus/federate.qtpl:37
+		streamprometheusFederateMetricNameEscapeUnderscore(qw422016, &rs.MetricName)
+//line app/vmselect/prometheus/federate.qtpl:37
+		qw422016.N().S(` `)
+//line app/vmselect/prometheus/federate.qtpl:39
+	case "":
+//line app/vmselect/prometheus/federate.qtpl:40
+		streamprometheusMetricName(qw422016, &rs.MetricName)
+//line app/vmselect/prometheus/federate.qtpl:40
+		qw422016.N().S(` `)
+//line app/vmselect/prometheus/federate.qtpl:41
+	}
+//line app/vmselect/prometheus/federate.qtpl:43
 	qw422016.N().F(lastValue)
-//line app/vmselect/prometheus/federate.qtpl:29
+//line app/vmselect/prometheus/federate.qtpl:43
 	qw422016.N().S(` `)
-//line app/vmselect/prometheus/federate.qtpl:30
+//line app/vmselect/prometheus/federate.qtpl:44
 	qw422016.N().DL(timestamps[len(timestamps)-1])
-//line app/vmselect/prometheus/federate.qtpl:30
+//line app/vmselect/prometheus/federate.qtpl:44
 	qw422016.N().S(`
 `)
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
 }
 
-//line app/vmselect/prometheus/federate.qtpl:31
-func WriteFederate(qq422016 qtio422016.Writer, rs *netstorage.Result) {
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
+func WriteFederate(qq422016 qtio422016.Writer, rs *netstorage.Result, escapeScheme string) {
+//line app/vmselect/prometheus/federate.qtpl:45
 	qw422016 := qt422016.AcquireWriter(qq422016)
-//line app/vmselect/prometheus/federate.qtpl:31
-	StreamFederate(qw422016, rs)
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
+	StreamFederate(qw422016, rs, escapeScheme)
+//line app/vmselect/prometheus/federate.qtpl:45
 	qt422016.ReleaseWriter(qw422016)
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
 }
 
-//line app/vmselect/prometheus/federate.qtpl:31
-func Federate(rs *netstorage.Result) string {
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
+func Federate(rs *netstorage.Result, escapeScheme string) string {
+//line app/vmselect/prometheus/federate.qtpl:45
 	qb422016 := qt422016.AcquireByteBuffer()
-//line app/vmselect/prometheus/federate.qtpl:31
-	WriteFederate(qb422016, rs)
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
+	WriteFederate(qb422016, rs, escapeScheme)
+//line app/vmselect/prometheus/federate.qtpl:45
 	qs422016 := string(qb422016.B)
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
 	qt422016.ReleaseByteBuffer(qb422016)
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
 	return qs422016
-//line app/vmselect/prometheus/federate.qtpl:31
+//line app/vmselect/prometheus/federate.qtpl:45
+}
+
+//line app/vmselect/prometheus/federate.qtpl:47
+func streamprometheusFederateMetricNameEscapeUnderscore(qw422016 *qt422016.Writer, mn *storage.MetricName) {
+//line app/vmselect/prometheus/federate.qtpl:48
+	qw422016.N().S(promrelabel.SanitizeMetricName(bytesutil.ToUnsafeString(mn.MetricGroup)))
+//line app/vmselect/prometheus/federate.qtpl:49
+	if len(mn.Tags) > 0 {
+//line app/vmselect/prometheus/federate.qtpl:49
+		qw422016.N().S(`{`)
+//line app/vmselect/prometheus/federate.qtpl:51
+		tags := mn.Tags
+
+//line app/vmselect/prometheus/federate.qtpl:52
+		qw422016.N().S(promrelabel.SanitizeLabelName(bytesutil.ToUnsafeString(tags[0].Key)))
+//line app/vmselect/prometheus/federate.qtpl:52
+		qw422016.N().S(`=`)
+//line app/vmselect/prometheus/federate.qtpl:52
+		streamescapePrometheusLabel(qw422016, tags[0].Value)
+//line app/vmselect/prometheus/federate.qtpl:53
+		tags = tags[1:]
+
+//line app/vmselect/prometheus/federate.qtpl:54
+		for i := range tags {
+//line app/vmselect/prometheus/federate.qtpl:55
+			tag := &tags[i]
+
+//line app/vmselect/prometheus/federate.qtpl:55
+			qw422016.N().S(`,`)
+//line app/vmselect/prometheus/federate.qtpl:56
+			qw422016.N().S(promrelabel.SanitizeLabelName(bytesutil.ToUnsafeString(tag.Key)))
+//line app/vmselect/prometheus/federate.qtpl:56
+			qw422016.N().S(`=`)
+//line app/vmselect/prometheus/federate.qtpl:56
+			streamescapePrometheusLabel(qw422016, tag.Value)
+//line app/vmselect/prometheus/federate.qtpl:57
+		}
+//line app/vmselect/prometheus/federate.qtpl:57
+		qw422016.N().S(`}`)
+//line app/vmselect/prometheus/federate.qtpl:59
+	}
+//line app/vmselect/prometheus/federate.qtpl:60
+}
+
+//line app/vmselect/prometheus/federate.qtpl:60
+func writeprometheusFederateMetricNameEscapeUnderscore(qq422016 qtio422016.Writer, mn *storage.MetricName) {
+//line app/vmselect/prometheus/federate.qtpl:60
+	qw422016 := qt422016.AcquireWriter(qq422016)
+//line app/vmselect/prometheus/federate.qtpl:60
+	streamprometheusFederateMetricNameEscapeUnderscore(qw422016, mn)
+//line app/vmselect/prometheus/federate.qtpl:60
+	qt422016.ReleaseWriter(qw422016)
+//line app/vmselect/prometheus/federate.qtpl:60
+}
+
+//line app/vmselect/prometheus/federate.qtpl:60
+func prometheusFederateMetricNameEscapeUnderscore(mn *storage.MetricName) string {
+//line app/vmselect/prometheus/federate.qtpl:60
+	qb422016 := qt422016.AcquireByteBuffer()
+//line app/vmselect/prometheus/federate.qtpl:60
+	writeprometheusFederateMetricNameEscapeUnderscore(qb422016, mn)
+//line app/vmselect/prometheus/federate.qtpl:60
+	qs422016 := string(qb422016.B)
+//line app/vmselect/prometheus/federate.qtpl:60
+	qt422016.ReleaseByteBuffer(qb422016)
+//line app/vmselect/prometheus/federate.qtpl:60
+	return qs422016
+//line app/vmselect/prometheus/federate.qtpl:60
+}
+
+//line app/vmselect/prometheus/federate.qtpl:62
+func streamprometheusFederateMetricNameUTF8(qw422016 *qt422016.Writer, mn *storage.MetricName) {
+//line app/vmselect/prometheus/federate.qtpl:62
+	qw422016.N().S(`{`)
+//line app/vmselect/prometheus/federate.qtpl:64
+	streamescapePrometheusLabel(qw422016, mn.MetricGroup)
+//line app/vmselect/prometheus/federate.qtpl:65
+	if len(mn.Tags) > 0 {
+//line app/vmselect/prometheus/federate.qtpl:65
+		qw422016.N().S(`,`)
+//line app/vmselect/prometheus/federate.qtpl:67
+		tags := mn.Tags
+
+//line app/vmselect/prometheus/federate.qtpl:68
+		streamescapePrometheusLabel(qw422016, tags[0].Key)
+//line app/vmselect/prometheus/federate.qtpl:68
+		qw422016.N().S(`=`)
+//line app/vmselect/prometheus/federate.qtpl:68
+		streamescapePrometheusLabel(qw422016, tags[0].Value)
+//line app/vmselect/prometheus/federate.qtpl:69
+		tags = tags[1:]
+
+//line app/vmselect/prometheus/federate.qtpl:70
+		for i := range tags {
+//line app/vmselect/prometheus/federate.qtpl:71
+			tag := &tags[i]
+
+//line app/vmselect/prometheus/federate.qtpl:71
+			qw422016.N().S(`,`)
+//line app/vmselect/prometheus/federate.qtpl:72
+			streamescapePrometheusLabel(qw422016, tag.Key)
+//line app/vmselect/prometheus/federate.qtpl:72
+			qw422016.N().S(`=`)
+//line app/vmselect/prometheus/federate.qtpl:72
+			streamescapePrometheusLabel(qw422016, tag.Value)
+//line app/vmselect/prometheus/federate.qtpl:73
+		}
+//line app/vmselect/prometheus/federate.qtpl:74
+	}
+//line app/vmselect/prometheus/federate.qtpl:74
+	qw422016.N().S(`}`)
+//line app/vmselect/prometheus/federate.qtpl:76
+}
+
+//line app/vmselect/prometheus/federate.qtpl:76
+func writeprometheusFederateMetricNameUTF8(qq422016 qtio422016.Writer, mn *storage.MetricName) {
+//line app/vmselect/prometheus/federate.qtpl:76
+	qw422016 := qt422016.AcquireWriter(qq422016)
+//line app/vmselect/prometheus/federate.qtpl:76
+	streamprometheusFederateMetricNameUTF8(qw422016, mn)
+//line app/vmselect/prometheus/federate.qtpl:76
+	qt422016.ReleaseWriter(qw422016)
+//line app/vmselect/prometheus/federate.qtpl:76
+}
+
+//line app/vmselect/prometheus/federate.qtpl:76
+func prometheusFederateMetricNameUTF8(mn *storage.MetricName) string {
+//line app/vmselect/prometheus/federate.qtpl:76
+	qb422016 := qt422016.AcquireByteBuffer()
+//line app/vmselect/prometheus/federate.qtpl:76
+	writeprometheusFederateMetricNameUTF8(qb422016, mn)
+//line app/vmselect/prometheus/federate.qtpl:76
+	qs422016 := string(qb422016.B)
+//line app/vmselect/prometheus/federate.qtpl:76
+	qt422016.ReleaseByteBuffer(qb422016)
+//line app/vmselect/prometheus/federate.qtpl:76
+	return qs422016
+//line app/vmselect/prometheus/federate.qtpl:76
 }

app/vmselect/prometheus/federate_test.go

@@ -8,15 +8,15 @@ import (
 )
 
 func TestFederate(t *testing.T) {
-	f := func(rs *netstorage.Result, expectedResult string) {
+	f := func(rs *netstorage.Result, escapeScheme string, expectedResult string) {
 		t.Helper()
-		result := Federate(rs)
+		result := Federate(rs, escapeScheme)
 		if result != expectedResult {
 			t.Fatalf("unexpected result; got\n%s\nwant\n%s", result, expectedResult)
 		}
 	}
 
-	f(&netstorage.Result{}, ``)
+	f(&netstorage.Result{}, ``, ``)
 
 	f(&netstorage.Result{
 		MetricName: storage.MetricName{
@@ -39,5 +39,60 @@ func TestFederate(t *testing.T) {
 		},
 		Values:     []float64{1.23},
 		Timestamps: []int64{123},
-	}, `foo{a="b",qqq="\\",abc="a<b\"\\c"} 1.23 123`+"\n")
+	}, ``, `foo{a="b",qqq="\\",abc="a<b\"\\c"} 1.23 123`+"\n")
+
+	f(&netstorage.Result{
+		MetricName: storage.MetricName{
+			MetricGroup: []byte("foo.bar"),
+			Tags: []storage.Tag{
+				{
+					Key:   []byte("some.!other"),
+					Value: []byte("value.unchanged!."),
+				},
+				{
+					Key:   []byte("qqq"),
+					Value: []byte("\\"),
+				},
+				{
+					Key:   []byte("!key"),
+					Value: []byte("value"),
+				},
+				{
+					Key: []byte("abc"),
+					// Verify that < isn't encoded. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5431
+					Value: []byte("a<b\"\\c"),
+				},
+			},
+		},
+		Values:     []float64{1.23},
+		Timestamps: []int64{123},
+	}, federateEscapeSchemeUnderscore, `foo_bar{some__other="value.unchanged!.",qqq="\\",_key="value",abc="a<b\"\\c"} 1.23 123`+"\n")
+
+	f(&netstorage.Result{
+		MetricName: storage.MetricName{
+			MetricGroup: []byte("foo.bar"),
+			Tags: []storage.Tag{
+				{
+					Key:   []byte("some.!other"),
+					Value: []byte("value.unchanged!."),
+				},
+				{
+					Key:   []byte("qqq"),
+					Value: []byte("\\"),
+				},
+				{
+					Key:   []byte("!key"),
+					Value: []byte("value"),
+				},
+				{
+					Key: []byte(`ab"c`),
+					// Verify that < isn't encoded. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5431
+					Value: []byte("a<b\"\\c"),
+				},
+			},
+		},
+		Values:     []float64{1.23},
+		Timestamps: []int64{123},
+	}, federateEscapeSchemeUTF8, `{"foo.bar","some.!other"="value.unchanged!.","qqq"="\\","!key"="value","ab\"c"="a<b\"\\c"} 1.23 123`+"\n")
+
 }

app/vmselect/prometheus/federate_timing_test.go

@@ -9,16 +9,17 @@ import (
 )
 
 func BenchmarkFederate(b *testing.B) {
+
 	rs := &netstorage.Result{
 		MetricName: storage.MetricName{
-			MetricGroup: []byte("foo_bar_bazaaaa_total"),
+			MetricGroup: []byte("foo_bar_?_._bazaaaa_total"),
 			Tags: []storage.Tag{
 				{
-					Key:   []byte("instance"),
+					Key:   []byte("instance:job"),
 					Value: []byte("foobarbaz:2344"),
 				},
 				{
-					Key:   []byte("job"),
+					Key:   []byte("job.name"),
 					Value: []byte("aaabbbccc"),
 				},
 			},
@@ -27,12 +28,22 @@ func BenchmarkFederate(b *testing.B) {
 		Timestamps: []int64{1234567890},
 	}
 
-	b.ReportAllocs()
-	b.RunParallel(func(pb *testing.PB) {
-		var bb bytes.Buffer
-		for pb.Next() {
-			bb.Reset()
-			WriteFederate(&bb, rs)
-		}
-	})
+	f := func(name, escapeScheme string) {
+		b.Helper()
+
+		b.Run(name, func(b *testing.B) {
+			b.ReportAllocs()
+			b.RunParallel(func(pb *testing.PB) {
+				var bb bytes.Buffer
+				for pb.Next() {
+					bb.Reset()
+					WriteFederate(&bb, rs, escapeScheme)
+				}
+			})
+		})
+	}
+
+	f("without escape", "")
+	f("allow-utf-8", federateEscapeSchemeUTF8)
+	f("legacy-underscore", federateEscapeSchemeUnderscore)
 }

app/vmselect/prometheus/prometheus.go

@@ -28,8 +28,6 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/httpserver"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/httputil"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/memory"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/netutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/querytracer"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
@@ -50,9 +48,6 @@ var (
 		"If set to true, the query model becomes closer to InfluxDB data model. If set to true, then -search.maxLookback and -search.maxStalenessInterval are ignored")
 	maxStepForPointsAdjustment = flag.Duration("search.maxStepForPointsAdjustment", time.Minute, "The maximum step when /api/v1/query_range handler adjusts "+
 		"points with timestamps closer than -search.latencyOffset to the current time. The adjustment is needed because such points may contain incomplete data")
-
-	maxUniqueTimeseries = flag.Int("search.maxUniqueTimeseries", 0, "The maximum number of unique time series, which can be selected during /api/v1/query and /api/v1/query_range queries. This option allows limiting memory usage. "+
-		"When set to zero, the limit is automatically calculated based on -search.maxConcurrentRequests (inversely proportional) and memory available to the process (proportional).")
 	maxFederateSeries       = flag.Int("search.maxFederateSeries", 1e6, "The maximum number of time series, which can be returned from /federate. This option allows limiting memory usage")
 	maxExportSeries         = flag.Int("search.maxExportSeries", 10e6, "The maximum number of time series, which can be returned from /api/v1/export* APIs. This option allows limiting memory usage")
 	maxTSDBStatusSeries     = flag.Int("search.maxTSDBStatusSeries", 10e6, "The maximum number of time series, which can be processed during the call to /api/v1/status/tsdb. This option allows limiting memory usage")
@@ -108,6 +103,11 @@ func PrettifyQuery(w http.ResponseWriter, r *http.Request) {
 	_ = bw.Flush()
 }
 
+const (
+	federateEscapeSchemeUnderscore = "underscore"
+	federateEscapeSchemeUTF8       = "utf-8"
+)
+
 // FederateHandler implements /federate . See https://prometheus.io/docs/prometheus/latest/federation/
 func FederateHandler(startTime time.Time, w http.ResponseWriter, r *http.Request) error {
 	defer federateDuration.UpdateDuration(startTime)
@@ -132,6 +132,21 @@ func FederateHandler(startTime time.Time, w http.ResponseWriter, r *http.Request
 		return fmt.Errorf("cannot fetch data for %q: %w", sq, err)
 	}
 
+	// add best-effort format negotiation
+	// modern version of Prometheus always set allow-utf-8 in order to properly parse utf-8 names and labels
+	// prometheus below v3 uses underscore escaping by default and it's the most common standard
+	var escapeScheme string
+	accept := r.Header.Get("Accept")
+	if len(accept) > 0 && strings.Contains(accept, "allow-utf-8") {
+		escapeScheme = federateEscapeSchemeUTF8
+	}
+	// try fallback to legacy underscore escaping if needed for Prometheus only,
+	// it's not widely used after Prometheus v3.0 release
+	// most of the Prometheus scrapers already use allow-utf-8 header
+	isPrometheus := strings.HasPrefix(r.UserAgent(), "Prometheus")
+	if len(escapeScheme) == 0 && isPrometheus {
+		escapeScheme = federateEscapeSchemeUnderscore
+	}
 	w.Header().Set("Content-Type", "text/plain; charset=utf-8")
 	bw := bufferedwriter.Get(w)
 	defer bufferedwriter.Put(bw)
@@ -141,7 +156,7 @@ func FederateHandler(startTime time.Time, w http.ResponseWriter, r *http.Request
 			return err
 		}
 		bb := sw.getBuffer(workerID)
-		WriteFederate(bb, rs)
+		WriteFederate(bb, rs, escapeScheme)
 		return sw.maybeFlushBuffer(bb)
 	})
 	if err == nil {
@@ -853,7 +868,7 @@ func QueryHandler(qt *querytracer.Tracer, startTime time.Time, w http.ResponseWr
 		End:                 start,
 		Step:                step,
 		MaxPointsPerSeries:  *maxPointsPerTimeseries,
-		MaxSeries:           GetMaxUniqueTimeSeries(),
+		MaxSeries:           0, // let vmstorage use maxUniqueTimeseries by default
 		QuotedRemoteAddr:    httpserver.GetQuotedRemoteAddr(r),
 		Deadline:            deadline,
 		MayCache:            mayCache,
@@ -964,7 +979,7 @@ func queryRangeHandler(qt *querytracer.Tracer, startTime time.Time, w http.Respo
 		End:                 end,
 		Step:                step,
 		MaxPointsPerSeries:  *maxPointsPerTimeseries,
-		MaxSeries:           GetMaxUniqueTimeSeries(),
+		MaxSeries:           0, // let vmstorage use maxUniqueTimeseries by default
 		QuotedRemoteAddr:    httpserver.GetQuotedRemoteAddr(r),
 		Deadline:            deadline,
 		MayCache:            mayCache,
@@ -1300,43 +1315,6 @@ func (sw *scalableWriter) flush() error {
 	return sw.bw.Flush()
 }
 
