extract vmsingle-specific code into vmstorageSingleNode type

Signed-off-by: Artem Fetishev <rtm@victoriametrics.com>

Makefile

@@ -485,8 +485,8 @@ apptest-legacy: victoria-metrics-race vmbackup-race vmrestore-race
 		curl --output-dir /tmp -LO $${URL}/$${VMSINGLE} && tar xzf /tmp/$${VMSINGLE} -C $${DIR} && \
 		curl --output-dir /tmp -LO $${URL}/$${VMCLUSTER} && tar xzf /tmp/$${VMCLUSTER} -C $${DIR} \
 	); \
-	VM_LEGACY_VMSINGLE_PATH=$${DIR}/victoria-metrics-prod \
-	VM_LEGACY_VMSTORAGE_PATH=$${DIR}/vmstorage-prod \
+	VMSINGLE_V1_132_0_PATH=$${DIR}/victoria-metrics-prod \
+	VMSTORAGE_V1_132_0_PATH=$${DIR}/vmstorage-prod \
 	go test ./apptest/tests -run="^TestLegacySingle.*"
 
 benchmark:

app/victoria-metrics/main.go

@@ -22,7 +22,6 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/procutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promscrape"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/pushmetrics"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
 )
 
 var (
@@ -30,23 +29,26 @@ var (
 	useProxyProtocol = flagutil.NewArrayBool("httpListenAddr.useProxyProtocol", "Whether to use proxy protocol for connections accepted at the corresponding -httpListenAddr . "+
 		"See https://www.haproxy.org/download/1.8/doc/proxy-protocol.txt . "+
 		"With enabled proxy protocol http server cannot serve regular /metrics endpoint. Use -pushmetrics.url for metrics pushing")
-	minScrapeInterval = flag.Duration("dedup.minScrapeInterval", 0, "Leave only the last sample in every time series per each discrete interval "+
-		"equal to -dedup.minScrapeInterval > 0. See also -streamAggr.dedupInterval and https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#deduplication")
 	dryRun = flag.Bool("dryRun", false, "Whether to check config files without running VictoriaMetrics. The following config files are checked: "+
 		"-promscrape.config, -relabelConfig and -streamAggr.config. Unknown config entries aren't allowed in -promscrape.config by default. "+
 		"This can be changed with -promscrape.config.strictParse=false command-line flag")
-	inmemoryDataFlushInterval = flag.Duration("inmemoryDataFlushInterval", 5*time.Second, "The interval for guaranteed saving of in-memory data to disk. "+
-		"The saved data survives unclean shutdowns such as OOM crash, hardware reset, SIGKILL, etc. "+
-		"Bigger intervals may help increase the lifetime of flash storage with limited write cycles (e.g. Raspberry PI). "+
-		"Smaller intervals increase disk IO load. Minimum supported value is 1s")
 	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.")
-	finalDedupScheduleInterval = flag.Duration("storage.finalDedupScheduleCheckInterval", time.Hour, "The interval for checking when final deduplication process should be started."+
-		"Storage unconditionally adds 25% jitter to the interval value on each check evaluation."+
-		" Changing the interval to the bigger values may delay downsampling, deduplication for historical data."+
-		" See also https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#deduplication")
+	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")
 )
 
+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,
@@ -87,14 +89,8 @@ func main() {
 	}
 	logger.Infof("starting VictoriaMetrics at %q...", listenAddrs)
 	startTime := time.Now()
-	storage.SetDedupInterval(*minScrapeInterval)
-	storage.SetDataFlushInterval(*inmemoryDataFlushInterval)
-	if *finalDedupScheduleInterval < time.Hour {
-		logger.Fatalf("-dedup.finalDedupScheduleCheckInterval cannot be smaller than 1 hour; got %s", *finalDedupScheduleInterval)
-	}
-	storage.SetFinalDedupScheduleInterval(*finalDedupScheduleInterval)
-	vmstorage.Init(promql.ResetRollupResultCacheIfNeeded)
-	vmselect.Init()
+	vmstorage.Init(*maxConcurrentRequests, promql.ResetRollupResultCacheIfNeeded)
+	vmselect.Init(*maxConcurrentRequests, *maxQueueDuration)
 	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/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/config/config.go

@@ -113,15 +113,15 @@ func (g *Group) Validate(validateTplFn ValidateTplFn, validateExpressions bool)
 			// because correct types must be inherited after unmarshalling.
 			exprValidator := g.Type.ValidateExpr
 			if err := exprValidator(r.Expr); err != nil {
-				return fmt.Errorf("invalid expression for rule  %q: %w", ruleName, err)
+				return fmt.Errorf("invalid expression for rule %q: %w", ruleName, err)
 			}
 		}
 		if validateTplFn != nil {
 			if err := validateTplFn(r.Annotations); err != nil {
-				return fmt.Errorf("invalid annotations for rule  %q: %w", ruleName, err)
+				return fmt.Errorf("invalid annotations for rule %q: %w", ruleName, err)
 			}
 			if err := validateTplFn(r.Labels); err != nil {
-				return fmt.Errorf("invalid labels for rule  %q: %w", ruleName, err)
+				return fmt.Errorf("invalid labels for rule %q: %w", ruleName, err)
 			}
 		}
 	}

app/vmalert/config/config_test.go

@@ -121,7 +121,7 @@ func TestParse_Failure(t *testing.T) {
 	f([]string{"testdata/dir/rules2-bad.rules"}, "function \"unknown\" not defined")
 	f([]string{"testdata/dir/rules3-bad.rules"}, "either `record` or `alert` must be set")
 	f([]string{"testdata/dir/rules4-bad.rules"}, "either `record` or `alert` must be set")
-	f([]string{"testdata/rules/rules1-bad.rules"}, "bad graphite expr")
+	f([]string{"testdata/rules/rules1-bad.rules"}, "bad GraphiteQL expr")
 	f([]string{"testdata/rules/vlog-rules0-bad.rules"}, "bad LogsQL expr")
 	f([]string{"testdata/dir/rules6-bad.rules"}, "missing ':' in header")
 	f([]string{"testdata/rules/rules-multi-doc-bad.rules"}, "unknown fields")
@@ -283,7 +283,7 @@ func TestGroupValidate_Failure(t *testing.T) {
 				Expr:   "up | 0",
 			},
 		},
-	}, true, "bad prometheus expr")
+	}, true, "bad MetricsQL expr")
 
 	f(&Group{
 		Name: "test graphite expr",
@@ -293,7 +293,7 @@ func TestGroupValidate_Failure(t *testing.T) {
 				"description": "some-description",
 			}},
 		},
-	}, true, "bad graphite expr")
+	}, true, "bad GraphiteQL expr")
 
 	f(&Group{
 		Name: "test vlogs expr",
@@ -327,7 +327,7 @@ func TestGroupValidate_Failure(t *testing.T) {
 				Expr:   "sum(up == 0 ) by (host)",
 			},
 		},
-	}, true, "bad graphite expr")
+	}, true, "bad GraphiteQL expr")
 
 	f(&Group{
 		Name: "test vlogs with prometheus exp",
@@ -351,7 +351,7 @@ func TestGroupValidate_Failure(t *testing.T) {
 				For:    promutil.NewDuration(10 * time.Millisecond),
 			},
 		},
-	}, true, "bad prometheus expr")
+	}, true, "bad MetricsQL expr")
 }
 
 func TestGroupValidate_Success(t *testing.T) {

app/vmalert/config/types.go

@@ -66,11 +66,11 @@ func (t *Type) ValidateExpr(expr string) error {
 	switch t.String() {
 	case "graphite":
 		if _, err := graphiteql.Parse(expr); err != nil {
-			return fmt.Errorf("bad graphite expr: %q, err: %w", expr, err)
+			return fmt.Errorf("bad GraphiteQL expr: %q, err: %w", expr, err)
 		}
 	case "prometheus":
 		if _, err := metricsql.Parse(expr); err != nil {
-			return fmt.Errorf("bad prometheus expr: %q, err: %w", expr, err)
+			return fmt.Errorf("bad MetricsQL expr: %q, err: %w", expr, err)
 		}
 	case "vlogs":
 		q, err := logstorage.ParseStatsQuery(expr, 0)

app/vmalert/rule/group.go

@@ -95,7 +95,6 @@ type groupMetrics struct {
 	iterationTotal    *metrics.Counter
 	iterationDuration *metrics.Summary
 	iterationMissed   *metrics.Counter
-	iterationReset    *metrics.Counter
 	iterationInterval *metrics.Gauge
 }
 
@@ -331,7 +330,6 @@ 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
@@ -476,16 +474,14 @@ 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,
-				// Reset the evalTS to the current time.
+				// or the system wall clock was changed backward
+				missed = 0
 				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/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
@@ -225,7 +222,7 @@ func metricsFind(tr storage.TimeRange, label, qHead, qTail string, delimiter byt
 	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, 0, deadline)
 		if err != nil {
 			return nil, err
 		}
@@ -245,7 +242,7 @@ 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, 0, 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")
@@ -50,27 +43,22 @@ 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(maxConcurrentRequests int, maxQueueDuration 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)
+	concurrencyLimitCh = make(chan struct{}, maxConcurrentRequests)
 	initVMUIConfig()
 	initVMAlertProxy()
 
 	flagutil.RegisterSecretFlag("vmalert.proxyURL")
+
+	RequestHandler = func(w http.ResponseWriter, r *http.Request) bool {
+		return requestHandler(w, r, maxConcurrentRequests, maxQueueDuration)
+	}
 }
 
 // Stop stops vmselect
@@ -90,9 +78,6 @@ 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
@@ -100,8 +85,10 @@ var vmuiFiles embed.FS
 
 var vmuiFileServer = http.FileServer(http.FS(vmuiFiles))
 
-// RequestHandler handles remote read API requests
-func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
+var RequestHandler func(w http.ResponseWriter, r *http.Request) bool
+
+// requestHandler handles remote read API requests
+func requestHandler(w http.ResponseWriter, r *http.Request, maxConcurrentRequests int, maxQueueDuration time.Duration) bool {
 	path := strings.ReplaceAll(r.URL.Path, "//", "/")
 
 	// Strip /prometheus and /graphite prefixes in order to provide path compatibility with cluster version
@@ -131,12 +118,12 @@ 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 +139,7 @@ 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 +941,12 @@ 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)
@@ -1010,6 +956,7 @@ func ExportBlocks(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline sear
 		mustStop      atomic.Bool
 	)
 	var wg sync.WaitGroup
+	tr := sq.GetTimeRange()
 	for workerID := range gomaxprocs {
 		wg.Go(func() {
 			for xw := range workCh {
@@ -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/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")
@@ -853,7 +848,7 @@ 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 +959,7 @@ 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 +1295,6 @@ 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/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")
@@ -51,6 +43,12 @@ var (
 	retentionTimezoneOffset = flag.Duration("retentionTimezoneOffset", 0, "The offset for performing indexdb rotation. "+
 		"If set to 0, then the indexdb rotation is performed at 4am UTC time per each -retentionPeriod. "+
 		"If set to 2h, then the indexdb rotation is performed at 4am EET time (the timezone with +2h offset)")
+	minScrapeInterval = flag.Duration("dedup.minScrapeInterval", 0, "Leave only the last sample in every time series per each discrete interval "+
+		"equal to -dedup.minScrapeInterval > 0. See also -streamAggr.dedupInterval and https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#deduplication")
+	inmemoryDataFlushInterval = flag.Duration("inmemoryDataFlushInterval", 5*time.Second, "The interval for guaranteed saving of in-memory data to disk. "+
+		"The saved data survives unclean shutdowns such as OOM crash, hardware reset, SIGKILL, etc. "+
+		"Bigger intervals may help increase the lifetime of flash storage with limited write cycles (e.g. Raspberry PI). "+
+		"Smaller intervals increase disk IO load. Minimum supported value is 1s")
 
 	logNewSeries = flag.Bool("logNewSeries", false, "Whether to log new series. This option is for debug purposes only. It can lead to performance issues "+
 		"when big number of new series are ingested into VictoriaMetrics")
@@ -68,6 +66,11 @@ var (
 
 	minFreeDiskSpaceBytes = flagutil.NewBytes("storage.minFreeDiskSpaceBytes", 100e6, "The minimum free disk space at -storageDataPath after which the storage stops accepting new data")
 