-var (
-	maxUniqueTimeseriesValueOnce sync.Once
-	maxUniqueTimeseriesValue     int
-)
-
-// InitMaxUniqueTimeseries init the max metrics limit calculated by available resources.
-// The calculation is split into calculateMaxUniqueTimeSeriesForResource for unit testing.
-func InitMaxUniqueTimeseries(maxConcurrentRequests int) {
-	maxUniqueTimeseriesValueOnce.Do(func() {
-		maxUniqueTimeseriesValue = *maxUniqueTimeseries
-		if maxUniqueTimeseriesValue <= 0 {
-			maxUniqueTimeseriesValue = calculateMaxUniqueTimeSeriesForResource(maxConcurrentRequests, memory.Remaining())
-		}
-	})
-}
-
-// calculateMaxUniqueTimeSeriesForResource calculate the max metrics limit calculated by available resources.
-func calculateMaxUniqueTimeSeriesForResource(maxConcurrentRequests, remainingMemory int) int {
-	if maxConcurrentRequests <= 0 {
-		// This line should NOT be reached unless the user has set an incorrect `search.maxConcurrentRequests`.
-		// In such cases, fallback to unlimited.
-		logger.Warnf("limiting -search.maxUniqueTimeseries to %v because -search.maxConcurrentRequests=%d.", 2e9, maxConcurrentRequests)
-		return 2e9
-	}
-
-	// Calculate the max metrics limit for a single request in the worst-case concurrent scenario.
-	// The approximate size of 1 unique series that could occupy in the vmstorage is 200 bytes.
-	mts := remainingMemory / 200 / maxConcurrentRequests
-	logger.Infof("limiting -search.maxUniqueTimeseries to %d according to -search.maxConcurrentRequests=%d and remaining memory=%d bytes. To increase the limit, reduce -search.maxConcurrentRequests or increase memory available to the process.", mts, maxConcurrentRequests, remainingMemory)
-	return mts
-}
-
-// GetMaxUniqueTimeSeries returns the max metrics limit calculated by available resources.
-func GetMaxUniqueTimeSeries() int {
-	return maxUniqueTimeseriesValue
-}
-
 // copied from https://github.com/prometheus/common/blob/adea6285c1c7447fcb7bfdeb6abfc6eff893e0a7/model/metric.go#L483
 // it's not possible to use direct import due to increased binary size
 func unescapePrometheusLabelName(name string) string {

app/vmselect/prometheus/prometheus_test.go

@@ -4,7 +4,6 @@ import (
 	"math"
 	"net/http"
 	"reflect"
-	"runtime"
 	"testing"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect/netstorage"
@@ -230,29 +229,3 @@ func TestGetLatencyOffsetMillisecondsFailure(t *testing.T) {
 	}
 	f("http://localhost?latency_offset=foobar")
 }
-
-func TestCalculateMaxMetricsLimitByResource(t *testing.T) {
-	f := func(maxConcurrentRequest, remainingMemory, expect int) {
-		t.Helper()
-		maxMetricsLimit := calculateMaxUniqueTimeSeriesForResource(maxConcurrentRequest, remainingMemory)
-		if maxMetricsLimit != expect {
-			t.Fatalf("unexpected max metrics limit: got %d, want %d", maxMetricsLimit, expect)
-		}
-	}
-
-	// Skip when GOARCH=386
-	if runtime.GOARCH != "386" {
-		// 8 CPU & 32 GiB
-		f(16, int(math.Round(32*1024*1024*1024*0.4)), 4294967)
-		// 4 CPU & 32 GiB
-		f(8, int(math.Round(32*1024*1024*1024*0.4)), 8589934)
-	}
-
-	// 2 CPU & 4 GiB
-	f(4, int(math.Round(4*1024*1024*1024*0.4)), 2147483)
-
-	// other edge cases
-	f(0, int(math.Round(4*1024*1024*1024*0.4)), 2e9)
-	f(4, 0, 0)
-
-}

app/vmselect/promql/rollup.go

@@ -2439,8 +2439,15 @@ func rollupIntegrate(rfa *rollupFuncArg) float64 {
 		prevTimestamp = timestamp
 		prevValue = v
 	}
-	dt := float64(rfa.currTimestamp-prevTimestamp) / 1e3
-	sum += prevValue * dt
+	// Only extrapolate the last value through to currTimestamp when the time
+	// series has any sample after the lookbehind window. When realNextValue is
+	// NaN the series has effectively ended at prevTimestamp, so accruing area
+	// past it would overcount the integral.
+	// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/9474
+	if !math.IsNaN(rfa.realNextValue) {
+		dt := float64(rfa.currTimestamp-prevTimestamp) / 1e3
+		sum += prevValue * dt
+	}
 	return sum
 }
 

app/vmselect/promql/rollup_result_cache.go

@@ -184,7 +184,7 @@ func InitRollupResultCache(cachePath string) {
 		rollupResultCacheRequests:    metrics.GetOrCreateCounter(`vm_rollup_result_cache_requests_total`),
 		rollupResultCacheFullHits:    metrics.GetOrCreateCounter(`vm_rollup_result_cache_full_hits_total`),
 		rollupResultCachePartialHits: metrics.GetOrCreateCounter(`vm_rollup_result_cache_partial_hits_total`),
-		rollupResultCacheMisses:      metrics.GetOrCreateCounter(`vm_rollup_result_cache_misses_total`),
+		rollupResultCacheMisses:      metrics.GetOrCreateCounter(`vm_rollup_result_cache_miss_total`),
 		rollupResultCacheResets:      metrics.GetOrCreateCounter(`vm_rollup_result_cache_resets_total`),
 	}
 }

app/vmselect/promql/rollup_test.go

@@ -1385,10 +1385,65 @@ func TestRollupFuncsNoWindow(t *testing.T) {
 		if samplesScanned != 24 {
 			t.Fatalf("expecting 24 samplesScanned from rollupConfig.Do; got %d", samplesScanned)
 		}
-		valuesExpected := []float64{nan, 2.148, 1.593, 1.156, 1.36}
+		// At tEnd=160 the series has no samples past the window (last sample is at
+		// ts=130), so integrate() must not extrapolate prevValue through tEnd.
+		// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/9474
+		valuesExpected := []float64{nan, 2.148, 1.593, 1.156, 0.34}
 		timestampsExpected := []int64{0, 40, 80, 120, 160}
 		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
 	})
+	t.Run("integrate_past_series_end", func(t *testing.T) {
+		// Constant series of value 1.0 from t=0..3600s (1h) at 60s step.
+		// Query integrate(metric[1h]) across t=0..10800s with 600s step.
+		// For t=0..3600s the window overlap with the data is [0,t], so the integral grows from 0 to 3600 (seconds).
+		// After the series ends, integrate must NOT keep accruing 3600 — it
+		// should taper to 0 once the lookbehind window is entirely past the
+		// last sample.
+		// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/9474
+		var testValues []int64
+		var testTimestamps []float64
+		for t := int64(0); t <= 3600_000; t += 60_000 {
+			testValues = append(testValues, t)
+			testTimestamps = append(testTimestamps, 1.0)
+		}
+		rc := rollupConfig{
+			Func:               rollupIntegrate,
+			Start:              0,
+			End:                10800_000,
+			Step:               600_000,
+			Window:             3600_000,
+			MaxPointsPerSeries: 1e4,
+		}
+		rc.Timestamps = rc.getTimestamps()
+		values, _ := rc.Do(nil, testTimestamps, testValues)
+		for i, ti := range rc.Timestamps {
+			v := values[i]
+
+			// For t<=3600s: window overlap is [0,ti], integral equals ti in seconds.
+			if ti <= 3600_000 {
+				expV := float64(ti / 1e3)
+				if v != expV {
+					t.Fatalf("unexpected integrate result at t=%ds, want=%.3f got=%.3f", ti/1e3, expV, v)
+				}
+				continue
+			}
+			// For 3600s<t<7200s: data is partially outside the window, so the
+			// integral shrinks linearly from 3600 to 0 as t approaches 7200s.
+			if ti > 3600_000 && ti < 7200_000 {
+				expV := float64((7200_000 - ti) / 1e3)
+				if v != expV {
+					t.Fatalf("unexpected integrate result at t=%ds, want=%.3f got=%.3f", ti/1e3, expV, v)
+				}
+				continue
+			}
+			if ti >= 7200_000 {
+				// Window entirely past data end: must be NaN.
+				if !math.IsNaN(v) {
+					t.Fatalf("unexpected integrate result at t=%ds, want=NaN got=%.3f", ti/1e3, v)
+				}
+			}
+		}
+	})
 	t.Run("distinct_over_time_1", func(t *testing.T) {
 		rc := rollupConfig{
 			Func:               rollupDistinct,

app/vmselect/vmui/index.html

@@ -37,11 +37,11 @@
   <meta property="og:title" content="UI for VictoriaMetrics">
   <meta property="og:url" content="https://victoriametrics.com/">
   <meta property="og:description" content="Explore and troubleshoot your VictoriaMetrics data">
-  <script type="module" crossorigin src="./assets/index-BjJ7fDL7.js"></script>
+  <script type="module" crossorigin src="./assets/index-CoGukb-x.js"></script>
   <link rel="modulepreload" crossorigin href="./assets/rolldown-runtime-COnpUsM8.js">
   <link rel="modulepreload" crossorigin href="./assets/vendor-C8Kwp93_.js">
   <link rel="stylesheet" crossorigin href="./assets/vendor-CnsZ1jie.css">
-  <link rel="stylesheet" crossorigin href="./assets/index-BL7jEFBa.css">
+  <link rel="stylesheet" crossorigin href="./assets/index-BBUnmLOr.css">
 </head>
 <body>
 <noscript>You need to enable JavaScript to run this app.</noscript>

app/vmstorage/main.go

@@ -1,7 +1,6 @@
 package vmstorage
 
 import (
-	"errors"
 	"flag"
 	"fmt"
 	"io"
@@ -9,12 +8,10 @@ import (
 	"net/http"
 	"strconv"
 	"strings"
-	"sync"
 	"time"
 
 	"github.com/VictoriaMetrics/metrics"
 
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
@@ -23,11 +20,9 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/mergeset"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/querytracer"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricnamestats"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricsmetadata"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/stringsutil"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/syncwg"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/timeutil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/vminsertapi"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/vmselectapi"
 )
 
 var (
@@ -39,11 +34,8 @@ var (
 	snapshotAuthKey   = flagutil.NewPassword("snapshotAuthKey", "authKey, which must be passed in query string to /snapshot* pages. It overrides -httpAuth.*")
 	forceMergeAuthKey = flagutil.NewPassword("forceMergeAuthKey", "authKey, which must be passed in query string to /internal/force_merge pages. It overrides -httpAuth.*")
 	forceFlushAuthKey = flagutil.NewPassword("forceFlushAuthKey", "authKey, which must be passed in query string to /internal/force_flush pages. It overrides -httpAuth.*")
-	snapshotsMaxAge   = flagutil.NewRetentionDuration("snapshotsMaxAge", "3d", "Automatically delete snapshots older than -snapshotsMaxAge if it is set to non-zero duration. Make sure that backup process has enough time to finish the backup before the corresponding snapshot is automatically deleted")
 	_                 = flag.Duration("snapshotCreateTimeout", 0, "Deprecated: this flag does nothing")
 
-	precisionBits = flag.Int("precisionBits", 64, "The number of precision bits to store per each value. Lower precision bits improves data compression at the cost of precision loss")
-
 	_ = flag.Duration("finalMergeDelay", 0, "Deprecated: this flag does nothing")
 	_ = flag.Int("bigMergeConcurrency", 0, "Deprecated: this flag does nothing")
 	_ = flag.Int("smallMergeConcurrency", 0, "Deprecated: this flag does nothing")
@@ -117,11 +109,7 @@ func DataPath() string {
 }
 
 // Init initializes vmstorage.
-func Init(resetCacheIfNeeded func(mrs []storage.MetricRow)) {
-	if err := encoding.CheckPrecisionBits(uint8(*precisionBits)); err != nil {
-		logger.Fatalf("invalid `-precisionBits`: %s", err)
-	}
-
+func Init(vmselectMaxConcurrentRequests int, resetCacheIfNeeded func(mrs []storage.MetricRow)) {
 	storage.SetDedupInterval(*minScrapeInterval)
 	storage.SetDataFlushInterval(*inmemoryDataFlushInterval)
 	storage.LegacySetRetentionTimezoneOffset(*retentionTimezoneOffset)
@@ -165,7 +153,7 @@ func Init(resetCacheIfNeeded func(mrs []storage.MetricRow)) {
 		LogNewSeries:                *logNewSeries,
 	}
 	strg := storage.MustOpenStorage(*storageDataPath, opts)
-	initStaleSnapshotsRemover(strg)
+	vmStorage = newVMStorageSingleNode(strg, vmselectMaxConcurrentRequests, resetCacheIfNeeded)
 
 	var m storage.Metrics
 	strg.UpdateMetrics(&m)
@@ -179,151 +167,32 @@ func Init(resetCacheIfNeeded func(mrs []storage.MetricRow)) {
 
 	// register storage metrics
 	storageMetrics = metrics.NewSet()
-	storageMetrics.RegisterMetricsWriter(func(w io.Writer) {
-		writeStorageMetrics(w, strg)
-	})
+	storageMetrics.RegisterMetricsWriter(vmStorage.writeStorageMetrics)
 	metrics.RegisterSet(storageMetrics)
 
-	WG = syncwg.WaitGroup{}
-	resetResponseCacheIfNeeded = resetCacheIfNeeded
-	Storage = strg
+	VMInsertAPI = vmStorage
+	VMSelectAPI = vmStorage
+	GetSearch = vmStorage.GetSearch
+	PutSearch = vmStorage.PutSearch
+	RequestHandler = vmStorage.requestHandler
+	DebugFlush = vmStorage.vms.s.DebugFlush
 }
 
 var storageMetrics *metrics.Set
 
-// Storage is a storage.
-//
-// Every storage call must be wrapped into WG.Add(1) ... WG.Done()
-// for proper graceful shutdown when Stop is called.
-var Storage *storage.Storage
+var (
+	// vmStorageSingleNode is an instance of vmstorage used by vminsert and
+	// vmselect for writing and reading data.
+	vmStorage      *VMStorageSingleNode
+	VMInsertAPI    vminsertapi.API
+	VMSelectAPI    vmselectapi.API
+	GetSearch      func(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (*storage.Search, int, error)
+	PutSearch      func(sr *storage.Search)
+	RequestHandler func(w http.ResponseWriter, r *http.Request) bool
 
-// WG must be incremented before Storage call.
-//
-// Use syncwg instead of sync, since Add is called from concurrent goroutines.
-var WG syncwg.WaitGroup
-
-// resetResponseCacheIfNeeded is a callback for automatic resetting of response cache if needed.
-var resetResponseCacheIfNeeded func(mrs []storage.MetricRow)
-
-// AddRows adds mrs to the storage.
-//
-// The caller should limit the number of concurrent calls to AddRows() in order to limit memory usage.
-func AddRows(mrs []storage.MetricRow) error {
-	if Storage.IsReadOnly() {
-		return errReadOnly
-	}
-	resetResponseCacheIfNeeded(mrs)
-	WG.Add(1)
-	Storage.AddRows(mrs, uint8(*precisionBits))
-	WG.Done()
-	return nil
-}
-
-// AddMetadataRows adds mrs to the storage.
-//
-// The caller should limit the number of concurrent calls to AddMetadataRows() in order to limit memory usage.
-func AddMetadataRows(mms []metricsmetadata.Row) error {
-	if Storage.IsReadOnly() {
-		return errReadOnly
-	}
-	WG.Add(1)
-	Storage.AddMetadataRows(mms)
-	WG.Done()
-	return nil
-}
-
-var errReadOnly = errors.New("the storage is in read-only mode; check -storage.minFreeDiskSpaceBytes command-line flag value")
-
-// RegisterMetricNames registers all the metrics from mrs in the storage.
-func RegisterMetricNames(qt *querytracer.Tracer, mrs []storage.MetricRow) {
-	WG.Add(1)
-	Storage.RegisterMetricNames(qt, mrs)
-	WG.Done()
-}
-
-// DeleteSeries deletes series matching tfss.
-//
-// Returns the number of deleted series.
-func DeleteSeries(qt *querytracer.Tracer, tfss []*storage.TagFilters, maxMetrics int) (int, error) {
-	WG.Add(1)
-	n, err := Storage.DeleteSeries(qt, tfss, maxMetrics)
-	WG.Done()
-	return n, err
-}
-
-// GetMetricNamesStats returns metric names usage stats with give limit and lte predicate
-func GetMetricNamesStats(qt *querytracer.Tracer, limit, le int, matchPattern string) (metricnamestats.StatsResult, error) {
-	WG.Add(1)
-	r := Storage.GetMetricNamesStats(qt, limit, le, matchPattern)
-	WG.Done()
-	return r, nil
-}
-
-// ResetMetricNamesStats resets state for metric names usage tracker
-func ResetMetricNamesStats(qt *querytracer.Tracer) {
-	WG.Add(1)
-	Storage.ResetMetricNamesStats(qt)
-	WG.Done()
-}
-
-// SearchMetricNames returns metric names for the given tfss on the given tr.
-func SearchMetricNames(qt *querytracer.Tracer, tfss []*storage.TagFilters, tr storage.TimeRange, maxMetrics int, deadline uint64) ([]string, error) {
-	WG.Add(1)
-	metricNames, err := Storage.SearchMetricNames(qt, tfss, tr, maxMetrics, deadline)
-	WG.Done()
-	return metricNames, err
-}
-
-// SearchLabelNames searches for tag keys matching the given tfss on tr.
-func SearchLabelNames(qt *querytracer.Tracer, tfss []*storage.TagFilters, tr storage.TimeRange, maxTagKeys, maxMetrics int, deadline uint64) ([]string, error) {
-	WG.Add(1)
-	labelNames, err := Storage.SearchLabelNames(qt, tfss, tr, maxTagKeys, maxMetrics, deadline)
-	WG.Done()
-	return labelNames, err
-}
-
-// SearchLabelValues searches for label values for the given labelName, tfss and
-// tr.
-func SearchLabelValues(qt *querytracer.Tracer, labelName string, tfss []*storage.TagFilters, tr storage.TimeRange, maxLabelValues, maxMetrics int, deadline uint64) ([]string, error) {
-	WG.Add(1)
-	labelValues, err := Storage.SearchLabelValues(qt, labelName, tfss, tr, maxLabelValues, maxMetrics, deadline)
-	WG.Done()
-	return labelValues, err
-}
-
-// SearchTagValueSuffixes returns all the tag value suffixes for the given tagKey and tagValuePrefix on the given tr.
-//
-// This allows implementing https://graphite-api.readthedocs.io/en/latest/api.html#metrics-find or similar APIs.
-func SearchTagValueSuffixes(qt *querytracer.Tracer, tr storage.TimeRange, tagKey, tagValuePrefix string, delimiter byte, maxTagValueSuffixes int, deadline uint64) ([]string, error) {
-	WG.Add(1)
-	suffixes, err := Storage.SearchTagValueSuffixes(qt, tr, tagKey, tagValuePrefix, delimiter, maxTagValueSuffixes, deadline)
-	WG.Done()
-	return suffixes, err
-}
-
-// SearchGraphitePaths returns all the metric names matching the given Graphite query.
-func SearchGraphitePaths(qt *querytracer.Tracer, tr storage.TimeRange, query []byte, maxPaths int, deadline uint64) ([]string, error) {
-	WG.Add(1)
-	paths, err := Storage.SearchGraphitePaths(qt, tr, query, maxPaths, deadline)
-	WG.Done()
-	return paths, err
-}
-
-// GetTSDBStatus returns TSDB status for given filters on the given date.
-func GetTSDBStatus(qt *querytracer.Tracer, tfss []*storage.TagFilters, date uint64, focusLabel string, topN, maxMetrics int, deadline uint64) (*storage.TSDBStatus, error) {
-	WG.Add(1)
-	status, err := Storage.GetTSDBStatus(qt, tfss, date, focusLabel, topN, maxMetrics, deadline)
-	WG.Done()
-	return status, err
-}
-
-// GetSeriesCount returns the number of time series in the storage.
-func GetSeriesCount(deadline uint64) (uint64, error) {
-	WG.Add(1)
-	n, err := Storage.GetSeriesCount(deadline)
-	WG.Done()
-	return n, err
-}
+	// TODO(@rtm0): Remove this dependency from vmalert-tool unit tests.
+	DebugFlush func()
+)
 
 // Stop stops the vmstorage
 func Stop() {
@@ -333,17 +202,22 @@ func Stop() {
 
 	logger.Infof("gracefully closing the storage at %s", *storageDataPath)
 	startTime := time.Now()
-	WG.WaitAndBlock()
-	stopStaleSnapshotsRemover()
-	Storage.MustClose()
+	vmStorage.Stop()
 	logger.Infof("successfully closed the storage in %.3f seconds", time.Since(startTime).Seconds())
 
 	fs.MustStopDirRemover()
-	logger.Infof("the storage has been stopped")
+	logger.Infof("the vmstorage has been stopped")
 }
 
-// RequestHandler is a storage request handler.
-func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
+func (vmssn *VMStorageSingleNode) requestHandler(w http.ResponseWriter, r *http.Request) bool {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.requestHandler(w, r)
+}
+
+// requestHandler is a storage request handler.
+// TODO(@rtm0): Move to a separate file, request_handler.go
+func (vms *VMStorage) requestHandler(w http.ResponseWriter, r *http.Request) bool {
 	path := r.URL.Path
 	if path == "/internal/force_merge" {
 		if !httpserver.CheckAuthFlag(w, r, forceMergeAuthKey) {
@@ -356,7 +230,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 			defer activeForceMerges.Dec()
 			logger.Infof("forced merge for partition_prefix=%q has been started", partitionNamePrefix)
 			startTime := time.Now()
-			if err := Storage.ForceMergePartitions(partitionNamePrefix); err != nil {
+			if err := vms.s.ForceMergePartitions(partitionNamePrefix); err != nil {
 				logger.Errorf("error in forced merge for partition_prefix=%q: %s", partitionNamePrefix, err)
 				return
 			}
@@ -369,7 +243,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 			return true
 		}
 		logger.Infof("flushing storage to make pending data available for reading")
-		Storage.DebugFlush()
+		vms.s.DebugFlush()
 		return true
 	}
 
@@ -389,7 +263,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 		}
 		logger.Infof("enabling logging of new series for the next %s. This may increase resource usage during this period.", time.Duration(dealine)*time.Second)
 		endTime := fasttime.UnixTimestamp() + uint64(dealine)
-		Storage.SetLogNewSeriesUntil(endTime)
+		vms.s.SetLogNewSeriesUntil(endTime)
 		fmt.Fprintf(w, `{"status":"success","data":{"logEndTime":%q}}`, time.Unix(int64(endTime), 0))
 		return true
 	}
@@ -411,13 +285,13 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	case "/create":
 		snapshotsCreateTotal.Inc()
 		w.Header().Set("Content-Type", "application/json")
-		snapshotName := Storage.MustCreateSnapshot()
+		snapshotName := vms.s.MustCreateSnapshot()
 
 		// Verify whether the client already closed the connection.
 		// In this case it is better to drop the created snapshot, since the client isn't interested in it.
 		if err := r.Context().Err(); err != nil {
 			logger.Infof("deleting already created snapshot at %s because the client canceled the request", snapshotName)
-			if err := deleteSnapshot(snapshotName); err != nil {
+			if err := vms.deleteSnapshot(snapshotName); err != nil {
 				logger.Infof("cannot delete just created snapshot: %s", err)
 				return true
 			}
@@ -433,7 +307,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	case "/list":
 		snapshotsListTotal.Inc()
 		w.Header().Set("Content-Type", "application/json")
-		snapshots := Storage.MustListSnapshots()
+		snapshots := vms.s.MustListSnapshots()
 		fmt.Fprintf(w, `{"status":"ok","snapshots":[`)
 		if len(snapshots) > 0 {
 			for _, snapshot := range snapshots[:len(snapshots)-1] {
@@ -447,7 +321,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 		snapshotsDeleteTotal.Inc()
 		w.Header().Set("Content-Type", "application/json")
 		snapshotName := r.FormValue("snapshot")
-		if err := deleteSnapshot(snapshotName); err != nil {
+		if err := vms.deleteSnapshot(snapshotName); err != nil {
 			jsonResponseError(w, err)
 			snapshotsDeleteErrorsTotal.Inc()
 			return true
@@ -457,9 +331,9 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	case "/delete_all":
 		snapshotsDeleteAllTotal.Inc()
 		w.Header().Set("Content-Type", "application/json")
-		snapshots := Storage.MustListSnapshots()
+		snapshots := vms.s.MustListSnapshots()
 		for _, snapshotName := range snapshots {
-			if err := Storage.DeleteSnapshot(snapshotName); err != nil {
+			if err := vms.s.DeleteSnapshot(snapshotName); err != nil {
 				err = fmt.Errorf("cannot delete snapshot %q: %w", snapshotName, err)
 				jsonResponseError(w, err)
 				snapshotsDeleteAllErrorsTotal.Inc()
@@ -473,50 +347,6 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	}
 }
 
-func deleteSnapshot(snapshotName string) error {
-	snapshots := Storage.MustListSnapshots()
-	for _, snName := range snapshots {
-		if snName == snapshotName {
-			if err := Storage.DeleteSnapshot(snName); err != nil {
-				return fmt.Errorf("cannot delete snapshot %q: %w", snName, err)
-			}
-			return nil
-		}
-	}
-	return fmt.Errorf("cannot find snapshot %q", snapshotName)
-}
-
-func initStaleSnapshotsRemover(strg *storage.Storage) {
-	staleSnapshotsRemoverCh = make(chan struct{})
-	if snapshotsMaxAge.Duration() <= 0 {
-		return
-	}
-	snapshotsMaxAgeDur := snapshotsMaxAge.Duration()
-	staleSnapshotsRemoverWG.Go(func() {
-		d := timeutil.AddJitterToDuration(time.Second * 11)
-		t := time.NewTicker(d)
-		defer t.Stop()
-		for {
-			select {
-			case <-staleSnapshotsRemoverCh:
-				return
-			case <-t.C:
-			}
-			strg.MustDeleteStaleSnapshots(snapshotsMaxAgeDur)
-		}
-	})
-}
-
-func stopStaleSnapshotsRemover() {
-	close(staleSnapshotsRemoverCh)
-	staleSnapshotsRemoverWG.Wait()
-}
-
-var (
-	staleSnapshotsRemoverCh chan struct{}
-	staleSnapshotsRemoverWG sync.WaitGroup
-)
-
 var (
 	activeForceMerges = metrics.NewCounter("vm_active_force_merges")
 
@@ -531,7 +361,16 @@ var (
 	snapshotsDeleteAllErrorsTotal = metrics.NewCounter(`vm_http_request_errors_total{path="/snapshot/delete_all"}`)
 )
 
-func writeStorageMetrics(w io.Writer, strg *storage.Storage) {
+// TODO(@rtm0): Move to metrics.go.
+func (vmssn *VMStorageSingleNode) writeStorageMetrics(w io.Writer) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	vmssn.vms.writeStorageMetrics(w)
+}
+
+// TODO(@rtm0): Move to metrics.go.
+func (vms *VMStorage) writeStorageMetrics(w io.Writer) {
+	strg := vms.s
 	var m storage.Metrics
 	strg.UpdateMetrics(&m)
 	tm := &m.TableMetrics
@@ -755,6 +594,8 @@ func writeStorageMetrics(w io.Writer, strg *storage.Storage) {
 	metrics.WriteGaugeUint64(w, `vm_downsampling_partitions_scheduled`, tm.ScheduledDownsamplingPartitions)
 	metrics.WriteGaugeUint64(w, `vm_downsampling_partitions_scheduled_size_bytes`, tm.ScheduledDownsamplingPartitionsSize)
 
+	metrics.WriteGaugeUint64(w, `vm_search_max_unique_timeseries`, uint64(vms.maxUniqueTimeSeriesCalculated))
+
 	metrics.WriteGaugeUint64(w, `vm_metrics_metadata_storage_items`, m.MetadataStorageItemsCurrent)
 	metrics.WriteCounterUint64(w, `vm_metrics_metadata_storage_size_bytes`, m.MetadataStorageCurrentSizeBytes)
 	metrics.WriteCounterUint64(w, `vm_metrics_metadata_storage_max_size_bytes`, m.MetadataStorageMaxSizeBytes)

app/vmstorage/vmstorage.go

@@ -0,0 +1,391 @@
+package vmstorage
+
+import (
+	"flag"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/memory"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/querytracer"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricnamestats"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricsmetadata"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/timeutil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/vmselectapi"
+)
+
+var (
+	precisionBits = flag.Int("precisionBits", 64, "The number of precision bits to store per each value. Lower precision bits improves data compression "+
+		"at the cost of precision loss")
+	maxUniqueTimeseries = flag.Int("search.maxUniqueTimeseries", 0, "The maximum number of unique time series, which can be scanned during every query. "+
+		"This allows protecting against heavy queries, which select unexpectedly high number of series. When set to zero, the limit is automatically calculated based on -search.maxConcurrentRequests (inversely proportional) and memory available to the process (proportional). See also -search.max* command-line flags at vmselect")
+	maxTagKeys = flag.Int("search.maxTagKeys", 100e3, "The maximum number of tag keys returned per search. "+
+		"See also -search.maxLabelsAPISeries and -search.maxLabelsAPIDuration")
+	maxTagValues = flag.Int("search.maxTagValues", 100e3, "The maximum number of tag values returned per search. "+
+		"See also -search.maxLabelsAPISeries and -search.maxLabelsAPIDuration")
+	maxTagValueSuffixesPerSearch = flag.Int("search.maxTagValueSuffixesPerSearch", 100e3, "The maximum number of tag value suffixes returned from /metrics/find")
+	snapshotsMaxAge              = flagutil.NewRetentionDuration("snapshotsMaxAge", "3d", "Automatically delete snapshots older than -snapshotsMaxAge if it is set to non-zero duration. Make sure that backup process has enough time to finish the backup before the corresponding snapshot is automatically deleted")
+)
+
+// newVMStorage creates a new instance of of VMStorage.
+//
+// The created VMStorage instance takes ownership of s.
+func newVMStorage(s *storage.Storage, vmselectMaxConcurrentRequests int) *VMStorage {
+	if err := encoding.CheckPrecisionBits(uint8(*precisionBits)); err != nil {
+		logger.Fatalf("invalid -precisionBits=%d: %s", *precisionBits, err)
+	}
+
+	maxUniqueTimeseriesCalculated := *maxUniqueTimeseries
+	if maxUniqueTimeseriesCalculated <= 0 {
+		maxUniqueTimeseriesCalculated = calculateMaxUniqueTimeseries(vmselectMaxConcurrentRequests, memory.Remaining())
+	}
+
+	vms := &VMStorage{
+		s:                             s,
+		maxUniqueTimeseries:           *maxUniqueTimeseries,
+		maxUniqueTimeSeriesCalculated: maxUniqueTimeseriesCalculated,
+		staleSnapshotsRemoverCh:       make(chan struct{}),
+	}
+	vms.initStaleSnapshotsRemover()
+	return vms
+}
+
+// calculateMaxUniqueTimeseries calculates the maxUniqueTimeseries based on the
+// available system resources.
+func calculateMaxUniqueTimeseries(maxConcurrentRequests, remainingMemory int) int {
+	if maxConcurrentRequests <= 0 {
+		// This line should NOT be reached unless the user has set an incorrect `search.maxConcurrentRequests`.
+		// In such cases, fallback to unlimited.
+		logger.Warnf("limiting -search.maxUniqueTimeseries to %v because -search.maxConcurrentRequests=%d.", 2e9, maxConcurrentRequests)
+		return 2e9
+	}
+
+	// Calculate the max metrics limit for a single request in the worst-case concurrent scenario.
+	// The approximate size of 1 unique series that could occupy in the vmstorage is 200 bytes.
+	mts := remainingMemory / 200 / maxConcurrentRequests
+	logger.Infof("limiting -search.maxUniqueTimeseries to %d according to -search.maxConcurrentRequests=%d and remaining memory=%d bytes. To increase the limit, reduce -search.maxConcurrentRequests or increase memory available to the process.", mts, maxConcurrentRequests, remainingMemory)
+	return mts
+}
+
+// VMStorage impelements vmselectapi.API and vminsertapi.API.
+type VMStorage struct {
+	s                             *storage.Storage
+	maxUniqueTimeseries           int
+	maxUniqueTimeSeriesCalculated int
+	staleSnapshotsRemoverCh       chan struct{}
+	staleSnapshotsRemoverWG       sync.WaitGroup
+}
+
+func (vms *VMStorage) initStaleSnapshotsRemover() {
+	if snapshotsMaxAge.Duration() <= 0 {
+		return
+	}
+	snapshotsMaxAgeDuration := snapshotsMaxAge.Duration()
+	vms.staleSnapshotsRemoverWG.Go(func() {
+		d := timeutil.AddJitterToDuration(time.Second * 11)
+		t := time.NewTicker(d)
+		defer t.Stop()
+		for {
+			select {
+			case <-vms.staleSnapshotsRemoverCh:
+				return
+			case <-t.C:
+			}
+			vms.s.MustDeleteStaleSnapshots(snapshotsMaxAgeDuration)
+		}
+	})
+}
+
+func (vms *VMStorage) Stop() {
+	close(vms.staleSnapshotsRemoverCh)
+	vms.staleSnapshotsRemoverWG.Wait()
+	vms.s.MustClose()
+}
+
+// WriteRows writes metric rows to the storage.
+//
+// The caller should limit the number of concurrent calls to WriteRows() in
+// order to limit memory usage.
+func (vms *VMStorage) WriteRows(rows []storage.MetricRow) error {
+	vms.s.AddRows(rows, uint8(*precisionBits))
+	return nil
+}
+
+// WriteMetadata writes metrics metadata to storage.
+//
+// The caller should limit the number of concurrent calls to WriteMetadata() in
+// order to limit memory usage.
+func (vms *VMStorage) WriteMetadata(rows []metricsmetadata.Row) error {
+	vms.s.AddMetadataRows(rows)
+	return nil
+}
+
+// IsReadOnly returns true is the storage is in read-only mode.
+func (vms *VMStorage) IsReadOnly() bool {
+	return vms.s.IsReadOnly()
+}
+
+func (vms *VMStorage) InitSearch(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (vmselectapi.BlockIterator, error) {
+	tr := sq.GetTimeRange()
+	maxMetrics := vms.getMaxMetrics(sq.MaxMetrics)
+	tfss, err := vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	if len(tfss) == 0 {
+		return nil, fmt.Errorf("missing tag filters")
+	}
+	bi := getBlockIterator()
+	bi.sr.Init(qt, vms.s, tfss, tr, maxMetrics, deadline)
+	if err := bi.sr.Error(); err != nil {
+		bi.MustClose()
+		return nil, err
+	}
+	return bi, nil
+}
+
+func (vms *VMStorage) getMaxMetrics(searchQueryLimit int) int {
+	if searchQueryLimit <= 0 {
+		return vms.maxUniqueTimeSeriesCalculated
+	}
+	// searchQueryLimit cannot exceed `-search.maxUniqueTimeseries`
+	if vms.maxUniqueTimeseries != 0 && searchQueryLimit > vms.maxUniqueTimeseries {
+		searchQueryLimit = vms.maxUniqueTimeseries
+	}
+	return searchQueryLimit
+}
+
+// blockIterator implements vmselectapi.BlockIterator
+type blockIterator struct {
+	sr storage.Search
+	mb storage.MetricBlock
+}
+
+var blockIteratorsPool sync.Pool
+
+func (bi *blockIterator) MustClose() {
+	bi.sr.MustClose()
+	bi.mb.MetricName = nil
+	bi.mb.Block.Reset()
+	blockIteratorsPool.Put(bi)
+}
+
+func getBlockIterator() *blockIterator {
+	v := blockIteratorsPool.Get()
+	if v == nil {
+		v = &blockIterator{}
+	}
+	return v.(*blockIterator)
+}
+
+func (bi *blockIterator) NextBlock(dst []byte) ([]byte, bool) {
+	if !bi.sr.NextMetricBlock() {
+		return dst, false
+	}
+	mb := bi.mb
+	mb.MetricName = bi.sr.MetricBlockRef.MetricName
+	bi.sr.MetricBlockRef.BlockRef.MustReadBlock(&mb.Block)
+	dst = mb.Marshal(dst[:0])
+	return dst, true
+}
+
+func (bi *blockIterator) Error() error {
+	return bi.sr.Error()
+}
+
+// SearchMetricNames returns metric names for the given tfss on the given tr.
+func (vms *VMStorage) SearchMetricNames(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) ([]string, error) {
+	tr := sq.GetTimeRange()
+	maxMetrics := sq.MaxMetrics
+	if maxMetrics <= 0 {
+		// fallback to maxUniqueTimeSeries if no limit is provided,
+		// see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/7857
+		maxMetrics = vms.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	if len(tfss) == 0 {
+		return nil, fmt.Errorf("missing tag filters")
+	}
+	return vms.s.SearchMetricNames(qt, tfss, tr, maxMetrics, deadline)
+}
+
+// SearchLabelValues searches for label values for the given labelName, tfss and
+// tr.
+func (vms *VMStorage) LabelValues(qt *querytracer.Tracer, sq *storage.SearchQuery, labelName string, maxLabelValues int, deadline uint64) ([]string, error) {
+	tr := sq.GetTimeRange()
+	if maxLabelValues <= 0 || maxLabelValues > *maxTagValues {
+		maxLabelValues = *maxTagValues
+	}
+	maxMetrics := sq.MaxMetrics
+	if maxMetrics <= 0 {
+		// fallback to maxUniqueTimeSeries if no limit is provided,
+		// see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/7857
+		maxMetrics = vms.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	return vms.s.SearchLabelValues(qt, labelName, tfss, tr, maxLabelValues, maxMetrics, deadline)
+}
+
+// TagValueSuffixes returns all the tag value suffixes for the given tagKey and
+// tagValuePrefix on the given tr.
+//
+// This allows implementing
+// https://graphite-api.readthedocs.io/en/latest/api.html#metrics-find or
+// similar APIs.
+func (vms *VMStorage) TagValueSuffixes(qt *querytracer.Tracer, _, _ uint32, tr storage.TimeRange, tagKey, tagValuePrefix string, delimiter byte,
+	maxSuffixes int, deadline uint64) ([]string, error) {
+	if maxSuffixes <= 0 || maxSuffixes > *maxTagValueSuffixesPerSearch {
+		maxSuffixes = *maxTagValueSuffixesPerSearch
+	}
+	suffixes, err := vms.s.SearchTagValueSuffixes(qt, tr, tagKey, tagValuePrefix, delimiter, maxSuffixes, deadline)
+	if err != nil {
+		return nil, err
+	}
+	if len(suffixes) >= maxSuffixes {
+		return nil, fmt.Errorf("more than -search.maxTagValueSuffixesPerSearch=%d suffixes returned; "+
+			"either narrow down the search or increase -search.maxTagValueSuffixesPerSearch command-line flag value", maxSuffixes)
+	}
+	return suffixes, nil
+}
+
+// SearchLabelNames searches for tag keys matching the given tfss on tr.
+func (vms *VMStorage) LabelNames(qt *querytracer.Tracer, sq *storage.SearchQuery, maxLabelNames int, deadline uint64) ([]string, error) {
+	tr := sq.GetTimeRange()
+	if maxLabelNames <= 0 || maxLabelNames > *maxTagKeys {
+		maxLabelNames = *maxTagKeys
+	}
+	maxMetrics := sq.MaxMetrics
+	if maxMetrics <= 0 {
+		// fallback to maxUniqueTimeSeries if no limit is provided,
+		// see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/7857
+		maxMetrics = vms.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	return vms.s.SearchLabelNames(qt, tfss, tr, maxLabelNames, maxMetrics, deadline)
+}
+
+func (vms *VMStorage) SeriesCount(_ *querytracer.Tracer, _, _ uint32, deadline uint64) (uint64, error) {
+	return vms.s.GetSeriesCount(deadline)
+}
+
+func (vms *VMStorage) Tenants(_ *querytracer.Tracer, _ storage.TimeRange, _ uint64) ([]string, error) {
+	return nil, nil
+}
+
+// GetTSDBStatus returns TSDB status for given filters on the given date.
+func (vms *VMStorage) TSDBStatus(qt *querytracer.Tracer, sq *storage.SearchQuery, focusLabel string, topN int, deadline uint64) (*storage.TSDBStatus, error) {
+	tr := sq.GetTimeRange()
+	maxMetrics := sq.MaxMetrics
+	if maxMetrics <= 0 {
+		// fallback to maxUniqueTimeSeries if no limit is provided,
+		// see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/7857
+		maxMetrics = vms.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	date := uint64(sq.MinTimestamp) / (24 * 3600 * 1000)
+	return vms.s.GetTSDBStatus(qt, tfss, date, focusLabel, topN, maxMetrics, deadline)
+}
+
+// DeleteSeries deletes series matching tfss.
+//
+// Returns the number of deleted series.
+func (vms *VMStorage) DeleteSeries(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (int, error) {
+	tr := sq.GetTimeRange()
+	maxMetrics := sq.MaxMetrics
+	if maxMetrics <= 0 {
+		// fallback to maxUniqueTimeSeries if no limit is provided,
+		// see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/7857
+		maxMetrics = vms.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return 0, err
+	}
+	if len(tfss) == 0 {
+		return 0, fmt.Errorf("missing tag filters")
+	}
+	return vms.s.DeleteSeries(qt, tfss, maxMetrics)
+}
+
+func (vms *VMStorage) RegisterMetricNames(qt *querytracer.Tracer, mrs []storage.MetricRow, _ uint64) error {
+	vms.s.RegisterMetricNames(qt, mrs)
+	return nil
+}
+
+// GetMetricNamesUsageStats returns metric name usage stats.
+func (vms *VMStorage) GetMetricNamesUsageStats(qt *querytracer.Tracer, _ *storage.TenantToken, limit, le int, matchPattern string, _ uint64) (metricnamestats.StatsResult, error) {
+	return vms.s.GetMetricNamesStats(qt, limit, le, matchPattern), nil
+}
+
+// ResetMetricNamesStats resets state for metric names usage tracker
+func (vms *VMStorage) ResetMetricNamesUsageStats(qt *querytracer.Tracer, _ uint64) error {
+	vms.s.ResetMetricNamesStats(qt)
+	return nil
+}
+
+func (vms *VMStorage) setupTfss(qt *querytracer.Tracer, sq *storage.SearchQuery, tr storage.TimeRange, maxMetrics int, deadline uint64) ([]*storage.TagFilters, error) {
+	tfss := make([]*storage.TagFilters, 0, len(sq.TagFilterss))
+	for _, tagFilters := range sq.TagFilterss {
+		tfs := storage.NewTagFilters()
+		for i := range tagFilters {
+			tf := &tagFilters[i]
+			if string(tf.Key) == "__graphite__" {
+				query := tf.Value
+				qtChild := qt.NewChild("searching for series matching __graphite__=%q", query)
+				paths, err := vms.s.SearchGraphitePaths(qtChild, tr, query, maxMetrics, deadline)
+				qtChild.Donef("found %d series", len(paths))
+				if err != nil {
+					return nil, fmt.Errorf("error when searching for Graphite paths for query %q: %w", query, err)
+				}
+				if len(paths) >= maxMetrics {
+					return nil, fmt.Errorf("more than %d time series match Graphite query %q; "+
+						"either narrow down the query or increase the corresponding -search.max* command-line flag value at vmselect nodes; "+
+						"see https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#resource-usage-limits", maxMetrics, query)
+				}
+				tfs.AddGraphiteQuery(query, paths, tf.IsNegative)
+				continue
+			}
+			if err := tfs.Add(tf.Key, tf.Value, tf.IsNegative, tf.IsRegexp); err != nil {
+				return nil, fmt.Errorf("cannot parse tag filter %s: %w", tf, err)
+			}
+		}
+		tfss = append(tfss, tfs)
+	}
+	return tfss, nil
+}
+
+func (vms *VMStorage) GetMetadataRecords(qt *querytracer.Tracer, _ *storage.TenantToken, limit int, metricName string, _ uint64) ([]*metricsmetadata.Row, error) {
+	return vms.s.GetMetadataRows(qt, limit, metricName), nil
+}
+
+// deleteSnapshot deletes a snapshot by its name.
+//
+// Callers must wrap the call with wg.Add(1)...wg.Done().
+func (vms *VMStorage) deleteSnapshot(snapshotName string) error {
+	snapshots := vms.s.MustListSnapshots()
+	for _, snName := range snapshots {
+		if snName == snapshotName {
+			if err := vms.s.DeleteSnapshot(snName); err != nil {
+				return fmt.Errorf("cannot delete snapshot %q: %w", snName, err)
+			}
+			return nil
+		}
+	}
+	return fmt.Errorf("cannot find snapshot %q", snapshotName)
+}