+	finalDedupScheduleInterval = flag.Duration("storage.finalDedupScheduleCheckInterval", time.Hour, "The interval for checking when final deduplication process should be started."+
+		"Storage unconditionally adds 25% jitter to the interval value on each check evaluation."+
+		" Changing the interval to the bigger values may delay downsampling, deduplication for historical data."+
+		" See also https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#deduplication")
+
 	cacheSizeStorageTSID = flagutil.NewBytes("storage.cacheSizeStorageTSID", 0, "Overrides max size for storage/tsid cache. "+
 		"See https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#cache-tuning")
 	cacheSizeStorageMetricName = flagutil.NewBytes("storage.cacheSizeStorageMetricName", 0, "Overrides max size for storage/metricName cache. "+
@@ -106,16 +109,17 @@ 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)
-	}
-
-	resetResponseCacheIfNeeded = resetCacheIfNeeded
+func Init(maxConcurrentRequests int, resetCacheIfNeeded func(mrs []storage.MetricRow)) {
+	storage.SetDedupInterval(*minScrapeInterval)
+	storage.SetDataFlushInterval(*inmemoryDataFlushInterval)
 	storage.LegacySetRetentionTimezoneOffset(*retentionTimezoneOffset)
 	storage.SetFreeDiskSpaceLimit(minFreeDiskSpaceBytes.N)
 	storage.SetTSIDCacheSize(cacheSizeStorageTSID.IntN())
 	storage.SetTagFiltersCacheSize(cacheSizeIndexDBTagFilters.IntN())
+	if *finalDedupScheduleInterval < time.Hour {
+		logger.Fatalf("-storage.finalDedupScheduleCheckInterval cannot be smaller than 1 hour; got %s", *finalDedupScheduleInterval)
+	}
+	storage.SetFinalDedupScheduleInterval(*finalDedupScheduleInterval)
 	storage.SetMetricNamesStatsCacheSize(cacheSizeMetricNamesStats.IntN())
 	storage.SetMetricNameCacheSize(cacheSizeStorageMetricName.IntN())
 	storage.SetMetadataStorageSize(metadataStorageSize.IntN())
@@ -134,9 +138,9 @@ func Init(resetCacheIfNeeded func(mrs []storage.MetricRow)) {
 	if *idbPrefillStart > 23*time.Hour {
 		logger.Panicf("-storage.idbPrefillStart cannot exceed 23 hours; got %s", idbPrefillStart)
 	}
+	fs.RegisterPathFsMetrics(*storageDataPath)
 	logger.Infof("opening storage at %q with -retentionPeriod=%s", *storageDataPath, retentionPeriod)
 	startTime := time.Now()
-	WG = syncwg.WaitGroup{}
 	opts := storage.OpenOptions{
 		Retention:                   retentionPeriod.Duration(),
 		FutureRetention:             futureRetention.Duration(),
@@ -149,8 +153,7 @@ func Init(resetCacheIfNeeded func(mrs []storage.MetricRow)) {
 		LogNewSeries:                *logNewSeries,
 	}
 	strg := storage.MustOpenStorage(*storageDataPath, opts)
-	Storage = strg
-	initStaleSnapshotsRemover(strg)
+	vmStorage = newVMStorageSingleNode(strg, maxConcurrentRequests, resetCacheIfNeeded)
 