app/vmstorage/vmstorage_single_node.go

@@ -0,0 +1,213 @@
+package vmstorage
+
+import (
+	"errors"
+	"fmt"
+	"sync"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/querytracer"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricnamestats"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricsmetadata"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/syncwg"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/vmselectapi"
+)
+
+// newVMStorageSingleNode creates a new instance of of VMStorage for vmsingle.
+func newVMStorageSingleNode(s *storage.Storage, maxConcurrentRequests int, resetCacheIfNeeded func(mrs []storage.MetricRow)) *VMStorageSingleNode {
+	vms := newVMStorage(s, maxConcurrentRequests)
+	return &VMStorageSingleNode{
+		vms:                vms,
+		wg:                 syncwg.WaitGroup{},
+		resetCacheIfNeeded: resetCacheIfNeeded,
+	}
+}
+
+type VMStorageSingleNode struct {
+	vms *VMStorage
+
+	// wg is used to wrap every storage call into wg.Add(1) ... wg.Done()
+	// for proper graceful shutdown when Stop is called.
+	//
+	// Use syncwg instead of sync, since Add is called from concurrent
+	// goroutines.
+	wg syncwg.WaitGroup
+
+	// resetCacheIfNeeded is a callback for automatic resetting of response
+	// cache if needed.
+	resetCacheIfNeeded func(mrs []storage.MetricRow)
+}
+
+func (vmssn *VMStorageSingleNode) Stop() {
+	vmssn.wg.WaitAndBlock()
+	vmssn.vms.Stop()
+}
+
+// WriteRows writes metric rows to the storage.
+//
+// Returns an error if the storage is in read-only mode.
+//
+// The caller should limit the number of concurrent calls to WriteRows() in
+// order to limit memory usage.
+func (vmssn *VMStorageSingleNode) WriteRows(rows []storage.MetricRow) error {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+
+	if vmssn.vms.IsReadOnly() {
+		return errReadOnly
+	}
+	vmssn.resetCacheIfNeeded(rows)
+	return vmssn.vms.WriteRows(rows)
+}
+
+// WriteMetadata writes metrics metadata to storage.
+//
+// Returns an error if the storage is in read-only mode.
+//
+// The caller should limit the number of concurrent calls to WriteMetadata() in
+// order to limit memory usage.
+func (vmssn *VMStorageSingleNode) WriteMetadata(rows []metricsmetadata.Row) error {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+
+	if vmssn.vms.IsReadOnly() {
+		return errReadOnly
+	}
+	return vmssn.vms.WriteMetadata(rows)
+}
+
+var errReadOnly = errors.New("the storage is in read-only mode; check -storage.minFreeDiskSpaceBytes command-line flag value")
+
+func (vmssn *VMStorageSingleNode) IsReadOnly() bool {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.IsReadOnly()
+}
+
+func (vmssn *VMStorageSingleNode) InitSearch(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (vmselectapi.BlockIterator, error) {
+	return nil, fmt.Errorf("not implemented in vmsingle")
+}
+
+// GetSearch sets up an instance of storage search and returns it to the caller
+// along with the max series count that the search can return.
+//
+// This method is not part of the vmselectapi.API and must only be used by
+// vmsingle HTTP handlers.
+//
+// Callers of this method must call PutSearch() once the search instance is not
+// needed anymore.
+func (vmssn *VMStorageSingleNode) GetSearch(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (*storage.Search, int, error) {
+	vmssn.wg.Add(1)
+
+	tr := sq.GetTimeRange()
+	maxMetrics := vmssn.vms.getMaxMetrics(sq.MaxMetrics)
+	tfss, err := vmssn.vms.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		vmssn.wg.Done()
+		return nil, 0, err
+	}
+
+	sr := getSearch()
+	maxSeriesCount := sr.Init(qt, vmssn.vms.s, tfss, tr, sq.MaxMetrics, deadline)
+	return sr, maxSeriesCount, nil
+}
+
+// PutSearch resets the search once it is not needed anymore and puts it aside
+// for future reuse.
+//
+// This method is not part of the vmselectapi.API and must only be used by
+// vmsingle HTTP handlers.
+//
+// The method must only be used on search instances that have been created with
+// GetSearch().
+func (vmssn *VMStorageSingleNode) PutSearch(sr *storage.Search) {
+	putSearch(sr)
+	vmssn.wg.Done()
+}
+
+func getSearch() *storage.Search {
+	v := ssPool.Get()
+	if v == nil {
+		return &storage.Search{}
+	}
+	return v.(*storage.Search)
+}
+
+func putSearch(sr *storage.Search) {
+	sr.MustClose()
+	ssPool.Put(sr)
+}
+
+var ssPool sync.Pool
+
+func (vmssn *VMStorageSingleNode) SearchMetricNames(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) ([]string, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.SearchMetricNames(qt, sq, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) LabelValues(qt *querytracer.Tracer, sq *storage.SearchQuery, labelName string, maxLabelValues int, deadline uint64) ([]string, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.LabelValues(qt, sq, labelName, maxLabelValues, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) TagValueSuffixes(qt *querytracer.Tracer, accountID, projectID uint32, tr storage.TimeRange, tagKey, tagValuePrefix string, delimiter byte, maxSuffixes int, deadline uint64) ([]string, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.TagValueSuffixes(qt, accountID, projectID, tr, tagKey, tagValuePrefix, delimiter, maxSuffixes, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) LabelNames(qt *querytracer.Tracer, sq *storage.SearchQuery, maxLabelNames int, deadline uint64) ([]string, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.LabelNames(qt, sq, maxLabelNames, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) SeriesCount(qt *querytracer.Tracer, accountID, projectID uint32, deadline uint64) (uint64, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.SeriesCount(qt, accountID, projectID, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) Tenants(qt *querytracer.Tracer, tr storage.TimeRange, deadline uint64) ([]string, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.Tenants(qt, tr, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) TSDBStatus(qt *querytracer.Tracer, sq *storage.SearchQuery, focusLabel string, topN int, deadline uint64) (*storage.TSDBStatus, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.TSDBStatus(qt, sq, focusLabel, topN, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) DeleteSeries(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (int, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.DeleteSeries(qt, sq, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) RegisterMetricNames(qt *querytracer.Tracer, mrs []storage.MetricRow, deadline uint64) error {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.RegisterMetricNames(qt, mrs, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) GetMetricNamesUsageStats(qt *querytracer.Tracer, tt *storage.TenantToken, limit, le int, matchPattern string, deadline uint64) (metricnamestats.StatsResult, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.GetMetricNamesUsageStats(qt, tt, limit, le, matchPattern, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) ResetMetricNamesUsageStats(qt *querytracer.Tracer, deadline uint64) error {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.ResetMetricNamesUsageStats(qt, deadline)
+}
+
+func (vmssn *VMStorageSingleNode) GetMetadataRecords(qt *querytracer.Tracer, tt *storage.TenantToken, limit int, metricName string, deadline uint64) ([]*metricsmetadata.Row, error) {
+	vmssn.wg.Add(1)
+	defer vmssn.wg.Done()
+	return vmssn.vms.GetMetadataRecords(qt, tt, limit, metricName, deadline)
+}

app/vmstorage/vmstorage_test.go

@@ -0,0 +1,62 @@
+package vmstorage
+
+import (
+	"math"
+	"strconv"
+	"testing"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/cgroup"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
+)
+
+func TestCalculateMaxMetricsLimitByResource(t *testing.T) {
+	f := func(maxConcurrentRequest, remainingMemory, expect int) {
+		t.Helper()
+		maxMetricsLimit := calculateMaxUniqueTimeseries(maxConcurrentRequest, remainingMemory)
+		if maxMetricsLimit != expect {
+			t.Fatalf("unexpected max metrics limit: got %d, want %d", maxMetricsLimit, expect)
+		}
+	}
+
+	// 64-bit architectures support memory sizes > 4GB.
+	if strconv.IntSize == 64 {
+		// 8 CPU & 32 GiB
+		f(16, int(math.Round(32*1024*1024*1024*0.4)), 4294967)
+		// 4 CPU & 32 GiB
+		f(8, int(math.Round(32*1024*1024*1024*0.4)), 8589934)
+	}
+
+	// 2 CPU & 4 GiB
+	f(4, int(math.Round(4*1024*1024*1024*0.4)), 2147483)
+
+	// other edge cases
+	f(0, int(math.Round(4*1024*1024*1024*0.4)), 2e9)
+	f(4, 0, 0)
+
+}
+
+func TestGetMaxMetrics(t *testing.T) {
+	originalMaxUniqueTimeSeries := *maxUniqueTimeseries
+	defer func() {
+		*maxUniqueTimeseries = originalMaxUniqueTimeSeries
+		fs.MustRemoveDir(t.Name())
+	}()
+
+	maxConcurrentRequests := 2 * cgroup.AvailableCPUs()
+	f := func(searchQueryLimit, storageMaxUniqueTimeseries, expect int) {
+		t.Helper()
+		*maxUniqueTimeseries = storageMaxUniqueTimeseries
+		s := storage.MustOpenStorage(t.Name(), storage.OpenOptions{})
+		vms := newVMStorage(s, maxConcurrentRequests)
+		defer vms.Stop()
+		maxMetrics := vms.getMaxMetrics(searchQueryLimit)
+		if maxMetrics != expect {
+			t.Fatalf("unexpected max metrics: got %d, want %d", maxMetrics, expect)
+		}
+	}
+
+	f(0, 1e6, 1e6)
+	f(2e6, 0, 2e6)
+	f(2e6, 1e6, 1e6)
+}

app/vmui/Dockerfile-web

@@ -1,4 +1,4 @@
-FROM golang:1.26.3 AS build-web-stage
+FROM golang:1.26.4 AS build-web-stage
 COPY build /build
 
 WORKDIR /build

app/vmui/packages/vmui/src/components/Configurators/QueryEditor/warningText.ts

@@ -1,7 +1,7 @@
-export const seriesFetchedWarning = `No match! 
+export const seriesFetchedWarning = `No match!
 This query hasn't selected any time series from database.
 Either the requested metrics are missing in the database,
 or there is a typo in series selector.`;
 
 export const partialWarning = `The shown results are marked as PARTIAL.
-The result is marked as partial if one or more vmstorage nodes failed to respond to the query.`;
+The result is marked as partial if one or more storage nodes failed to respond to the query.`;

app/vmui/packages/vmui/src/components/ExploreAlerts/RulesHeader/index.tsx

@@ -71,7 +71,7 @@ const RulesHeader = ({
           <TextField
             label="Search"
             value={search}
-            placeholder="Filter by rule, name or labels"
+            placeholder="Filter by group or rule name"
             startIcon={<SearchIcon />}
             onChange={onChangeSearch}
           />

app/vmui/packages/vmui/src/components/Main/Tabs/TabItem.tsx

@@ -1,6 +1,5 @@
 import { Component, FC, Ref } from "preact/compat";
 import classNames from "classnames";
-import { getCssVariable } from "../../../utils/theme";
 import { TabItemType } from "./Tabs";
 import TabItemWrapper from "./TabItemWrapper";
 import "./style.scss";
@@ -8,7 +7,6 @@ import "./style.scss";
 interface TabItemProps {
   activeItem: string
   item: TabItemType
-  color?: string
   onChange?: (value: string) => void
   activeNavRef: Ref<Component>
   isNavLink?: boolean
@@ -17,7 +15,6 @@ interface TabItemProps {
 const TabItem: FC<TabItemProps> = ({
   activeItem,
   item,
-  color = getCssVariable("color-primary"),
   activeNavRef,
   onChange,
   isNavLink
@@ -35,7 +32,6 @@ const TabItem: FC<TabItemProps> = ({
       })}
       isNavLink={isNavLink}
       to={item.value}
-      style={{ color: color }}
       onClick={createHandlerClickTab(item.value)}
       ref={activeItem === item.value ? activeNavRef : undefined}
     >

app/vmui/packages/vmui/src/components/Main/Tabs/TabItemWrapper.tsx

@@ -6,7 +6,6 @@ interface TabItemWrapperProps {
   to: string
   isNavLink?: boolean
   className: string
-  style: { color: string }
   children: ReactNode
   onClick: () => void
 }

app/vmui/packages/vmui/src/components/Main/Tabs/Tabs.tsx

@@ -1,6 +1,5 @@
 import { Component, FC, useRef, useState } from "preact/compat";
 import { ReactNode, useEffect } from "react";
-import { getCssVariable } from "../../../utils/theme";
 import TabItem from "./TabItem";
 import "./style.scss";
 import useWindowSize from "../../../hooks/useWindowSize";
@@ -15,7 +14,6 @@ export interface TabItemType {
 interface TabsProps {
   activeItem: string
   items: TabItemType[]
-  color?: string
   onChange?: (value: string) => void
   indicatorPlacement?: "bottom" | "top"
   isNavLink?: boolean
@@ -24,7 +22,6 @@ interface TabsProps {
 const Tabs: FC<TabsProps> = ({
   activeItem,
   items,
-  color = getCssVariable("color-primary"),
   onChange,
   indicatorPlacement = "bottom",
   isNavLink,
@@ -48,14 +45,13 @@ const Tabs: FC<TabsProps> = ({
         activeItem={activeItem}
         item={item}
         onChange={onChange}
-        color={color}
         activeNavRef={activeNavRef}
         isNavLink={isNavLink}
       />
     ))}
     <div
       className="vm-tabs__indicator"
-      style={{ ...indicatorPosition, borderColor: color }}
+      style={{ ...indicatorPosition }}
     />
   </div>;
 };

app/vmui/packages/vmui/src/components/Main/Tabs/style.scss

@@ -14,7 +14,7 @@
     align-items: center;
     justify-content: center;
     padding: $padding-global $padding-small;
-    color: inherit;
+    color: $color-primary;
     text-decoration: none;
     text-transform: capitalize;
     font-size: inherit;
@@ -46,5 +46,6 @@
     position: absolute;
     border-bottom: 2px solid;
     transition: width 200ms ease, left 300ms cubic-bezier(0.280, 0.840, 0.420, 1);
+    border-color: $color-primary;
   }
 }

app/vmui/packages/vmui/src/state/customPanel/reducer.test.ts

@@ -0,0 +1,35 @@
+import { afterEach, describe, expect, it, vi, type Mock } from "vitest";
+import { getFromStorage, saveToStorage } from "../../utils/storage";
+
+vi.mock("../../utils/storage", () => ({
+  getFromStorage: vi.fn(),
+  saveToStorage: vi.fn(),
+}));
+
+describe("customPanel reducer", () => {
+  afterEach(() => {
+    vi.resetAllMocks();
+    vi.resetModules();
+  });
+
+  it("persists reduceMemUsage under its own storage key", async () => {
+    const { reducer, initialCustomPanelState } = await import("./reducer");
+
+    reducer(initialCustomPanelState, { type: "TOGGLE_REDUCE_MEM_USAGE" });
+
+    expect(saveToStorage).toHaveBeenCalledWith("REDUCE_MEM_USAGE", true);
+    expect(saveToStorage).not.toHaveBeenCalledWith("TABLE_COMPACT", true);
+  });
+
+  it("hydrates reduceMemUsage from storage", async () => {
+    const getFromStorageMock = getFromStorage as Mock;
+    getFromStorageMock.mockImplementation((key: string) => {
+      if (key === "REDUCE_MEM_USAGE") return true;
+      return undefined;
+    });
+
+    const { initialCustomPanelState } = await import("./reducer");
+
+    expect(initialCustomPanelState.reduceMemUsage).toBe(true);
+  });
+});

app/vmui/packages/vmui/src/state/customPanel/reducer.ts

@@ -35,7 +35,7 @@ export const initialCustomPanelState: CustomPanelState = {
   isTracingEnabled: false,
   seriesLimits: limitsStorage ? JSON.parse(limitsStorage) : DEFAULT_MAX_SERIES,
   tableCompact: getFromStorage("TABLE_COMPACT") as boolean || false,
-  reduceMemUsage: false
+  reduceMemUsage: getFromStorage("REDUCE_MEM_USAGE") as boolean || false
 };
 
 export function reducer(state: CustomPanelState, action: CustomPanelAction): CustomPanelState {
@@ -69,7 +69,7 @@ export function reducer(state: CustomPanelState, action: CustomPanelAction): Cus
         tableCompact: !state.tableCompact
       };
     case "TOGGLE_REDUCE_MEM_USAGE":
-      saveToStorage("TABLE_COMPACT", !state.reduceMemUsage);
+      saveToStorage("REDUCE_MEM_USAGE", !state.reduceMemUsage);
       return {
         ...state,
         reduceMemUsage: !state.reduceMemUsage

app/vmui/packages/vmui/src/utils/storage.ts

@@ -7,6 +7,7 @@ export const ALL_STORAGE_KEYS = [
   "SERIES_LIMITS",
   "LEGEND_AUTO_COLLAPSE",
   "TABLE_COMPACT",
+  "REDUCE_MEM_USAGE",
   "TIMEZONE",
   "DISABLED_DEFAULT_TIMEZONE",
   "THEME",

apptest/model.go

@@ -79,24 +79,25 @@ type PrometheusWriteQuerier interface {
 
 // QueryOpts contains various params used for querying or ingesting data
 type QueryOpts struct {
-	Tenant         string
-	Timeout        string
-	Start          string
-	End            string
-	Time           string
-	Step           string
-	ExtraFilters   []string
-	ExtraLabels    []string
-	Trace          string
-	ReduceMemUsage string
-	MaxLookback    string
-	LatencyOffset  string
-	Format         string
-	NoCache        string
-	Headers        http.Header
-	From           string
-	Until          string
-	StorageStep    string
+	Tenant              string
+	Timeout             string
+	Start               string
+	End                 string
+	Time                string
+	Step                string
+	ExtraFilters        []string
+	ExtraLabels         []string
+	Trace               string
+	ReduceMemUsage      string
+	MaxLookback         string
+	LatencyOffset       string
+	Format              string
+	NoCache             string
+	Headers             http.Header
+	From                string
+	Until               string
+	StorageStep         string
+	DenyPartialResponse string
 }
 
 func (qos *QueryOpts) getHeaders() http.Header {
@@ -132,6 +133,7 @@ func (qos *QueryOpts) asURLValues() url.Values {
 	addNonEmpty("from", qos.From)
 	addNonEmpty("until", qos.Until)
 	addNonEmpty("storage_step", qos.StorageStep)
+	addNonEmpty("deny_partial_response", qos.DenyPartialResponse)
 
 	return uv
 }

apptest/tests/apptest_test.go

@@ -0,0 +1,40 @@
+package tests
+
+import (
+	"fmt"
+	"os"
+	"testing"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+func TestMain(m *testing.M) {
+	// check if necessary binaries are there.
+	checkBinaryRequirement("../../bin/victoria-metrics-race")
+	checkBinaryRequirement("../../bin/vmagent-race")
+	checkBinaryRequirement("../../bin/vmauth-race")
+	checkBinaryRequirement("../../bin/vmctl-race")
+	checkBinaryRequirement("../../bin/vmbackup-race")
+	checkBinaryRequirement("../../bin/vmrestore-race")
+
+	// check if the test is run via make command
+	checkIntegrationTestEnv()
+
+	// start the integration test.
+	os.Exit(m.Run())
+}
+
+// checkBinaryRequirement panic if required binary not exist.
+func checkBinaryRequirement(path string) {
+	if _, err := os.Stat(path); err != nil {
+		if os.IsNotExist(err) {
+			panic(fmt.Sprintf("apptest failed: %s not found. please run `make apptest` to execute apptest. check how different tests are executed: https://docs.victoriametrics.com/victoriametrics/contributing/#testing", path))
+		}
+	}
+}
+
+func checkIntegrationTestEnv() {
+	if os.Getenv("VM_APPTEST") == "" {
+		logger.Warnf("executing apptest with potential outdated binaries. it's recommended to execute via `make apptest` command. check this doc for more details: https://docs.victoriametrics.com/victoriametrics/contributing/#testing")
+	}
+}

apptest/tests/replication_test.go

@@ -1015,35 +1015,42 @@ func testGroupSkipSlowReplicas(tc *apptest.TestCase, opts *testGroupReplicationO
 func testGroupPartialResponse(tc *apptest.TestCase, opts *testGroupReplicationOpts) {
 	t := tc.T()
 
-	assertSeries := func(app *apptest.Vmselect, wantPartial bool) {
+	assertSeries := func(app *apptest.Vmselect, denyPartialResponse string, want *apptest.PrometheusAPIV1SeriesResponse) {
 		t.Helper()
 		tc.Assert(&apptest.AssertOptions{
 			Msg: "unexpected /api/v1/series response",
 			Got: func() any {
 				return app.PrometheusAPIV1Series(t, `{__name__=~".*"}`, apptest.QueryOpts{
-					Start: "2024-01-01T00:00:00Z",
-					End:   "2024-01-31T00:00:00Z",
+					Start:               "2024-01-01T00:00:00Z",
+					End:                 "2024-01-31T00:00:00Z",
+					DenyPartialResponse: denyPartialResponse,
 				}).Sort()
 			},
-			Want: &apptest.PrometheusAPIV1SeriesResponse{
-				Status:    "success",
-				IsPartial: wantPartial,
-			},
+			Want: want,
 			CmpOpts: []cmp.Option{
-				cmpopts.IgnoreFields(apptest.PrometheusAPIV1SeriesResponse{}, "Data"),
+				cmpopts.IgnoreFields(apptest.PrometheusAPIV1SeriesResponse{}, "Data", "Error"),
 			},
 		})
 	}
 
-	mustReturnPartialResponse := true
-	mustReturnFullResponse := false
+	allowPartialResponse := ""
+	denyPartialResponse := "1"
+
+	mustReturnPartialResponse := &apptest.PrometheusAPIV1SeriesResponse{
+		Status:    "success",
+		IsPartial: true,
+	}
+	mustReturnFullResponse := &apptest.PrometheusAPIV1SeriesResponse{
+		Status:    "success",
+		IsPartial: false,
+	}
 
 	// All vmstorage replicas are available so both vmselects must return full
 	// response.
-	assertSeries(opts.c.vmselect, mustReturnFullResponse)
-	assertSeries(opts.c.vmselectGroupRF, mustReturnFullResponse)
-	assertSeries(opts.c.vmselectGlobalRF, mustReturnFullResponse)
-	assertSeries(opts.c.vmselectGroupGlobalRF, mustReturnFullResponse)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnFullResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnFullResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnFullResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnFullResponse)
 
 	// Stop groupRF-1 vmstorage nodes in first group.
 	//
@@ -1053,10 +1060,10 @@ func testGroupPartialResponse(tc *apptest.TestCase, opts *testGroupReplicationOp
 	// about the replication factor and therefore they must still be able to
 	// return full dataset.
 	opts.c.storageGroups[0].stopNodes(tc, opts.groupRF-1)
-	assertSeries(opts.c.vmselect, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupRF, mustReturnFullResponse)
-	assertSeries(opts.c.vmselectGlobalRF, mustReturnFullResponse)
-	assertSeries(opts.c.vmselectGroupGlobalRF, mustReturnFullResponse)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnFullResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnFullResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnFullResponse)
 
 	// Stop groupRF-1 vmstorages in the remaining groups.
 	//
@@ -1066,10 +1073,10 @@ func testGroupPartialResponse(tc *apptest.TestCase, opts *testGroupReplicationOp
 	for g := 1; g < len(opts.c.storageGroups); g++ {
 		opts.c.storageGroups[g].stopNodes(tc, opts.groupRF-1)
 	}
-	assertSeries(opts.c.vmselect, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupRF, mustReturnFullResponse)
-	assertSeries(opts.c.vmselectGlobalRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupGlobalRF, mustReturnFullResponse)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnFullResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnFullResponse)
 
 	// Stop one more vmstorage in the first group.
 	//
@@ -1077,10 +1084,10 @@ func testGroupPartialResponse(tc *apptest.TestCase, opts *testGroupReplicationOp
 	// because it is unaware of replication across groups. vmselectGroupGlobalRF
 	// will continue retuning full dataset.
 	opts.c.storageGroups[0].stopNodes(tc, 1)
-	assertSeries(opts.c.vmselect, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGlobalRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupGlobalRF, mustReturnFullResponse)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnFullResponse)
 
 	// Stop one more vmstoarge in remaining globarRF-1 groups.
 	//
@@ -1089,19 +1096,56 @@ func testGroupPartialResponse(tc *apptest.TestCase, opts *testGroupReplicationOp
 	for g := 1; g < opts.globalRF-1; g++ {
 		opts.c.storageGroups[g].stopNodes(tc, 1)
 	}
-	assertSeries(opts.c.vmselect, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGlobalRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupGlobalRF, mustReturnFullResponse)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnFullResponse)
 
 	// Stop one more vmstoarge in one more group.
 	//
 	// vmselectGroupGlobalRF must now return partial dataset.
 	opts.c.storageGroups[opts.globalRF].stopNodes(tc, 1)
-	assertSeries(opts.c.vmselect, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGlobalRF, mustReturnPartialResponse)
-	assertSeries(opts.c.vmselectGroupGlobalRF, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+
+	// Stop all the remaining vmstorage nodes except a single node.
+	//
+	// At this point vmselects still must be able to return partial response
+	// because at least one vmstorage node has successfully returned results.
+	n := len(opts.c.storageGroups[0].vmstorages)
+	opts.c.storageGroups[0].stopNodes(tc, n-1)
+	for g := 1; g < len(opts.c.storageGroups); g++ {
+		n := len(opts.c.storageGroups[g].vmstorages)
+		opts.c.storageGroups[g].stopNodes(tc, n)
+	}
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnPartialResponse)
+
+	mustReturnUnavailableError := &apptest.PrometheusAPIV1SeriesResponse{
+		Status:    "error",
+		ErrorType: "503",
+	}
+
+	// vmselects must return an error for the same request when partial
+	// responses are denied explicitly.
+	assertSeries(opts.c.vmselect, denyPartialResponse, mustReturnUnavailableError)
+	assertSeries(opts.c.vmselectGroupRF, denyPartialResponse, mustReturnUnavailableError)
+	assertSeries(opts.c.vmselectGlobalRF, denyPartialResponse, mustReturnUnavailableError)
+	assertSeries(opts.c.vmselectGroupGlobalRF, denyPartialResponse, mustReturnUnavailableError)
+
+	// Stop the last remaining vmstorage node.
+	//
+	// vmselects must return an error when there are no successful vmstorage
+	// responses.
+	opts.c.storageGroups[0].stopNodes(tc, 1)
+	assertSeries(opts.c.vmselect, allowPartialResponse, mustReturnUnavailableError)
+	assertSeries(opts.c.vmselectGroupRF, allowPartialResponse, mustReturnUnavailableError)
+	assertSeries(opts.c.vmselectGlobalRF, allowPartialResponse, mustReturnUnavailableError)
+	assertSeries(opts.c.vmselectGroupGlobalRF, allowPartialResponse, mustReturnUnavailableError)
 }
 
 // TestClusterReplication_PartialResponseMultitenant checks how vmselect handles some

apptest/tests/testdata/mimir-tsdb/anonymous/01JFJBS3YP1SHZ3PJQ6HK76EC3/meta.json

@@ -0,0 +1,51 @@
+{
+	"ulid": "01JFJBS3YP1SHZ3PJQ6HK76EC3",
+	"minTime": 1734709200000,
+	"maxTime": 1734709320000,
+	"stats": {
+		"numSamples": 400,
+		"numSeries": 100,
+		"numChunks": 100
+	},
+	"compaction": {
+		"level": 1,
+		"sources": [
+			"01JFJBS3YP1SHZ3PJQ6HK76EC3"
+		],
+		"parents": [
+			{
+				"ulid": "00000000000000000000000000",
+				"minTime": 0,
+				"maxTime": 0
+			}
+		],
+		"hints": [
+			"from-out-of-order"
+		]
+	},
+	"version": 1,
+	"out_of_order": false,
+	"thanos": {
+		"labels": {},
+		"downsample": {
+			"resolution": 0
+		},
+		"source": "receive",
+		"segment_files": [
+			"000001"
+		],
+		"files": [
+			{
+				"rel_path": "chunks/000001",
+				"size_bytes": 4808
+			},
+			{
+				"rel_path": "index",
+				"size_bytes": 55021
+			},
+			{
+				"rel_path": "meta.json"
+			}
+		]
+	}
+}

apptest/tests/vmctl_mimir_migration_test.go

@@ -0,0 +1,139 @@
+package tests
+
+import (
+	"encoding/json"
+	"fmt"
+	"io"
+	"os"
+	"testing"
+
+	"github.com/google/go-cmp/cmp"
+	"github.com/google/go-cmp/cmp/cmpopts"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/apptest"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
+)
+
+const (
+	testMimirPath             = "testdata/mimir-tsdb"
+	expectedMimirResponseFile = "./testdata/mimir-tsdb/expected_response.json"
+)
+
+func TestSingleVmctlMimirProtocol(t *testing.T) {
+	fs.MustRemoveDir(t.Name())
+
+	tc := apptest.NewTestCase(t)
+	defer tc.Stop()
+
+	vmsingleDst := tc.MustStartDefaultVmsingle()
+	vmAddr := fmt.Sprintf("http://%s/", vmsingleDst.HTTPAddr())
+	dir, err := os.Getwd()
+	if err != nil {
+		t.Fatalf("cannot get current working directory: %s", err)
+	}
+
+	path := fmt.Sprintf("fs://%s/%s", dir, testMimirPath)
+	vmctlFlags := []string{
+		`mimir`,
+		`--mimir-tenant-id=anonymous`,
+		`--mimir-filter-time-start=2024-12-01T00:00:00Z`,
+		`--mimir-filter-time-end=2024-12-31T23:59:59Z`,
+		`--mimir-custom-s3-endpoint=http://localhost:9000`,
+		`--mimir-path=` + path,
+		`--vm-addr=` + vmAddr,
+		`--disable-progress-bar=true`,
+		`--vm-concurrency=6`,
+		`--mimir-concurrency=6`,
+	}
+
+	testMimirProtocol(tc, vmsingleDst, vmctlFlags)
+}
+
+func TestClusterVmctlMimirProtocol(t *testing.T) {
+	fs.MustRemoveDir(t.Name())
+
+	tc := apptest.NewTestCase(t)
+	defer tc.Stop()
+
+	cluster := tc.MustStartDefaultCluster()
+	vmAddr := fmt.Sprintf("http://%s/", cluster.Vminsert.HTTPAddr())
+	dir, err := os.Getwd()
+	if err != nil {
+		t.Fatalf("cannot get current working directory: %s", err)
+	}
+
+	path := fmt.Sprintf("fs://%s/%s", dir, testMimirPath)
+
+	vmctlFlags := []string{
+		`mimir`,
+		`--mimir-tenant-id=anonymous`,
+		`--mimir-filter-time-start=2024-12-01T00:00:00Z`,
+		`--mimir-filter-time-end=2024-12-31T23:59:59Z`,
+		`--mimir-custom-s3-endpoint=http://localhost:9000`,
+		`--mimir-path=` + path,
+		`--vm-addr=` + vmAddr,
+		`--disable-progress-bar=true`,
+		`--vm-concurrency=6`,
+		`--mimir-concurrency=6`,
+	}
+
+	testMimirProtocol(tc, cluster, vmctlFlags)
+}
+
+func testMimirProtocol(tc *apptest.TestCase, sut apptest.PrometheusWriteQuerier, vmctlFlags []string) {
+	t := tc.T()
+	t.Helper()
+
+	cmpOpt := cmpopts.IgnoreFields(apptest.PrometheusAPIV1QueryResponse{}, "Status", "Data.ResultType")
+
+	// test for empty data request
+	got := sut.PrometheusAPIV1Query(t, `{__name__=~".*"}`, apptest.QueryOpts{
+		Step: "5m",
+		Time: "2025-06-02T17:14:00Z",
+	})
+
+	want := apptest.NewPrometheusAPIV1QueryResponse(t, `{"data":{"result":[]}}`)
+	if diff := cmp.Diff(want, got, cmpOpt); diff != "" {
+		t.Errorf("unexpected response (-want, +got):\n%s", diff)
+	}
+
+	tc.MustStartVmctl("vmctl", vmctlFlags)
+
+	sut.ForceFlush(t)
+
+	// open the expected series response file
+	file, err := os.Open(expectedMimirResponseFile)
+	if err != nil {
+		t.Fatalf("cannot open expected series response file: %s", err)
+	}
+	defer file.Close()
+
+	bytes, err := io.ReadAll(file)
+	if err != nil {
+		t.Fatalf("cannot read expected series response file: %s", err)
+	}
+
+	var wantResponse apptest.PrometheusAPIV1QueryResponse
+	if err := json.Unmarshal(bytes, &wantResponse); err != nil {
+		t.Fatalf("cannot unmarshal expected series response file: %s", err)
+	}
+	wantResponse.Sort()
+
+	tc.Assert(&apptest.AssertOptions{
+		// For cluster version, we need to wait longer for the metrics to be stored
+		Retries: 300,
+		Msg:     `unexpected metrics stored on vmsingle via the prometheus protocol`,
+		Got: func() any {
+			expected := sut.PrometheusAPIV1Export(t, `{__name__=~".*"}`, apptest.QueryOpts{
+				Start: "2024-12-01T15:31:10Z",
+				End:   "2024-12-31T15:32:20Z",
+			})
+			expected.Sort()
+			return expected.Data.Result
+		},
+		Want: wantResponse.Data.Result,
+		CmpOpts: []cmp.Option{
+			cmpopts.IgnoreFields(apptest.PrometheusAPIV1QueryResponse{}, "Status", "Data.ResultType"),
+		},
+	})
+}

dashboards/operator.json

@@ -130,7 +130,7 @@
           "calcs": [
             "lastNotNull"
           ],
-          "fields": "/^short_version$/",
+          "fields": "/^version$/",
           "values": false
         },
         "showPercentChange": false,
@@ -146,11 +146,10 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "vm_app_version{job=~\"$job\",instance=~\"$instance\"}",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "interval": "",
-          "legendFormat": "{{short_version}}",
           "range": false,
           "refId": "A"
         }
@@ -1696,7 +1695,7 @@
         "type": "prometheus",
         "uid": "$ds"
       },
-      "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+      "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
       "fieldConfig": {
         "defaults": {
           "color": {