 	var m storage.Metrics
 	strg.UpdateMetrics(&m)
@@ -165,147 +168,41 @@ func Init(resetCacheIfNeeded func(mrs []storage.MetricRow)) {
 	// register storage metrics
 	storageMetrics = metrics.NewSet()
 	storageMetrics.RegisterMetricsWriter(func(w io.Writer) {
-		writeStorageMetrics(w, strg)
+		vmStorage.writeStorageMetrics(w)
 	})
 	metrics.RegisterSet(storageMetrics)
-	fs.RegisterPathFsMetrics(*storageDataPath)
+
+	VMInsertAPI = vmStorage
+	VMSelectAPI = vmStorage
+	GetSearch = func(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (*storage.Search, int, error) {
+		return vmStorage.GetSearch(qt, sq, deadline)
+	}
+	PutSearch = func(sr *storage.Search) {
+		vmStorage.PutSearch(sr)
+	}
+	RequestHandler = func(w http.ResponseWriter, r *http.Request) bool {
+		return vmStorage.requestHandler(w, r)
+	}
+	DebugFlush = func() {
+		vmStorage.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() {
@@ -315,17 +212,22 @@ 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 (api *VMStorageSingleNode) requestHandler(w http.ResponseWriter, r *http.Request) bool {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.requestHandler(w, r)
+}
+
+// requestHandler is a storage request handler.
+// TODO(@rtm0): Move to a separate file, request_handler.go
+func (api *VMStorage) requestHandler(w http.ResponseWriter, r *http.Request) bool {
 	path := r.URL.Path
 	if path == "/internal/force_merge" {
 		if !httpserver.CheckAuthFlag(w, r, forceMergeAuthKey) {
@@ -338,8 +240,9 @@ 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 := api.s.ForceMergePartitions(partitionNamePrefix); err != nil {
 				logger.Errorf("error in forced merge for partition_prefix=%q: %s", partitionNamePrefix, err)
+				return
 			}
 			logger.Infof("forced merge for partition_prefix=%q has been successfully finished in %.3f seconds", partitionNamePrefix, time.Since(startTime).Seconds())
 		}()
@@ -350,9 +253,10 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 			return true
 		}
 		logger.Infof("flushing storage to make pending data available for reading")
-		Storage.DebugFlush()
+		api.s.DebugFlush()
 		return true
 	}
+
 	if path == "/internal/log_new_series" {
 		if !httpserver.CheckAuthFlag(w, r, logNewSeriesAuthKey) {
 			return true
@@ -369,7 +273,7 @@ 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)
+		api.s.SetLogNewSeriesUntil(endTime)
 		fmt.Fprintf(w, `{"status":"success","data":{"logEndTime":%q}}`, time.Unix(int64(endTime), 0))
 		return true
 	}
@@ -391,13 +295,13 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	case "/create":
 		snapshotsCreateTotal.Inc()
 		w.Header().Set("Content-Type", "application/json")
-		snapshotName := Storage.MustCreateSnapshot()
+		snapshotName := api.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 := api.deleteSnapshot(snapshotName); err != nil {
 				logger.Infof("cannot delete just created snapshot: %s", err)
 				return true
 			}
@@ -413,7 +317,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 	case "/list":
 		snapshotsListTotal.Inc()
 		w.Header().Set("Content-Type", "application/json")
-		snapshots := Storage.MustListSnapshots()
+		snapshots := api.s.MustListSnapshots()
 		fmt.Fprintf(w, `{"status":"ok","snapshots":[`)
 		if len(snapshots) > 0 {
 			for _, snapshot := range snapshots[:len(snapshots)-1] {
@@ -427,7 +331,7 @@ 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 := api.deleteSnapshot(snapshotName); err != nil {
 			jsonResponseError(w, err)
 			snapshotsDeleteErrorsTotal.Inc()
 			return true
@@ -437,9 +341,10 @@ 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 := api.s.MustListSnapshots()
 		for _, snapshotName := range snapshots {
-			if err := Storage.DeleteSnapshot(snapshotName); err != nil {
+			// TODO(@rtm0): Use VMStorage.deleteSnapshot()?
+			if err := api.s.DeleteSnapshot(snapshotName); err != nil {
 				err = fmt.Errorf("cannot delete snapshot %q: %w", snapshotName, err)
 				jsonResponseError(w, err)
 				snapshotsDeleteAllErrorsTotal.Inc()
@@ -453,50 +358,6 @@ 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")
 
@@ -511,7 +372,9 @@ 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 (api *VMStorage) writeStorageMetrics(w io.Writer) {
+	strg := api.s
 	var m storage.Metrics
 	strg.UpdateMetrics(&m)
 	tm := &m.TableMetrics
@@ -735,6 +598,8 @@ 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(api.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 +1,396 @@
+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, maxConcurrentRequests int) *VMStorage {
+	if err := encoding.CheckPrecisionBits(uint8(*precisionBits)); err != nil {
+		logger.Fatalf("invalid -precisionBits: %d", err)
+	}
+
+	maxUniqueTimeseriesCalculated := *maxUniqueTimeseries
+	if maxUniqueTimeseriesCalculated <= 0 {
+		maxUniqueTimeseriesCalculated = calculateMaxUniqueTimeseries(maxConcurrentRequests, 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 (api *VMStorage) initStaleSnapshotsRemover() {
+	if snapshotsMaxAge.Duration() <= 0 {
+		return
+	}
+	snapshotsMaxAgeDuration := snapshotsMaxAge.Duration()
+	api.staleSnapshotsRemoverWG.Go(func() {
+		d := timeutil.AddJitterToDuration(time.Second * 11)
+		t := time.NewTicker(d)
+		defer t.Stop()
+		for {
+			select {
+			case <-api.staleSnapshotsRemoverCh:
+				return
+			case <-t.C:
+			}
+			api.s.MustDeleteStaleSnapshots(snapshotsMaxAgeDuration)
+		}
+	})
+}
+
+func (api *VMStorage) Stop() {
+	close(api.staleSnapshotsRemoverCh)
+	api.staleSnapshotsRemoverWG.Wait()
+	api.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 (api *VMStorage) WriteRows(rows []storage.MetricRow) error {
+	api.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 (api *VMStorage) WriteMetadata(rows []metricsmetadata.Row) error {
+	api.s.AddMetadataRows(rows)
+	return nil
+}
+
+// IsReadOnly returns true is the storage is in read-only mode.
+func (api *VMStorage) IsReadOnly() bool {
+	return api.s.IsReadOnly()
+}
+
+func (api *VMStorage) InitSearch(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (vmselectapi.BlockIterator, error) {
+	tr := sq.GetTimeRange()
+	maxMetrics := api.getMaxMetrics(sq.MaxMetrics)
+	tfss, err := api.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, api.s, tfss, tr, maxMetrics, deadline)
+	if err := bi.sr.Error(); err != nil {
+		bi.MustClose()
+		return nil, err
+	}
+	return bi, nil
+}
+
+func (api *VMStorage) getMaxMetrics(searchQueryLimit int) int {
+	if searchQueryLimit <= 0 {
+		return api.maxUniqueTimeSeriesCalculated
+	}
+	// searchQueryLimit cannot exceed `-search.maxUniqueTimeseries`
+	if api.maxUniqueTimeseries != 0 && searchQueryLimit > api.maxUniqueTimeseries {
+		searchQueryLimit = api.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 (api *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 = api.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := api.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	if len(tfss) == 0 {
+		return nil, fmt.Errorf("missing tag filters")
+	}
+	return api.s.SearchMetricNames(qt, tfss, tr, maxMetrics, deadline)
+}
+
+// SearchLabelValues searches for label values for the given labelName, tfss and
+// tr.
+func (api *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 = api.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := api.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	return api.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 (api *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 := api.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 (api *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 = api.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := api.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	return api.s.SearchLabelNames(qt, tfss, tr, maxLabelNames, maxMetrics, deadline)
+}
+
+func (api *VMStorage) SeriesCount(_ *querytracer.Tracer, _, _ uint32, deadline uint64) (uint64, error) {
+	return api.s.GetSeriesCount(deadline)
+}
+
+func (api *VMStorage) Tenants(_ *querytracer.Tracer, _ storage.TimeRange, _ uint64) ([]string, error) {
+	return nil, nil
+}
+
+// GetTSDBStatus returns TSDB status for given filters on the given date.
+func (api *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 = api.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := api.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, err
+	}
+	date := uint64(sq.MinTimestamp) / (24 * 3600 * 1000)
+	return api.s.GetTSDBStatus(qt, tfss, date, focusLabel, topN, maxMetrics, deadline)
+}
+
+// DeleteSeries deletes series matching tfss.
+//
+// Returns the number of deleted series.
+func (api *VMStorage) DeleteSeries(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (int, error) {
+	// TODO(@rtm0): Return an error if the storage is in read-only mode?
+
+	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 = api.maxUniqueTimeSeriesCalculated
+	}
+	tfss, err := api.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return 0, err
+	}
+	if len(tfss) == 0 {
+		return 0, fmt.Errorf("missing tag filters")
+	}
+	return api.s.DeleteSeries(qt, tfss, maxMetrics)
+}
+
+func (api *VMStorage) RegisterMetricNames(qt *querytracer.Tracer, mrs []storage.MetricRow, _ uint64) error {
+	// TODO(@rtm0): Return an error if the storage is in read-only mode?
+
+	api.s.RegisterMetricNames(qt, mrs)
+	return nil
+}
+
+// GetMetricNamesUsageStats returns metric name usage stats.
+func (api *VMStorage) GetMetricNamesUsageStats(qt *querytracer.Tracer, _ *storage.TenantToken, limit, le int, matchPattern string, _ uint64) (metricnamestats.StatsResult, error) {
+
+	return api.s.GetMetricNamesStats(qt, limit, le, matchPattern), nil
+}
+
+// ResetMetricNamesStats resets state for metric names usage tracker
+func (api *VMStorage) ResetMetricNamesUsageStats(qt *querytracer.Tracer, _ uint64) error {
+	api.s.ResetMetricNamesStats(qt)
+	return nil
+}
+
+func (api *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 := api.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 (api *VMStorage) GetMetadataRecords(qt *querytracer.Tracer, _ *storage.TenantToken, limit int, metricName string, _ uint64) ([]*metricsmetadata.Row, error) {
+	return api.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 (api *VMStorage) deleteSnapshot(snapshotName string) error {
+	snapshots := api.s.MustListSnapshots()
+	for _, snName := range snapshots {
+		if snName == snapshotName {
+			if err := api.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 +1,202 @@
+package vmstorage
+
+import (
+	"errors"
+	"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"
+)
+
+// 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{
+		VMStorage:          vms,
+		wg:                 syncwg.WaitGroup{},
+		resetCacheIfNeeded: resetCacheIfNeeded,
+	}
+}
+
+type VMStorageSingleNode struct {
+	*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 (api *VMStorageSingleNode) Stop() {
+	api.wg.WaitAndBlock()
+	api.VMStorage.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 (api *VMStorageSingleNode) WriteRows(rows []storage.MetricRow) error {
+	api.wg.Add(1)
+	defer api.wg.Done()
+
+	if api.s.IsReadOnly() {
+		return errReadOnly
+	}
+	api.resetCacheIfNeeded(rows)
+	return api.VMStorage.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 (api *VMStorageSingleNode) WriteMetadata(rows []metricsmetadata.Row) error {
+	api.wg.Add(1)
+	defer api.wg.Done()
+
+	if api.s.IsReadOnly() {
+		return errReadOnly
+	}
+	return api.VMStorage.WriteMetadata(rows)
+}
+
+var errReadOnly = errors.New("the storage is in read-only mode; check -storage.minFreeDiskSpaceBytes command-line flag value")
+
+func (api *VMStorageSingleNode) IsReadOnly() bool {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.IsReadOnly()
+}
+
+// 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 (api *VMStorageSingleNode) GetSearch(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (*storage.Search, int, error) {
+	api.wg.Add(1)
+
+	tr := sq.GetTimeRange()
+	maxMetrics := api.getMaxMetrics(sq.MaxMetrics)
+	tfss, err := api.setupTfss(qt, sq, tr, maxMetrics, deadline)
+	if err != nil {
+		return nil, 0, err
+	}
+
+	sr := getSearch()
+	maxSeriesCount := sr.Init(qt, api.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 (api *VMStorageSingleNode) PutSearch(sr *storage.Search) {
+	api.wg.Done()
+	putSearch(sr)
+}
+
+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 (api *VMStorageSingleNode) SearchMetricNames(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) ([]string, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.SearchMetricNames(qt, sq, deadline)
+}
+
+func (api *VMStorageSingleNode) LabelValues(qt *querytracer.Tracer, sq *storage.SearchQuery, labelName string, maxLabelValues int, deadline uint64) ([]string, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.LabelValues(qt, sq, labelName, maxLabelValues, deadline)
+}
+
+func (api *VMStorageSingleNode) TagValueSuffixes(qt *querytracer.Tracer, accountID, projectID uint32, tr storage.TimeRange, tagKey, tagValuePrefix string, delimiter byte, maxSuffixes int, deadline uint64) ([]string, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.TagValueSuffixes(qt, accountID, projectID, tr, tagKey, tagValuePrefix, delimiter, maxSuffixes, deadline)
+}
+
+func (api *VMStorageSingleNode) LabelNames(qt *querytracer.Tracer, sq *storage.SearchQuery, maxLabelNames int, deadline uint64) ([]string, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.LabelNames(qt, sq, maxLabelNames, deadline)
+}
+
+func (api *VMStorageSingleNode) SeriesCount(qt *querytracer.Tracer, accountID, projectID uint32, deadline uint64) (uint64, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.SeriesCount(qt, accountID, projectID, deadline)
+}
+
+func (api *VMStorageSingleNode) TSDBStatus(qt *querytracer.Tracer, sq *storage.SearchQuery, focusLabel string, topN int, deadline uint64) (*storage.TSDBStatus, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.TSDBStatus(qt, sq, focusLabel, topN, deadline)
+}
+
+func (api *VMStorageSingleNode) DeleteSeries(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (int, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	// TODO(@rtm0): Return an error if the storage is in read-only mode?
+	return api.VMStorage.DeleteSeries(qt, sq, deadline)
+}
+
+func (api *VMStorageSingleNode) RegisterMetricNames(qt *querytracer.Tracer, mrs []storage.MetricRow, deadline uint64) error {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	// TODO(@rtm0): Return an error if the storage is in read-only mode?
+	return api.VMStorage.RegisterMetricNames(qt, mrs, deadline)
+}
+
+func (api *VMStorageSingleNode) GetMetricNamesUsageStats(qt *querytracer.Tracer, tt *storage.TenantToken, limit, le int, matchPattern string, deadline uint64) (metricnamestats.StatsResult, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.GetMetricNamesUsageStats(qt, tt, limit, le, matchPattern, deadline)
+}
+
+func (api *VMStorageSingleNode) ResetMetricNamesUsageStats(qt *querytracer.Tracer, deadline uint64) error {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.ResetMetricNamesUsageStats(qt, deadline)
+}
+
+func (api *VMStorageSingleNode) GetMetadataRecords(qt *querytracer.Tracer, tt *storage.TenantToken, limit int, metricName string, deadline uint64) ([]*metricsmetadata.Row, error) {
+	api.wg.Add(1)
+	defer api.wg.Done()
+	return api.VMStorage.GetMetadataRecords(qt, tt, limit, metricName, deadline)
+}