dashboards/victoriametrics-cluster.json

@@ -791,7 +791,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -1960,7 +1960,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2083,7 +2083,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.\n\nSee also major page faults rate panel in Troubleshooting section if this metric continued to be high.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2727,7 +2727,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nThe lower the better.\n\n- waiting: at least one runnable thread blocked on block-`I/O` (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`; no useful user code ran during these periods → true `I/O` thrashing.\n\nIf stalled > 0 while querying, it's recommended to increase queue depth on NVMe, raise blk-mq budgets, or relax cgroup I/O limits.",
+          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one runnable thread blocked on IO (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`.\n\nIf stalled > 0, consider increasing queue depth on NVMe, raising blk-mq budgets, or relaxing cgroup I/O limits.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3391,7 +3391,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -4110,7 +4110,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows cache miss ratio. Lower is better.",
+          "description": "Shows cache miss ratio.\n\n**Lower is better.**",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -5715,106 +5715,6 @@
           ],
           "title": "Pending",
           "type": "timeseries"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "$ds"
-          },
-          "description": "Network usage by internal VictoriaMetrics RPC protocol",
-          "fieldConfig": {
-            "defaults": {
-              "color": {
-                "mode": "palette-classic"
-              },
-              "custom": {
-                "axisCenteredZero": false,
-                "axisColorMode": "text",
-                "axisLabel": "",
-                "axisPlacement": "auto",
-                "barAlignment": 0,
-                "drawStyle": "line",
-                "fillOpacity": 10,
-                "gradientMode": "none",
-                "hideFrom": {
-                  "legend": false,
-                  "tooltip": false,
-                  "viz": false
-                },
-                "lineInterpolation": "linear",
-                "lineWidth": 1,
-                "pointSize": 5,
-                "scaleDistribution": {
-                  "type": "linear"
-                },
-                "showPoints": "never",
-                "spanNulls": false,
-                "stacking": {
-                  "group": "A",
-                  "mode": "none"
-                },
-                "thresholdsStyle": {
-                  "mode": "off"
-                }
-              },
-              "links": [],
-              "mappings": [],
-              "min": 0,
-              "thresholds": {
-                "mode": "absolute",
-                "steps": [
-                  {
-                    "color": "green"
-                  },
-                  {
-                    "color": "red",
-                    "value": 80
-                  }
-                ]
-              },
-              "unit": "bps"
-            },
-            "overrides": []
-          },
-          "gridPos": {
-            "h": 8,
-            "w": 12,
-            "x": 12,
-            "y": 8388
-          },
-          "id": 74,
-          "options": {
-            "legend": {
-              "calcs": [
-                "mean",
-                "lastNotNull",
-                "max"
-              ],
-              "displayMode": "table",
-              "placement": "bottom",
-              "showLegend": true
-            },
-            "tooltip": {
-              "mode": "multi",
-              "sort": "none"
-            }
-          },
-          "pluginVersion": "9.1.0",
-          "targets": [
-            {
-              "datasource": {
-                "type": "prometheus",
-                "uid": "$ds"
-              },
-              "editorMode": "code",
-              "expr": "sum(rate(vm_tcpdialer_written_bytes_total{job=~\"$job\", instance=~\"$instance\"}[$__rate_interval])) * 8",
-              "legendFormat": "network usage",
-              "range": true,
-              "refId": "A"
-            }
-          ],
-          "title": "RPC network usage ($instance)",
-          "type": "timeseries"
         }
       ],
       "title": "Interconnection ($job)",
@@ -9062,7 +8962,7 @@
                 "uid": "$ds"
               },
               "editorMode": "code",
-              "expr": "sum(rate(vm_rollup_result_cache_misses_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))\n/\nsum(rate(vm_rollup_result_cache_requests_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))",
+              "expr": "sum(rate(vm_rollup_result_cache_miss_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))\n/\nsum(rate(vm_rollup_result_cache_requests_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))",
               "format": "time_series",
               "intervalFactor": 1,
               "legendFormat": "miss",

dashboards/victoriametrics.json

@@ -789,7 +789,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -1954,7 +1954,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2063,7 +2063,7 @@
               "format": "time_series",
               "interval": "",
               "intervalFactor": 2,
-              "legendFormat": "{{job}}-{{instance}} - waiting",
+              "legendFormat": "{{job}}-{{instance}} - stalled",
               "range": true,
               "refId": "B"
             }
@@ -2388,7 +2388,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.\n\nSee also major page faults rate panel in Troubleshooting section if this metric continued to be high.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3023,7 +3023,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\nThe lower the better.",
+          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one runnable thread blocked on IO (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`.\n\nIf stalled > 0, consider increasing queue depth on NVMe, raising blk-mq budgets, or relaxing cgroup I/O limits.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3690,7 +3690,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -4510,7 +4510,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows cache miss ratio. Lower is better.",
+          "description": "Shows cache miss ratio.\n\n**Lower is better.**",
           "fieldConfig": {
             "defaults": {
               "color": {

dashboards/vm/operator.json

@@ -131,7 +131,7 @@
           "calcs": [
             "lastNotNull"
           ],
-          "fields": "/^short_version$/",
+          "fields": "/^version$/",
           "values": false
         },
         "showPercentChange": false,
@@ -147,11 +147,10 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "vm_app_version{job=~\"$job\",instance=~\"$instance\"}",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "interval": "",
-          "legendFormat": "{{short_version}}",
           "range": false,
           "refId": "A"
         }
@@ -1697,7 +1696,7 @@
         "type": "victoriametrics-metrics-datasource",
         "uid": "$ds"
       },
-      "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+      "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
       "fieldConfig": {
         "defaults": {
           "color": {

dashboards/vm/victoriametrics-cluster.json

@@ -792,7 +792,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -1961,7 +1961,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2084,7 +2084,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.\n\nSee also major page faults rate panel in Troubleshooting section if this metric continued to be high.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2728,7 +2728,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nThe lower the better.\n\n- waiting: at least one runnable thread blocked on block-`I/O` (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`; no useful user code ran during these periods → true `I/O` thrashing.\n\nIf stalled > 0 while querying, it's recommended to increase queue depth on NVMe, raise blk-mq budgets, or relax cgroup I/O limits.",
+          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one runnable thread blocked on IO (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`.\n\nIf stalled > 0, consider increasing queue depth on NVMe, raising blk-mq budgets, or relaxing cgroup I/O limits.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3392,7 +3392,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -4111,7 +4111,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows cache miss ratio. Lower is better.",
+          "description": "Shows cache miss ratio.\n\n**Lower is better.**",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -5716,106 +5716,6 @@
           ],
           "title": "Pending",
           "type": "timeseries"
-        },
-        {
-          "datasource": {
-            "type": "victoriametrics-metrics-datasource",
-            "uid": "$ds"
-          },
-          "description": "Network usage by internal VictoriaMetrics RPC protocol",
-          "fieldConfig": {
-            "defaults": {
-              "color": {
-                "mode": "palette-classic"
-              },
-              "custom": {
-                "axisCenteredZero": false,
-                "axisColorMode": "text",
-                "axisLabel": "",
-                "axisPlacement": "auto",
-                "barAlignment": 0,
-                "drawStyle": "line",
-                "fillOpacity": 10,
-                "gradientMode": "none",
-                "hideFrom": {
-                  "legend": false,
-                  "tooltip": false,
-                  "viz": false
-                },
-                "lineInterpolation": "linear",
-                "lineWidth": 1,
-                "pointSize": 5,
-                "scaleDistribution": {
-                  "type": "linear"
-                },
-                "showPoints": "never",
-                "spanNulls": false,
-                "stacking": {
-                  "group": "A",
-                  "mode": "none"
-                },
-                "thresholdsStyle": {
-                  "mode": "off"
-                }
-              },
-              "links": [],
-              "mappings": [],
-              "min": 0,
-              "thresholds": {
-                "mode": "absolute",
-                "steps": [
-                  {
-                    "color": "green"
-                  },
-                  {
-                    "color": "red",
-                    "value": 80
-                  }
-                ]
-              },
-              "unit": "bps"
-            },
-            "overrides": []
-          },
-          "gridPos": {
-            "h": 8,
-            "w": 12,
-            "x": 12,
-            "y": 8388
-          },
-          "id": 74,
-          "options": {
-            "legend": {
-              "calcs": [
-                "mean",
-                "lastNotNull",
-                "max"
-              ],
-              "displayMode": "table",
-              "placement": "bottom",
-              "showLegend": true
-            },
-            "tooltip": {
-              "mode": "multi",
-              "sort": "none"
-            }
-          },
-          "pluginVersion": "9.1.0",
-          "targets": [
-            {
-              "datasource": {
-                "type": "victoriametrics-metrics-datasource",
-                "uid": "$ds"
-              },
-              "editorMode": "code",
-              "expr": "sum(rate(vm_tcpdialer_written_bytes_total{job=~\"$job\", instance=~\"$instance\"}[$__rate_interval])) * 8",
-              "legendFormat": "network usage",
-              "range": true,
-              "refId": "A"
-            }
-          ],
-          "title": "RPC network usage ($instance)",
-          "type": "timeseries"
         }
       ],
       "title": "Interconnection ($job)",
@@ -9063,7 +8963,7 @@
                 "uid": "$ds"
               },
               "editorMode": "code",
-              "expr": "sum(rate(vm_rollup_result_cache_misses_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))\n/\nsum(rate(vm_rollup_result_cache_requests_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))",
+              "expr": "sum(rate(vm_rollup_result_cache_miss_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))\n/\nsum(rate(vm_rollup_result_cache_requests_total{job=~\"$job_select\", instance=~\"$instance\"}[$__rate_interval]))",
               "format": "time_series",
               "intervalFactor": 1,
               "legendFormat": "miss",

dashboards/vm/victoriametrics.json

@@ -790,7 +790,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -1955,7 +1955,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2064,7 +2064,7 @@
               "format": "time_series",
               "interval": "",
               "intervalFactor": 2,
-              "legendFormat": "{{job}}-{{instance}} - waiting",
+              "legendFormat": "{{job}}-{{instance}} - stalled",
               "range": true,
               "refId": "B"
             }
@@ -2389,7 +2389,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.\n\nSee also major page faults rate panel in Troubleshooting section if this metric continued to be high.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3024,7 +3024,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\nThe lower the better.",
+          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one runnable thread blocked on IO (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`.\n\nIf stalled > 0, consider increasing queue depth on NVMe, raising blk-mq budgets, or relaxing cgroup I/O limits.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3691,7 +3691,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -4511,7 +4511,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows cache miss ratio. Lower is better.",
+          "description": "Shows cache miss ratio.\n\n**Lower is better.**",
           "fieldConfig": {
             "defaults": {
               "color": {

dashboards/vm/vmagent.json

@@ -785,7 +785,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -2042,7 +2042,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2165,7 +2165,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2542,7 +2542,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\nThe lower the better.",
+          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one runnable thread blocked on IO (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`.\n\nIf stalled > 0, consider increasing queue depth on NVMe, raising blk-mq budgets, or relaxing cgroup I/O limits.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3015,7 +3015,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {

dashboards/vm/vmalert.json

@@ -566,7 +566,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -1719,7 +1719,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -1840,7 +1840,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2177,7 +2177,7 @@
             "type": "victoriametrics-metrics-datasource",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2804,10 +2804,10 @@
             "overrides": []
           },
           "gridPos": {
-            "h": 8,
+            "h": 7,
             "w": 12,
             "x": 0,
-            "y": 352
+            "y": 11
           },
           "id": 63,
           "options": {
@@ -2843,7 +2843,113 @@
           ],
           "title": "Restarts ($job)",
           "type": "timeseries"
-        }
+        },
+        {
+          "datasource": {
+            "type": "victoriametrics-metrics-datasource",
+            "uid": "$ds"
+          },
+          "description": "Group iteration reset can be caused by irregular delays during evaluation or by the system wall clock being moved backward.\nIf it is caused by host clock changes, vmalert could generate duplicate results for the group rules, since some evaluations could be repeated.\nCheck the host clock time synchronization configuration if this happens frequently.\n",
+          "fieldConfig": {
+            "defaults": {
+              "color": {
+                "mode": "palette-classic"
+              },
+              "custom": {
+                "axisBorderShow": false,
+                "axisCenteredZero": false,
+                "axisColorMode": "text",
+                "axisLabel": "",
+                "axisPlacement": "auto",
+                "barAlignment": 0,
+                "barWidthFactor": 0.6,
+                "drawStyle": "bars",
+                "fillOpacity": 10,
+                "gradientMode": "none",
+                "hideFrom": {
+                  "legend": false,
+                  "tooltip": false,
+                  "viz": false
+                },
+                "insertNulls": false,
+                "lineInterpolation": "linear",
+                "lineWidth": 1,
+                "pointSize": 5,
+                "scaleDistribution": {
+                  "type": "linear"
+                },
+                "showPoints": "never",
+                "showValues": false,
+                "spanNulls": false,
+                "stacking": {
+                  "group": "A",
+                  "mode": "none"
+                },
+                "thresholdsStyle": {
+                  "mode": "off"
+                }
+              },
+              "mappings": [],
+              "thresholds": {
+                "mode": "absolute",
+                "steps": [
+                  {
+                    "color": "green",
+                    "value": 0
+                  },
+                  {
+                    "color": "red",
+                    "value": 80
+                  }
+                ]
+              },
+              "unit": "short"
+            },
+            "overrides": []
+          },
+          "gridPos": {
+            "h": 7,
+            "w": 12,
+            "x": 12,
+            "y": 11
+          },
+          "id": 70,
+          "options": {
+            "legend": {
+              "calcs": [
+                "mean",
+                "lastNotNull",
+                "max"
+              ],
+              "displayMode": "table",
+              "placement": "bottom",
+              "showLegend": true
+            },
+            "tooltip": {
+              "hideZeros": false,
+              "mode": "multi",
+              "sort": "none"
+            }
+          },
+          "pluginVersion": "12.2.0",
+          "targets": [
+            {
+              "datasource": {
+                "type": "victoriametrics-metrics-datasource",
+                "uid": "$ds"
+              },
+              "editorMode": "code",
+              "exemplar": false,
+              "expr": "sum(increase(vmalert_iteration_reset_total{job=~\"$job\", instance=~\"$instance\"}[$__rate_interval])) by(job, group, file) > 0",
+              "interval": "1m",
+              "legendFormat": "({{job}}) {{group}}({{file}})",
+              "range": true,
+              "refId": "A"
+            }
+          ],
+          "title": "Group Iteration Reset ($instance)",
+          "type": "timeseries"
+        }        
       ],
       "title": "Troubleshooting",
       "type": "row"

dashboards/vm/vmauth.json

@@ -492,14 +492,14 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by (job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
           "refId": "A"
         }
       ],
-      "title": "Version",
+      "title": "",
       "type": "table"
     },
     {

dashboards/vmagent.json

@@ -784,7 +784,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -2041,7 +2041,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2164,7 +2164,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2541,7 +2541,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\nThe lower the better.",
+          "description": "Shows IO pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one runnable thread blocked on IO (disk, NVMe, network-storage) while others could still make progress.\n- stalled: all non-idle threads simultaneously waiting on `I/O`.\n\nIf stalled > 0, consider increasing queue depth on NVMe, raising blk-mq budgets, or relaxing cgroup I/O limits.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -3014,7 +3014,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {

dashboards/vmalert.json

@@ -565,7 +565,7 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by(job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
@@ -1718,7 +1718,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html). Helps troubleshoot high CPU usage or throttling.\nLower is better.\n\n- waiting: The percentage of time at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: The percentage of time all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 1%. It only becomes a concern if it consistently climbs above 5–10% and aligns with latency spikes or GC slowdowns.",
+          "description": "CPU pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one task in the process was ready to run (runnable) but couldn't get scheduled on the CPU.\n- stalled: all tasks in the process (except idle ones) were unable to get CPU time — a full CPU stall.\n\nIf there's a CPU burst, it's normal to see waiting or stalled > 100ms. It only becomes a concern if it consistently climbs above 50-100ms and aligns with latency spikes or GC slowdowns.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -1839,7 +1839,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\nLower is better.\n\n- waiting: Time fraction where at least one thread was blocked on memory.\n- stalled: Time fraction where every thread was blocked on memory (severe pressure).\n\nIf queries slow down and both series spike, the host is likely limited by RAM or I/O throughput.",
+          "description": "Shows memory pressure based on [Pressure Stall Information](https://docs.kernel.org/accounting/psi.html).\n\n**Lower is better.**\n\nPressure is measured as amount of time within 1sec time window the process was:\n- waiting: at least one thread was blocked on memory.\n- stalled: every thread was blocked on memory (severe pressure).\n\nElevated memory pressure can slowdown the process performance by utilizing more disk IO. Consider increasing amount of available RAM limit or decreasing the load on the process.\n\nSeу major page faults rate panel in Troubleshooting section if this metric continued to be high.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2176,7 +2176,7 @@
             "type": "prometheus",
             "uid": "$ds"
           },
-          "description": "Shows the time goroutines have spent in runnable state before actually running. The lower is better.\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
+          "description": "Shows the time goroutines have spent in runnable state before actually running.\n\n**Lower is better.**\n\nHigh values or values exceeding the threshold is usually a sign of            insufficient CPU resources or CPU throttling. \n\nVerify that service has enough CPU resources. Otherwise, the service could work unreliably with delays in processing.",
           "fieldConfig": {
             "defaults": {
               "color": {
@@ -2803,10 +2803,10 @@
             "overrides": []
           },
           "gridPos": {
-            "h": 8,
+            "h": 7,
             "w": 12,
             "x": 0,
-            "y": 352
+            "y": 11
           },
           "id": 63,
           "options": {
@@ -2842,7 +2842,113 @@
           ],
           "title": "Restarts ($job)",
           "type": "timeseries"
-        }
+        },
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "$ds"
+          },
+          "description": "Group iteration reset can be caused by irregular delays during evaluation or by the system wall clock being moved backward.\nIf it is caused by host clock changes, vmalert could generate duplicate results for the group rules, since some evaluations could be repeated.\nCheck the host clock time synchronization configuration if this happens frequently.\n",
+          "fieldConfig": {
+            "defaults": {
+              "color": {
+                "mode": "palette-classic"
+              },
+              "custom": {
+                "axisBorderShow": false,
+                "axisCenteredZero": false,
+                "axisColorMode": "text",
+                "axisLabel": "",
+                "axisPlacement": "auto",
+                "barAlignment": 0,
+                "barWidthFactor": 0.6,
+                "drawStyle": "bars",
+                "fillOpacity": 10,
+                "gradientMode": "none",
+                "hideFrom": {
+                  "legend": false,
+                  "tooltip": false,
+                  "viz": false
+                },
+                "insertNulls": false,
+                "lineInterpolation": "linear",
+                "lineWidth": 1,
+                "pointSize": 5,
+                "scaleDistribution": {
+                  "type": "linear"
+                },
+                "showPoints": "never",
+                "showValues": false,
+                "spanNulls": false,
+                "stacking": {
+                  "group": "A",
+                  "mode": "none"
+                },
+                "thresholdsStyle": {
+                  "mode": "off"
+                }
+              },
+              "mappings": [],
+              "thresholds": {
+                "mode": "absolute",
+                "steps": [
+                  {
+                    "color": "green",
+                    "value": 0
+                  },
+                  {
+                    "color": "red",
+                    "value": 80
+                  }
+                ]
+              },
+              "unit": "short"
+            },
+            "overrides": []
+          },
+          "gridPos": {
+            "h": 7,
+            "w": 12,
+            "x": 12,
+            "y": 11
+          },
+          "id": 70,
+          "options": {
+            "legend": {
+              "calcs": [
+                "mean",
+                "lastNotNull",
+                "max"
+              ],
+              "displayMode": "table",
+              "placement": "bottom",
+              "showLegend": true
+            },
+            "tooltip": {
+              "hideZeros": false,
+              "mode": "multi",
+              "sort": "none"
+            }
+          },
+          "pluginVersion": "12.2.0",
+          "targets": [
+            {
+              "datasource": {
+                "type": "prometheus",
+                "uid": "$ds"
+              },
+              "editorMode": "code",
+              "exemplar": false,
+              "expr": "sum(increase(vmalert_iteration_reset_total{job=~\"$job\", instance=~\"$instance\"}[$__rate_interval])) by(job, group, file) > 0",
+              "interval": "1m",
+              "legendFormat": "({{job}}) {{group}}({{file}})",
+              "range": true,
+              "refId": "A"
+            }
+          ],
+          "title": "Group Iteration Reset ($instance)",
+          "type": "timeseries"
+        }        
       ],
       "title": "Troubleshooting",
       "type": "row"

dashboards/vmauth.json

@@ -491,14 +491,14 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "sum(vm_app_version{job=~\"$job\", instance=~\"$instance\"}) by (job, short_version)",
+          "expr": "sum by(job, version) (label_replace(vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version!=\"\"}, \"version\", \"$1\", \"short_version\", \"(.*)\") or vm_app_version{job=~\"$job\", instance=~\"$instance\", short_version=\"\"})",
           "format": "table",
           "instant": true,
           "range": false,
           "refId": "A"
         }
       ],
-      "title": "Version",
+      "title": "",
       "type": "table"
     },
     {

deployment/docker/Makefile

@@ -7,7 +7,7 @@ ROOT_IMAGE ?= alpine:3.23.4
 ROOT_IMAGE_SCRATCH ?= scratch
 CERTS_IMAGE := alpine:3.23.4
 