apptest/app.go

@@ -22,6 +22,7 @@ var (
 	vminsertAddrRE              = regexp.MustCompile(`accepting vminsert conns at (.*:\d{1,5})$`)
 	vminsertClusterNativeAddrRE = regexp.MustCompile(`started TCP clusternative server at "(.*:\d{1,5})"`)
 	vmselectAddrRE              = regexp.MustCompile(`accepting vmselect conns at (.*:\d{1,5})$`)
+	vmauthHttpListenAddrRE      = regexp.MustCompile(`pprof handlers are exposed at http://(.*:\d{1,5})/debug/pprof/`)
 )
 
 // app represents an instance of some VictoriaMetrics server (such as vmstorage,

apptest/testcase.go

@@ -88,19 +88,11 @@ func (tc *TestCase) MustStartDefaultVmsingle() *Vmsingle {
 }
 
 // MustStartVmsingle is a test helper function that starts an instance of
-// vmsingle located at ../../bin/victoria-metrics-race and fails the test if the app
-// fails to start.
+// vmsingle (latest version) and fails the test if the app fails to start.
 func (tc *TestCase) MustStartVmsingle(instance string, flags []string) *Vmsingle {
 	tc.t.Helper()
-	return tc.MustStartVmsingleAt(instance, "../../bin/victoria-metrics-race", flags)
-}
 
-// MustStartVmsingleAt is a test helper function that starts an instance of
-// vmsingle and fails the test if the app fails to start.
-func (tc *TestCase) MustStartVmsingleAt(instance, binary string, flags []string) *Vmsingle {
-	tc.t.Helper()
-
-	app, err := StartVmsingleAt(instance, binary, flags, tc.cli, tc.output)
+	app, err := StartVmsingle(instance, flags, tc.cli, tc.output)
 	if err != nil {
 		tc.t.Fatalf("Could not start %s: %v", instance, err)
 	}
@@ -109,19 +101,11 @@ func (tc *TestCase) MustStartVmsingleAt(instance, binary string, flags []string)
 }
 
 // MustStartVmstorage is a test helper function that starts an instance of
-// vmstorage located at ../../bin/vmstorage-race and fails the test if the app fails
-// to start.
+// vmstorage (latest version) and fails the test if the app fails to start.
 func (tc *TestCase) MustStartVmstorage(instance string, flags []string) *Vmstorage {
 	tc.t.Helper()
-	return tc.MustStartVmstorageAt(instance, "../../bin/vmstorage-race", flags)
-}
 
-// MustStartVmstorageAt is a test helper function that starts an instance of
-// vmstorage and fails the test if the app fails to start.
-func (tc *TestCase) MustStartVmstorageAt(instance string, binary string, flags []string) *Vmstorage {
-	tc.t.Helper()
-
-	app, err := StartVmstorageAt(instance, binary, flags, tc.cli, tc.output)
+	app, err := StartVmstorage(instance, flags, tc.cli, tc.output)
 	if err != nil {
 		tc.t.Fatalf("Could not start %s: %v", instance, err)
 	}
@@ -130,7 +114,7 @@ func (tc *TestCase) MustStartVmstorageAt(instance string, binary string, flags [
 }
 
 // MustStartVmselect is a test helper function that starts an instance of
-// vmselect and fails the test if the app fails to start.
+// vmselect (latest version) and fails the test if the app fails to start.
 func (tc *TestCase) MustStartVmselect(instance string, flags []string) *Vmselect {
 	tc.t.Helper()
 
@@ -290,10 +274,8 @@ func (tc *TestCase) MustStartDefaultCluster() *Vmcluster {
 // tests usually come paired with corresponding vmsingle tests.
 type ClusterOptions struct {
 	Vmstorage1Instance string
-	Vmstorage1Binary   string
 	Vmstorage1Flags    []string
 	Vmstorage2Instance string
-	Vmstorage2Binary   string
 	Vmstorage2Flags    []string
 	VminsertInstance   string
 	VminsertFlags      []string
@@ -305,15 +287,8 @@ type ClusterOptions struct {
 func (tc *TestCase) MustStartCluster(opts *ClusterOptions) *Vmcluster {
 	tc.t.Helper()
 
-	if opts.Vmstorage1Binary == "" {
-		opts.Vmstorage1Binary = "../../bin/vmstorage-race"
-	}
-	vmstorage1 := tc.MustStartVmstorageAt(opts.Vmstorage1Instance, opts.Vmstorage1Binary, opts.Vmstorage1Flags)
-
-	if opts.Vmstorage2Binary == "" {
-		opts.Vmstorage2Binary = "../../bin/vmstorage-race"
-	}
-	vmstorage2 := tc.MustStartVmstorageAt(opts.Vmstorage2Instance, opts.Vmstorage2Binary, opts.Vmstorage2Flags)
+	vmstorage1 := tc.MustStartVmstorage(opts.Vmstorage1Instance, opts.Vmstorage1Flags)
+	vmstorage2 := tc.MustStartVmstorage(opts.Vmstorage2Instance, opts.Vmstorage2Flags)
 
 	opts.VminsertFlags = append(opts.VminsertFlags, []string{
 		"-storageNode=" + vmstorage1.VminsertAddr() + "," + vmstorage2.VminsertAddr(),

apptest/testcase_legacy.go

@@ -0,0 +1,50 @@
+package apptest
+
+// MustStartVmsingle_v1_132_0 is a test helper function that starts an instance
+// of vmsingle-v1.132.0 (last version that uses legacy index) and fails the test
+// if the app fails to start.
+func (tc *TestCase) MustStartVmsingle_v1_132_0(instance string, flags []string) *Vmsingle {
+	tc.t.Helper()
+
+	app, err := StartVmsingle_v1_132_0(instance, flags, tc.cli, tc.output)
+	if err != nil {
+		tc.t.Fatalf("Could not start %s: %v", instance, err)
+	}
+	tc.addApp(instance, app)
+	return app
+}
+
+// MustStartVmstorage_v1_132_0 is a test helper function that starts an instance
+// of vmstorage-v1.132.0 (last version that uses legacy index)  and fails the
+// test if the app fails to start.
+func (tc *TestCase) MustStartVmstorage_v1_132_0(instance string, flags []string) *Vmstorage {
+	tc.t.Helper()
+
+	app, err := StartVmstorage_v1_132_0(instance, flags, tc.cli, tc.output)
+	if err != nil {
+		tc.t.Fatalf("Could not start %s: %v", instance, err)
+	}
+	tc.addApp(instance, app)
+	return app
+}
+
+// MustStartCluster_v1_132_0 starts a cluster with vmstorage-v1.132.0 with
+// custom flags.
+func (tc *TestCase) MustStartCluster_v1_132_0(opts *ClusterOptions) *Vmcluster {
+	tc.t.Helper()
+
+	vmstorage1 := tc.MustStartVmstorage_v1_132_0(opts.Vmstorage1Instance, opts.Vmstorage1Flags)
+	vmstorage2 := tc.MustStartVmstorage_v1_132_0(opts.Vmstorage2Instance, opts.Vmstorage2Flags)
+
+	opts.VminsertFlags = append(opts.VminsertFlags, []string{
+		"-storageNode=" + vmstorage1.VminsertAddr() + "," + vmstorage2.VminsertAddr(),
+	}...)
+	vminsert := tc.MustStartVminsert(opts.VminsertInstance, opts.VminsertFlags)
+
+	opts.VmselectFlags = append(opts.VmselectFlags, []string{
+		"-storageNode=" + vmstorage1.VmselectAddr() + "," + vmstorage2.VmselectAddr(),
+	}...)
+	vmselect := tc.MustStartVmselect(opts.VmselectInstance, opts.VmselectFlags)
+
+	return &Vmcluster{vminsert, vmselect, []*Vmstorage{vmstorage1, vmstorage2}}
+}

apptest/tests/legacy_indexdb_test.go

@@ -2,7 +2,6 @@ package tests
 
 import (
 	"fmt"
-	"os"
 	"path/filepath"
 	"slices"
 	"testing"
@@ -11,11 +10,6 @@ import (
 	at "github.com/VictoriaMetrics/VictoriaMetrics/apptest"
 )
 
-var (
-	legacyVmsinglePath  = os.Getenv("VM_LEGACY_VMSINGLE_PATH")
-	legacyVmstoragePath = os.Getenv("VM_LEGACY_VMSTORAGE_PATH")
-)
-
 type testLegacyDeleteSeriesOpts struct {
 	startLegacySUT func() at.PrometheusWriteQuerier
 	startNewSUT    func() at.PrometheusWriteQuerier
@@ -31,7 +25,7 @@ func TestLegacySingleDeleteSeries(t *testing.T) {
 
 	opts := testLegacyDeleteSeriesOpts{
 		startLegacySUT: func() at.PrometheusWriteQuerier {
-			return tc.MustStartVmsingleAt("vmsingle-legacy", legacyVmsinglePath, []string{
+			return tc.MustStartVmsingle_v1_132_0("vmsingle-legacy", []string{
 				"-storageDataPath=" + storageDataPath,
 				"-retentionPeriod=100y",
 				"-search.maxStalenessInterval=1m",
@@ -64,15 +58,13 @@ func TestLegacyClusterDeleteSeries(t *testing.T) {
 
 	opts := testLegacyDeleteSeriesOpts{
 		startLegacySUT: func() at.PrometheusWriteQuerier {
-			return tc.MustStartCluster(&at.ClusterOptions{
+			return tc.MustStartCluster_v1_132_0(&at.ClusterOptions{
 				Vmstorage1Instance: "vmstorage1-legacy",
-				Vmstorage1Binary:   legacyVmstoragePath,
 				Vmstorage1Flags: []string{
 					"-storageDataPath=" + storage1DataPath,
 					"-retentionPeriod=100y",
 				},
 				Vmstorage2Instance: "vmstorage2-legacy",
-				Vmstorage2Binary:   legacyVmstoragePath,
 				Vmstorage2Flags: []string{
 					"-storageDataPath=" + storage2DataPath,
 					"-retentionPeriod=100y",
@@ -255,7 +247,7 @@ func TestLegacySingleBackupRestore(t *testing.T) {
 
 	opts := testLegacyBackupRestoreOpts{
 		startLegacySUT: func() at.PrometheusWriteQuerier {
-			return tc.MustStartVmsingleAt("vmsingle-legacy", legacyVmsinglePath, []string{
+			return tc.MustStartVmsingle_v1_132_0("vmsingle-legacy", []string{
 				"-storageDataPath=" + storageDataPath,
 				"-retentionPeriod=100y",
 				"-search.disableCache=true",
@@ -298,15 +290,13 @@ func TestLegacyClusterBackupRestore(t *testing.T) {
 
 	opts := testLegacyBackupRestoreOpts{
 		startLegacySUT: func() at.PrometheusWriteQuerier {
-			return tc.MustStartCluster(&at.ClusterOptions{
+			return tc.MustStartCluster_v1_132_0(&at.ClusterOptions{
 				Vmstorage1Instance: "vmstorage1-legacy",
-				Vmstorage1Binary:   legacyVmstoragePath,
 				Vmstorage1Flags: []string{
 					"-storageDataPath=" + storage1DataPath,
 					"-retentionPeriod=100y",
 				},
 				Vmstorage2Instance: "vmstorage2-legacy",
-				Vmstorage2Binary:   legacyVmstoragePath,
 				Vmstorage2Flags: []string{
 					"-storageDataPath=" + storage2DataPath,
 					"-retentionPeriod=100y",
@@ -583,7 +573,7 @@ func TestLegacySingleDowngrade(t *testing.T) {
 
 	opts := testLegacyDowngradeOpts{
 		startLegacySUT: func() at.PrometheusWriteQuerier {
-			return tc.MustStartVmsingleAt("vmsingle-legacy", legacyVmsinglePath, []string{
+			return tc.MustStartVmsingle_v1_132_0("vmsingle-legacy", []string{
 				"-storageDataPath=" + storageDataPath,
 				"-retentionPeriod=100y",
 				"-search.disableCache=true",
@@ -618,15 +608,13 @@ func TestLegacyClusterDowngrade(t *testing.T) {
 
 	opts := testLegacyDowngradeOpts{
 		startLegacySUT: func() at.PrometheusWriteQuerier {
-			return tc.MustStartCluster(&at.ClusterOptions{
+			return tc.MustStartCluster_v1_132_0(&at.ClusterOptions{
 				Vmstorage1Instance: "vmstorage1-legacy",
-				Vmstorage1Binary:   legacyVmstoragePath,
 				Vmstorage1Flags: []string{
 					"-storageDataPath=" + storage1DataPath,
 					"-retentionPeriod=100y",
 				},
 				Vmstorage2Instance: "vmstorage2-legacy",
-				Vmstorage2Binary:   legacyVmstoragePath,
 				Vmstorage2Flags: []string{
 					"-storageDataPath=" + storage2DataPath,
 					"-retentionPeriod=100y",

apptest/vmagent.go

@@ -16,43 +16,63 @@ import (
 	"github.com/golang/snappy"
 )
 
-// Vmagent holds the state of a vmagent app and provides vmagent-specific functions
-type Vmagent struct {
-	*app
-	*metricsClient
-
-	httpListenAddr string
-
-	cli *Client
-}
-
-// StartVmagent starts an instance of vmagent with the given flags. It also
-// sets the default flags and populates the app instance state with runtime
-// values extracted from the application log (such as httpListenAddr)
+// StartVmagent starts the latest version of vmagent.
+//
+// The path to the binary can be provided via VMAGENT_PATH environment
+// variable. If the variable is not set, ../../bin/vmagent-race will be
+// used.
 func StartVmagent(instance string, flags []string, cli *Client, promScrapeConfigFilePath string, output io.Writer) (*Vmagent, error) {
-	extractREs := []*regexp.Regexp{
-		httpListenAddrRE,
+	binary := os.Getenv("VMAGENT_PATH")
+	if binary == "" {
+		binary = "../../bin/vmagent-race"
 	}
-
-	app, stderrExtracts, err := startApp(instance, "../../bin/vmagent-race", flags, &appOptions{
+	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
 		defaultFlags: map[string]string{
 			"-httpListenAddr":          "127.0.0.1:0",
 			"-promscrape.config":       promScrapeConfigFilePath,
 			"-remoteWrite.tmpDataPath": fmt.Sprintf("%s/%s-%d", os.TempDir(), instance, time.Now().UnixNano()),
 		},
-		extractREs: extractREs,
-		output:     output,
+		extractREs: []*regexp.Regexp{
+			httpListenAddrRE,
+		},
+		output: output,
 	})
 	if err != nil {
 		return nil, err
 	}
 
+	return newVmagent(app, cli, vmagentRuntimeValues{
+		httpListenAddr: stderrExtracts[0],
+	}), nil
+}
+
+type vmagentRuntimeValues struct {
+	httpListenAddr string
+}
+
+func newVmagent(app *app, cli *Client, rt vmagentRuntimeValues) *Vmagent {
 	return &Vmagent{
 		app:            app,
-		metricsClient:  newMetricsClient(cli, stderrExtracts[0]),
-		httpListenAddr: stderrExtracts[0],
 		cli:            cli,
-	}, nil
+		metricsClient:  newMetricsClient(cli, rt.httpListenAddr),
+		httpListenAddr: rt.httpListenAddr,
+	}
+}
+
+// Vmagent holds the state of a vmagent app and provides vmagent-specific
+// functions.
+type Vmagent struct {
+	*app
+	*metricsClient
+
+	cli            *Client
+	httpListenAddr string
+}
+
+// HTTPAddr returns the address at which the vmagent process is listening
+// for http connections.
+func (app *Vmagent) HTTPAddr() string {
+	return app.httpListenAddr
 }
 
 // APIV1ImportPrometheus is a test helper function that inserts a
@@ -203,12 +223,6 @@ func (app *Vmagent) PrometheusAPIV1Write(t *testing.T, wr prompb.WriteRequest, o
 	})
 }
 
-// HTTPAddr returns the address at which the vmagent process is listening
-// for http connections.
-func (app *Vmagent) HTTPAddr() string {
-	return app