-GO_BUILDER_IMAGE := golang:1.26.3
+GO_BUILDER_IMAGE := golang:1.26.4
 
 BUILDER_IMAGE := local/builder:2.0.0-$(shell echo $(GO_BUILDER_IMAGE) | tr :/ __)-1
 BASE_IMAGE := local/base:1.1.4-$(shell echo $(ROOT_IMAGE) | tr :/ __)-$(shell echo $(CERTS_IMAGE) | tr :/ __)

deployment/docker/compose-vm-cluster.yml

@@ -3,7 +3,7 @@ services:
   #  It scrapes targets defined in --promscrape.config
   #  And forward them to --remoteWrite.url
   vmagent:
-    image: victoriametrics/vmagent:v1.144.0
+    image: victoriametrics/vmagent:v1.145.0
     depends_on:
       - "vmauth"
     ports:
@@ -42,14 +42,14 @@ services:
   # vmstorage shards. Each shard receives 1/N of all metrics sent to vminserts,
   # where N is number of vmstorages (2 in this case).
   vmstorage-1:
-    image: victoriametrics/vmstorage:v1.144.0-cluster
+    image: victoriametrics/vmstorage:v1.145.0-cluster
     volumes:
       - strgdata-1:/storage
     command:
       - "--storageDataPath=/storage"
     restart: always
   vmstorage-2:
-    image: victoriametrics/vmstorage:v1.144.0-cluster
+    image: victoriametrics/vmstorage:v1.145.0-cluster
     volumes:
       - strgdata-2:/storage
     command:
@@ -59,7 +59,7 @@ services:
   # vminsert is ingestion frontend. It receives metrics pushed by vmagent,
   # pre-process them and distributes across configured vmstorage shards.
   vminsert-1:
-    image: victoriametrics/vminsert:v1.144.0-cluster
+    image: victoriametrics/vminsert:v1.145.0-cluster
     depends_on:
       - "vmstorage-1"
       - "vmstorage-2"
@@ -68,7 +68,7 @@ services:
       - "--storageNode=vmstorage-2:8400"
     restart: always
   vminsert-2:
-    image: victoriametrics/vminsert:v1.144.0-cluster
+    image: victoriametrics/vminsert:v1.145.0-cluster
     depends_on:
       - "vmstorage-1"
       - "vmstorage-2"
@@ -80,7 +80,7 @@ services:
   # vmselect is a query fronted. It serves read queries in MetricsQL or PromQL.
   # vmselect collects results from configured `--storageNode` shards.
   vmselect-1:
-    image: victoriametrics/vmselect:v1.144.0-cluster
+    image: victoriametrics/vmselect:v1.145.0-cluster
     depends_on:
       - "vmstorage-1"
       - "vmstorage-2"
@@ -90,7 +90,7 @@ services:
       - "--vmalert.proxyURL=http://vmalert:8880"
     restart: always
   vmselect-2:
-    image: victoriametrics/vmselect:v1.144.0-cluster
+    image: victoriametrics/vmselect:v1.145.0-cluster
     depends_on:
       - "vmstorage-1"
       - "vmstorage-2"
@@ -105,7 +105,7 @@ services:
   # read requests from Grafana, vmui, vmalert among vmselects.
   # It can be used as an authentication proxy.
   vmauth:
-    image: victoriametrics/vmauth:v1.144.0
+    image: victoriametrics/vmauth:v1.145.0
     depends_on:
       - "vmselect-1"
       - "vmselect-2"
@@ -119,7 +119,7 @@ services:
 
   # vmalert executes alerting and recording rules
   vmalert:
-    image: victoriametrics/vmalert:v1.144.0
+    image: victoriametrics/vmalert:v1.145.0
     depends_on:
       - "vmauth"
     ports:

deployment/docker/compose-vm-single.yml

@@ -3,7 +3,7 @@ services:
   #  It scrapes targets defined in --promscrape.config
   #  And forward them to --remoteWrite.url
   vmagent:
-    image: victoriametrics/vmagent:v1.144.0
+    image: victoriametrics/vmagent:v1.145.0
     depends_on:
       - "victoriametrics"
     ports:
@@ -18,7 +18,7 @@ services:
   # VictoriaMetrics instance, a single process responsible for
   # storing metrics and serve read requests.
   victoriametrics:
-    image: victoriametrics/victoria-metrics:v1.144.0
+    image: victoriametrics/victoria-metrics:v1.145.0
     ports:
       - 8428:8428
       - 8089:8089
@@ -59,7 +59,7 @@ services:
 
   # vmalert executes alerting and recording rules
   vmalert:
-    image: victoriametrics/vmalert:v1.144.0
+    image: victoriametrics/vmalert:v1.145.0
     depends_on:
       - "victoriametrics"
       - "alertmanager"

deployment/docker/rules/alerts-vmalert.yml

@@ -64,6 +64,18 @@ groups:
             group \"{{ $labels.group }}\". See https://docs.victoriametrics.com/victoriametrics/vmalert/#groups. 
             If rule expressions are taking longer than expected, please see https://docs.victoriametrics.com/victoriametrics/troubleshooting/#slow-queries."
 
+      - alert: GroupIterationReset
+        expr: increase(vmalert_iteration_reset_total[5m]) > 0
+        for: 5m
+        labels:
+          severity: warning
+        annotations:
+          summary: "Evaluation iteration for group {{ $labels.group }} in file {{ $labels.file }} is reset"
+          description: "Evaluation iteration for group \"{{ $labels.group }}\" in file \"{{ $labels.file }}\" is reset on vmalert instance {{ $labels.instance }}.
+            This can be caused by irregular delays during evaluation or by the system wall clock being moved backward. If it is caused by host clock changes, vmalert could
+            generate duplicate results for the group rules since some evaluations could be repeated. Check host clock time synchronization configurations if this happens frequently."
+
+
       - alert: RemoteWriteErrors
         expr: increase(vmalert_remotewrite_errors_total[5m]) > 0
         for: 15m
@@ -108,4 +120,3 @@ groups:
           summary: "vmalert instance {{ $labels.instance }} is failing to send notifications to Alertmanager"
           description: "vmalert instance {{ $labels.instance }} is failing to send alert notifications to \"{{ $labels.addr }}\".
             Check vmalert's logs for detailed error message."
-

deployment/docker/rules/alerts-vmanomaly.yml

@@ -40,7 +40,7 @@ groups:
         annotations:
           summary: "Metrics have not been seen from \"{{ $labels.job }}\"(\"{{ $labels.instance }}\") for {{ $value }} seconds"
           description: >
-           The missing metric may indicate that vmanomaly is not running or is inaccessible from vmagent or the remotewrite endpoint.
+            The missing metric may indicate that vmanomaly is not running or is inaccessible from vmagent or the remotewrite endpoint.
 
       - alert: ProcessNearFDLimits
         expr: (process_max_fds{job=~".*vmanomaly.*"} - process_open_fds{job=~".*vmanomaly.*"}) < 100

deployment/docker/vmanomaly/vmanomaly-integration/compose.yml

@@ -1,6 +1,6 @@
 services:
   vmagent:
-    image: victoriametrics/vmagent:v1.144.0
+    image: victoriametrics/vmagent:v1.145.0
     depends_on:
       - "victoriametrics"
     ports:
@@ -14,7 +14,7 @@ services:
     restart: always
 
   victoriametrics:
-    image: victoriametrics/victoria-metrics:v1.144.0
+    image: victoriametrics/victoria-metrics:v1.145.0
     ports:
       - 8428:8428
     volumes:
@@ -40,7 +40,7 @@ services:
     restart: always
 
   vmalert:
-    image: victoriametrics/vmalert:v1.144.0
+    image: victoriametrics/vmalert:v1.145.0
     depends_on:
       - "victoriametrics"
     ports:

docs/Makefile

@@ -271,6 +271,8 @@ endif
 	(cd /tmp/vm-opensource && ./bin/vmctl remote-read -help > /tmp/vmctl_remote-read_flags_tmp.md)
 	(cd /tmp/vm-opensource && ./bin/vmctl prometheus -help > /tmp/vmctl_prometheus_flags_tmp.md)
 	(cd /tmp/vm-opensource && ./bin/vmctl vm-native -help > /tmp/vmctl_vm-native_flags_tmp.md)
+	(cd /tmp/vm-opensource && ./bin/vmctl thanos -help > /tmp/vmctl_thanos_flags_tmp.md)
+	(cd /tmp/vm-opensource && ./bin/vmctl mimir -help > /tmp/vmctl_mimir_flags_tmp.md)
 
 	echo "$$FLAGS_HEADER" > docs/victoriametrics/vmctl/vmctl_flags.md && \
 	cat /tmp/vmctl_flags_tmp.md >> docs/victoriametrics/vmctl/vmctl_flags.md && \
@@ -296,6 +298,14 @@ endif
 	cat /tmp/vmctl_vm-native_flags_tmp.md >> docs/victoriametrics/vmctl/vmctl_vm-native_flags.md && \
 	printf '```\n' >> docs/victoriametrics/vmctl/vmctl_vm-native_flags.md
 
+	echo "$$FLAGS_HEADER" > docs/victoriametrics/vmctl/vmctl_thanos_flags.md && \
+	cat /tmp/vmctl_thanos_flags_tmp.md >> docs/victoriametrics/vmctl/vmctl_thanos_flags.md && \
+	printf '```\n' >> docs/victoriametrics/vmctl/vmctl_thanos_flags.md
+
+	echo "$$FLAGS_HEADER" > docs/victoriametrics/vmctl/vmctl_mimir_flags.md && \
+	cat /tmp/vmctl_mimir_flags_tmp.md >> docs/victoriametrics/vmctl/vmctl_mimir_flags.md && \
+	printf '```\n' >> docs/victoriametrics/vmctl/vmctl_mimir_flags.md
+
 	# remove Total time line from all vmctl flag files to reduce diffs noise
 	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_flags.md
 	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_opentsdb_flags.md
@@ -303,6 +313,8 @@ endif
 	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_remote-read_flags.md
 	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_prometheus_flags.md
 	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_vm-native_flags.md
+	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_thanos_flags.md
+	sed -i '/Total time:/d' docs/victoriametrics/vmctl/vmctl_mimir_flags.md
 
 	# remove Version line and the actual version line from vmctl_flags.md to reduce diffs noise
 	sed -i '/^VERSION:/,+1d' docs/victoriametrics/vmctl/vmctl_flags.md

docs/anomaly-detection/CHANGELOG.md

@@ -130,7 +130,7 @@ Released: 2025-11-05
 ## v1.27.0
 Released: 2025-10-31
 
-- FEATURE: Added runtime state compatibility guard for [stateful](https://docs.victoriametrics.com/anomaly-detection/components/settings/#restore-state) deployments. The service now persists normalized versions, evaluates an [upgrade/downgrade compatibility matrix](https://docs.victoriametrics.com/anomaly-detection/migration/#compatibility-matrix), and selectively drops or reuses DB records and on-disk artifacts to keep migrations safe and automatic. Please refer to the [migration page](https://docs.victoriametrics.com/anomaly-detection/migration/) for more details.
+- FEATURE: Added runtime state compatibility guard for [stateful](https://docs.victoriametrics.com/anomaly-detection/components/settings/#state-restoration) deployments. The service now persists normalized versions, evaluates an [upgrade/downgrade compatibility matrix](https://docs.victoriametrics.com/anomaly-detection/migration/#compatibility-matrix), and selectively drops or reuses DB records and on-disk artifacts to keep migrations safe and automatic. Please refer to the [migration page](https://docs.victoriametrics.com/anomaly-detection/migration/) for more details.
 
 - IMPROVEMENT: Parallelization now honours container cgroup CPU/RAM limits, so `settings.n_workers` in the [settings section](https://docs.victoriametrics.com/anomaly-detection/components/settings/#parallelization), internal routines and the `vmanomaly_available_memory_bytes`/`vmanomaly_cpu_cores_available` [startup metrics](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#startup-metrics) report or use container resources instead of host totals, keeping the [self-monitoring dashboard](https://docs.victoriametrics.com/anomaly-detection/self-monitoring/#grafana-dashboard) accurate.
 
@@ -164,7 +164,7 @@ Released: 2025-10-02
 
 - FEATURE: Introduced vmui-like [UI](https://docs.victoriametrics.com/anomaly-detection/ui/) for `vmanomaly` service to simplify the configuration and backtesting of anomaly detection models before it goes to production. It provides an intuitive interface to finetune model configurations, visualize its predictions and anomaly scores, and perform backtesting on historical data. The UI is accessible via a web browser and can be run as a [standalone service](https://docs.victoriametrics.com/anomaly-detection/ui/#preset-usage) or [integrated with productionalized deployments](https://docs.victoriametrics.com/anomaly-detection/ui/#mixed-usage). For more details, refer to the [documentation](https://docs.victoriametrics.com/anomaly-detection/ui/).
 
-- FEATURE: Added support for reading data from [VictoriaLogs stats queries](https://docs.victoriametrics.com/victorialogs/querying/#querying-log-range-stats) with `VLogsReader`. This reader allows querying and analyzing log data stored in VictoriaLogs, enabling anomaly detection on metrics generated from logs. It supports similar configuration options as `VmReader`, including `datasource_url`, `tenant_id`, `queries`, etc. For more details, refer to the [documentation](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vlogs-reader). It can be also used in [UI mode](https://docs.victoriametrics.com/anomaly-detection/ui/) for backtesting log-based anomaly detection configurations.
+- FEATURE: Added support for reading data from [VictoriaLogs stats queries](https://docs.victoriametrics.com/victorialogs/querying/#querying-log-range-stats) with `VLogsReader`. This reader allows querying and analyzing log data stored in VictoriaLogs, enabling anomaly detection on metrics generated from logs. It supports similar configuration options as `VmReader`, including `datasource_url`, `tenant_id`, `queries`, etc. For more details, refer to the [documentation](https://docs.victoriametrics.com/anomaly-detection/components/reader/#victorialogs-reader). It can be also used in [UI mode](https://docs.victoriametrics.com/anomaly-detection/ui/) for backtesting log-based anomaly detection configurations.
 
 - IMPROVEMENT: Resolved the case in the [`IsolationForestModel`](https://docs.victoriametrics.com/anomaly-detection/components/models/#isolation-forest-multivariate) with `provide_series` common model [argument](https://docs.victoriametrics.com/anomaly-detection/components/models/#provide-series) including `yhat.*` series (prediction and confidence boundaries), which are not produced by this model. Now config validation will fail with a clear error message if such series names are requested.
 
@@ -190,7 +190,7 @@ Released: 2025-08-19
 ## v1.25.2
 Released: 2025-07-30
 
-- BUGFIX: Resolved inconsistent state between in-memory models and state database (if [stateful mode](https://docs.victoriametrics.com/anomaly-detection/components/settings/#stateful-mode) is enabled). This bug caused `Model instance not found` warnings during inference calls and prevented proper cleanup of stale models from disk. The fix also prevents state updates when operations are terminated mid-execution of scheduled fit/infer jobs.
+- BUGFIX: Resolved inconsistent state between in-memory models and state database (if [stateful mode](https://docs.victoriametrics.com/anomaly-detection/components/settings/#state-restoration) is enabled). This bug caused `Model instance not found` warnings during inference calls and prevented proper cleanup of stale models from disk. The fix also prevents state updates when operations are terminated mid-execution of scheduled fit/infer jobs.
 
 - BUGFIX: Added explicit handling for inference calls on models that were deleted from disk by the time of their usage, but still referenced in the state database, preventing `'NoneType' object has no attribute 'infer'` rows in logs. Now a warning is logged and the inference call is skipped, which is expected behavior for deleted models.
 
@@ -210,7 +210,7 @@ Released: 2025-07-24
 
 - BUGFIX: Prevented `OneOffScheduler` and `BacktestingScheduler` [schedulers](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/) from receiving no data (when [state restoration](https://docs.victoriametrics.com/anomaly-detection/components/settings/#state-restoration) is enabled). Now a warning is logged and such scheduler types are implicitly used without state restoration, which is expected behavior for these one-time-job schedulers.
 
-- BUGFIX: Now the paths to artifact database (if [stateful mode](https://docs.victoriametrics.com/anomaly-detection/components/settings/#stateful-mode) is enabled) are properly resolved to absolute, preventing errors at initialization time (like `sqlalchemy.exc.OperationalError: (sqlite3.OperationalError) unable to open database file`) or warnings (like `SAWarning: fully NULL primary key identity cannot load any object.`).
+- BUGFIX: Now the paths to artifact database (if [stateful mode](https://docs.victoriametrics.com/anomaly-detection/components/settings/#state-restoration) is enabled) are properly resolved to absolute, preventing errors at initialization time (like `sqlalchemy.exc.OperationalError: (sqlite3.OperationalError) unable to open database file`) or warnings (like `SAWarning: fully NULL primary key identity cannot load any object.`).
 
 ## v1.25.0
 Released: 2025-07-17
@@ -273,7 +273,7 @@ Released: 2025-06-05
 
 - FEATURE: Added `decay` [argument](https://docs.victoriametrics.com/anomaly-detection/components/models/#decay) to [online models](https://docs.victoriametrics.com/anomaly-detection/components/models/#online-models). This parameters allows for newer data to be weighted more heavily in online models. By default this is set to 1 which means all data points are weighted the same to maintain backward compatibility with existing configs. The closer this value is to 0 the more important new data is.
 
-- IMPROVEMENT: **Restored back parallelization** in the read/fit/infer pipeline, previously disabled in [v1.22.0](#v1220-experimental) due to deadlock issues. The new implementation prevents deadlocks, allowing to control the parallelization level via `n_workers` in [settings section](https://docs.victoriametrics.com/anomaly-detection/components/settings/). It's suggested to upgrade from [v1.22.0](#v1220) - [v1.22.1](#v1221) to this version to regain the performance benefits of parallel processing.
+- IMPROVEMENT: **Restored back parallelization** in the read/fit/infer pipeline, previously disabled in [v1.22.0](#v1220-experimental) due to deadlock issues. The new implementation prevents deadlocks, allowing to control the parallelization level via `n_workers` in [settings section](https://docs.victoriametrics.com/anomaly-detection/components/settings/). It's suggested to upgrade from [v1.22.0](#v1220-experimental) - [v1.22.1](#v1221) to this version to regain the performance benefits of parallel processing.
 
 - IMPROVEMENT: Added `--dryRun` [argument](https://docs.victoriametrics.com/anomaly-detection/quickstart/#command-line-arguments) to `vmanomaly` to enable dry run mode. This mode allows to validate configuration without executing any actual operations and doesn't require a license. It is particularly useful to test the configurations before deploying them in a production environment.
 
@@ -559,7 +559,7 @@ Released: 2024-08-10
   - **Lowest anomaly scores** (=0) when the *model's predictions (`yhat`) fall outside the expected range*, signaling uncertain predictions.
   - For more details, please refer to the [documentation](https://docs.victoriametrics.com/anomaly-detection/components/reader/#per-query-parameters).
 
-- IMPROVEMENT: Added `latency_offset` argument to the [VmReader](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader) to override the default `-search.latencyOffset` [flag of VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/#list-of-command-line-flags) (30s). The default value is set to 1ms, which should help in cases where `sampling_frequency` is low (10-60s) and `sampling_frequency` equals `infer_every` in the [PeriodicScheduler](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/#periodic-scheduler). This prevents users from receiving `service - WARNING - [Scheduler [scheduler_alias]] No data available for inference.` warnings in logs and allows for consecutive `infer` calls without gaps. To restore the backward compatible behavior, set it equal to your `-search.latencyOffset` value in [VmReader](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader) config section.
+- IMPROVEMENT: Added `latency_offset` argument to the [VmReader](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader) to override the default `-search.latencyOffset` [flag of VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/#list-of-command-line-flags) (30s). The default value is set to 1ms, which should help in cases where `sampling_period` is low (10-60s) and `sampling_period` equals `infer_every` in the [PeriodicScheduler](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/#periodic-scheduler). This prevents users from receiving `service - WARNING - [Scheduler [scheduler_alias]] No data available for inference.` warnings in logs and allows for consecutive `infer` calls without gaps. To restore the backward compatible behavior, set it equal to your `-search.latencyOffset` value in [VmReader](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader) config section.
 
 - BUGFIX: Ensure the `use_transform` argument of the [`OnlineQuantileModel`](https://docs.victoriametrics.com/anomaly-detection/components/models/#online-seasonal-quantile) functions as intended.
 - BUGFIX: Add a docstring for `query_from_last_seen_timestamp` arg of [VmReader](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader).

docs/anomaly-detection/FAQ.md

@@ -53,7 +53,7 @@ Please see example graph illustrating this logic below:
 
 **VictoriaMetrics (metrics):** use full [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/) for selection, sampling, and processing; [global filters](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#prometheus-querying-api-enhancements) are also supported. See the [VmReader](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader) for the details.
 
-**VictoriaLogs (logs → metrics):** {{% available_from "v1.26.0" anomaly %}} use [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/) via the [`VLogsReader`](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vlogs-reader) to create log-derived or traces-derived metrics for anomaly detection (e.g., error rates, request latencies, error spans count). 
+**VictoriaLogs (logs → metrics):** {{% available_from "v1.26.0" anomaly %}} use [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/) via the [`VLogsReader`](https://docs.victoriametrics.com/anomaly-detection/components/reader/#victorialogs-reader) to create log-derived or traces-derived metrics for anomaly detection (e.g., error rates, request latencies, error spans count). 
 
 > [!NOTE]
 > Please note that only LogsQL queries with [stats pipe](https://docs.victoriametrics.com/victorialogs/logsql/#stats-pipe) functions [subset](https://docs.victoriametrics.com/anomaly-detection/components/reader/#valid-stats-functions) are supported, as they produce **numeric** time series.
@@ -281,7 +281,7 @@ reader:
   datasource_url: 'some_url_to_read_data_from'
   queries:
     query_alias1: 'some_metricsql_query'
-  sampling_frequency: '1m'  # change to whatever you need in data granularity
+  sampling_period: '1m'  # change to whatever you need in data granularity
   # other params if needed
   # https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader
 
@@ -294,7 +294,7 @@ writer:
 # https://docs.victoriametrics.com/anomaly-detection/components/monitoring/
 ```
 
-Configuration above will produce N intervals of full length (`fit_window`=14d + `fit_every`=1h) until `to_iso` timestamp is reached to run N consecutive `fit` calls to train models; Then these models will be used to produce `M = [fit_every / sampling_frequency]` infer datapoints for `fit_every` range at the end of each such interval, imitating M consecutive calls of `infer_every` in `PeriodicScheduler` [config](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/#periodic-scheduler). These datapoints then will be written back to VictoriaMetrics TSDB, defined in `writer` [section](https://docs.victoriametrics.com/anomaly-detection/components/writer/#vm-writer) for further visualization (i.e. in VMUI or Grafana)
+Configuration above will produce N intervals of full length (`fit_window`=14d + `fit_every`=1h) until `to_iso` timestamp is reached to run N consecutive `fit` calls to train models; Then these models will be used to produce `M = [fit_every / sampling_period]` infer datapoints for `fit_every` range at the end of each such interval, imitating M consecutive calls of `infer_every` in `PeriodicScheduler` [config](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/#periodic-scheduler). These datapoints then will be written back to VictoriaMetrics TSDB, defined in `writer` [section](https://docs.victoriametrics.com/anomaly-detection/components/writer/#vm-writer) for further visualization (i.e. in VMUI or Grafana)
 
 ## Forecasting
 
@@ -499,7 +499,7 @@ schedulers:
 models:
   zscore_example:
     class: 'zscore_online'
-    min_n_samples_seen: 120  # i.e. minimal relevant seasonality or (initial) fit_window / sampling_frequency
+    min_n_samples_seen: 120  # i.e. minimal relevant seasonality or (initial) fit_window / sampling_period
     decay: 0.999  # decay factor to control how fast the model adapts to new data, the lower, the faster it adapts
     schedulers: ['periodic']
     # other model params ...

docs/anomaly-detection/QuickStart.md

@@ -406,10 +406,10 @@ For optimal service behavior, consider the following tweaks when configuring `vm
 
 **Reader**:
 - Setup the datasource to read data from in the [reader](https://docs.victoriametrics.com/anomaly-detection/components/reader/) section. Include tenant ID if using a [cluster version of VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/) (`multitenant` value {{% available_from "v1.16.2" anomaly %}} can be also used here).
-- Define queries for input data using [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/) under `reader.queries` section. Note, it's possible to override reader-level arguments at query level for increased flexibility, e.g. specifying per-query [timezone](https://docs.victoriametrics.com/anomaly-detection/faq/#handling-timezones) or [sampling period](https://docs.victoriametrics.com/anomaly-detection/components/reader/#sampling-period).
+- Define queries for input data using [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/) under `reader.queries` section. Note, it's possible to override reader-level arguments at query level for increased flexibility, e.g. specifying per-query [timezone](https://docs.victoriametrics.com/anomaly-detection/faq/#handling-timezones) or [sampling period](https://docs.victoriametrics.com/anomaly-detection/components/reader/#config-parameters).
 - For longer `fit_window` intervals in scheduler, consider splitting queries into smaller time ranges to avoid excessive memory usage, timeouts and hitting server-side constraints, so they can be queried separately and reconstructed on `vmanomaly` side. Please refer to this [example](https://docs.victoriametrics.com/anomaly-detection/faq/#handling-large-queries-in-vmanomaly) for more details.
 
-> If applicable - consider [`VLogsReader`](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vlogs-reader) {{% available_from "v1.26.0" anomaly %}} to perform anomaly detection on **log-derived metrics**. This is particularly useful for scenarios where log data needs to be analyzed for unusual patterns or behaviors, such as error rates or request latencies.
+> If applicable - consider [`VLogsReader`](https://docs.victoriametrics.com/anomaly-detection/components/reader/#victorialogs-reader) {{% available_from "v1.26.0" anomaly %}} to perform anomaly detection on **log-derived metrics**. This is particularly useful for scenarios where log data needs to be analyzed for unusual patterns or behaviors, such as error rates or request latencies.
 
 **Writer**:
 - Specify where and how to store anomaly detection metrics in the [writer](https://docs.victoriametrics.com/anomaly-detection/components/writer/) section.

docs/anomaly-detection/UI.md

@@ -694,7 +694,7 @@ vmanomaly version: [v1.29.1](https://docs.victoriametrics.com/anomaly-detection/
 
 - BUGFIX: Now Visualization Panel correctly switches in between "query" and "detect" modes when respective buttons are hit in the [Visualization Panel](#visualization-panel), without showing stale results from the previous mode, once running anomaly detection task is explicitly cancelled (regression introduced in [v1.5.0](#v150)).
 
-- BUGFIX: Fixed an issue with [crypto.randomUUID](https://developer.mozilla.org/en-US/docs/Web/API/Crypto/randomUUID) introduced in [v1.29.0](#v1290) in [UI copilot](https://docs.victoriametrics.com/anomaly-detection/ui/#ai-assistance) that led to the front app showing a blank page.
+- BUGFIX: Fixed an issue with [crypto.randomUUID](https://developer.mozilla.org/en-US/docs/Web/API/Crypto/randomUUID) introduced in [v1.29.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1290) in [UI copilot](https://docs.victoriametrics.com/anomaly-detection/ui/#ai-assistance) that led to the front app showing a blank page.
 
 ### v1.5.0
 Released: 2026-03-05

docs/anomaly-detection/components/README.md

@@ -41,7 +41,7 @@ settings:
   restore_state: True  # restore state from previous run, if available
   retention:  # how long to keep stale models on disk/in memory
     ttl: "1d"  # time-to-live duration, if the model was not used for inference within this duration, it will be considered stale
-    check_every: "1h"  # how often to check for stale models and remove them
+    check_interval: "1h"  # how often to check for stale models and remove them
 
 # how and when to run the models is defined by schedulers
 # https://docs.victoriametrics.com/anomaly-detection/components/scheduler/
@@ -143,15 +143,15 @@ server:
 
 > This feature is better used in conjunction with [stateful service](https://docs.victoriametrics.com/anomaly-detection/components/settings/#state-restoration) to preserve the state of the models and schedulers between restarts and reuse what can be reused, thus avoiding unnecessary re-training of models, re-initialization of schedulers and re-reading of data.
 
-{{% available_from "v1.25.0" anomaly %}} Service supports hot reload of configuration files, which allows for automatic reloading of configurations on config files change filesystem events without the need of explicit service restart. This can be enabled via the `--watch` [CLI argument](https://docs.victoriametrics.com/anomaly-detection/quickstart/#command-line-arguments). `vmanomaly_hot_reload_enabled` flag in [self-monitoring metrics](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#startup-metrics) will be set to 1 (if enabled) or 0 (if disabled).
+{{% available_from "v1.25.0" anomaly %}} Service supports hot reload of configuration files, which allows for automatic reloading of configurations on config files change filesystem events without the need of explicit service restart. This can be enabled via the `--watch` [CLI argument](https://docs.victoriametrics.com/anomaly-detection/quickstart/#command-line-arguments). `vmanomaly_config_reload_enabled` flag in [self-monitoring metrics](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#startup-metrics) will be set to 1 (if enabled) or 0 (if disabled).
 
 ### How it works
 
-It works by watching for file system events, such as modifications, creations, or deletions of `.yml|.yaml` files in the specified directories. When a change is detected, the service will attempt to reload the configuration files, rebuild the [global config](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#global-config) and reinitialize the components. If the reload is successful, the `vmanomaly_hot_reload_events_total` metric will be incremented for `status="success"` label, otherwise it will be incremented with `status="failure"` label and a respective error message on config validation failure(s) will be logged.
+It works by watching for file system events, such as modifications, creations, or deletions of `.yml|.yaml` files in the specified directories. When a change is detected, the service will attempt to reload the configuration files, rebuild the [global config](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#global-configuration) and reinitialize the components. If the reload is successful, the `vmanomaly_config_reloads_total` metric will be incremented for `status="success"` label, otherwise it will be incremented with `status="failure"` label and a respective error message on config validation failure(s) will be logged.
 
 > If the reload fails, the service will log an error message indicating the reason for the failure, and the **previous configuration will remain active until a successful reload occurs** to preserve the service's stability. This means that if there are errors in the new configuration, the service will continue to operate with the last valid configuration until the issues are resolved.
 
-If used on [sharded setup](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#horizontal-scalability), upon [global config](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#global-config) change, all shards will be reinitialized with the new configurations.
+If used on [sharded setup](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#horizontal-scalability), upon [global config](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#global-configuration) change, all shards will be reinitialized with the new configurations.
 
 > Please note, that even if [state restoration](https://docs.victoriametrics.com/anomaly-detection/components/settings/#state-restoration) is enabled, the models, queries and schedulers might "migrate" to new shards if the order or the amount of [sub-configs](https://docs.victoriametrics.com/anomaly-detection/scaling-vmanomaly/#sub-configuration) changes after new config is hot-reloaded, so the state restoration won't be **fully** efficient in this case.
 
@@ -219,7 +219,7 @@ reader:
 # ... (rest of the config remains unchanged)
 ```
 
-After saving the changes, hot reload will automatically detect the changes in `config.yaml` and attempt to reload the configuration. As the changes are valid, the service will log a success message and increment the `vmanomaly_hot_reload_events_total` metric with `status="success"` label:
+After saving the changes, hot reload will automatically detect the changes in `config.yaml` and attempt to reload the configuration. As the changes are valid, the service will log a success message and increment the `vmanomaly_config_reloads_total` metric with `status="success"` label:
 
 - All the model instances of class `zscore_online`, that were trained on `host_network_receive_errors` can be reused as they are still valid and "fresh" for making inference on new datapoints until the next `fit_every` happens.
 - All the model instances of class `zscore_online`, that were trained on `cpu_seconds_total` will be re-trained with the new query expression and frequency, as old model instances are not valid anymore.

docs/anomaly-detection/components/models.md

@@ -41,7 +41,7 @@ models:
 # ...
 ```  
 
-Old-style configs (< [1.10.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#1100))
+Old-style configs (< [1.10.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1100))
 
 ```yaml
 model:
@@ -66,7 +66,7 @@ models:
 
 ## Common args
 
-From [1.10.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#1100), **common args**, supported by *every model (and model type)* were introduced.
+From [1.10.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1100), **common args**, supported by *every model (and model type)* were introduced.
 
 ### Queries
 

docs/anomaly-detection/components/reader.md

@@ -369,7 +369,7 @@ If True, then query will be performed from the last seen timestamp for a given s
 `1ms`
             </td>
             <td>
-It allows overriding the default `-search.latencyOffset`{{% available_from "v1.15.1" anomaly %}} [flag of VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/#list-of-command-line-flags) (30s). The default value is set to 1ms, which should help in cases where `sampling_frequency` is low (10-60s) and `sampling_frequency` equals `infer_every` in the [PeriodicScheduler](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/#periodic-scheduler). This prevents users from receiving `service - WARNING - [Scheduler [scheduler_alias]] No data available for inference.` warnings in logs and allows for consecutive `infer` calls without gaps. To restore the old behavior, set it equal to your `-search.latencyOffset` [flag value](https://docs.victoriametrics.com/victoriametrics/#list-of-command-line-flags).
+It allows overriding the default `-search.latencyOffset`{{% available_from "v1.15.1" anomaly %}} [flag of VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/#list-of-command-line-flags) (30s). The default value is set to 1ms, which should help in cases where `sampling_period` is low (10-60s) and `sampling_period` equals `infer_every` in the [PeriodicScheduler](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/#periodic-scheduler). This prevents users from receiving `service - WARNING - [Scheduler [scheduler_alias]] No data available for inference.` warnings in logs and allows for consecutive `infer` calls without gaps. To restore the old behavior, set it equal to your `-search.latencyOffset` [flag value](https://docs.victoriametrics.com/victoriametrics/#list-of-command-line-flags).
             </td>
         </tr>
         <tr>
@@ -760,7 +760,7 @@ Frequency of the points returned. Will be converted to `/select/stats_query_rang
 `10000`
             </td>
             <td>
-(Optional) For splitting long `fit_window` [queries](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vlogs-reader) into smaller sub-intervals. This helps users avoid hitting the timeout limits for individual queries by distributing initial query across multiple subquery requests with minimal overhead. Can be also set on [per-query](#per-query-parameters-1) basis to override reader-level settings.
+(Optional) For splitting long `fit_window` [queries](https://docs.victoriametrics.com/anomaly-detection/components/reader/#victorialogs-reader) into smaller sub-intervals. This helps users avoid hitting the timeout limits for individual queries by distributing initial query across multiple subquery requests with minimal overhead. Can be also set on [per-query](#per-query-parameters-1) basis to override reader-level settings.
             </td>
         </tr>
         <tr>

docs/guides/grafana-vmauth-openid-configuration/README.md

@@ -240,23 +240,23 @@ vmagent will write data into VictoriaMetrics single-node and cluster (with tenan
 # compose.yaml
 services:
   vmsingle:
-    image: victoriametrics/victoria-metrics:v1.144.0
+    image: victoriametrics/victoria-metrics:v1.145.0
 
   vmstorage:
-    image: victoriametrics/vmstorage:v1.144.0-cluster
+    image: victoriametrics/vmstorage:v1.145.0-cluster
 
   vminsert:
-    image: victoriametrics/vminsert:v1.144.0-cluster
+    image: victoriametrics/vminsert:v1.145.0-cluster
     command:
       - -storageNode=vmstorage:8400
 
   vmselect:
-    image: victoriametrics/vmselect:v1.144.0-cluster
+    image: victoriametrics/vmselect:v1.145.0-cluster
     command:
       - -storageNode=vmstorage:8401
 
   vmagent:
-    image: victoriametrics/vmagent:v1.144.0
+    image: victoriametrics/vmagent:v1.145.0
     volumes:
       - ./scrape.yaml:/etc/vmagent/config.yaml
     command:
@@ -308,7 +308,7 @@ Now add the vmauth service to `compose.yaml`:
 # compose.yaml
 services:
   vmauth:
-    image: docker.io/victoriametrics/vmauth:v1.144.0
+    image: docker.io/victoriametrics/vmauth:v1.145.0
     ports:
       - 8427:8427
     volumes:

docs/guides/vmagent-openid-configuration/README.md

@@ -155,15 +155,15 @@ These services will store and query the metrics scraped by vmagent.
 # compose.yaml
 services:
   vmstorage:
-    image: victoriametrics/vmstorage:v1.144.0-cluster
+    image: victoriametrics/vmstorage:v1.145.0-cluster
 
   vminsert:
-    image: victoriametrics/vminsert:v1.144.0-cluster
+    image: victoriametrics/vminsert:v1.145.0-cluster
     command:
       - -storageNode=vmstorage:8400
 
   vmselect:
-    image: victoriametrics/vmselect:v1.144.0-cluster
+    image: victoriametrics/vmselect:v1.145.0-cluster
     command:
       - -storageNode=vmstorage:8401
     ports:
@@ -196,7 +196,7 @@ Add the vmauth service to `compose.yaml`:
 # compose.yaml
 services:
   vmauth:
-    image: victoriametrics/vmauth:v1.144.0-enterprise
+    image: victoriametrics/vmauth:v1.145.0-enterprise
     ports:
       - 8427:8427
     volumes:
@@ -251,7 +251,7 @@ Add the vmagent service to `compose.yaml` with OAuth2 configuration:
 # compose.yaml
 services:
   vmagent:
-    image: victoriametrics/vmagent:v1.144.0
+    image: victoriametrics/vmagent:v1.145.0
     volumes:
       - ./scrape.yaml:/etc/vmagent/config.yaml
     command: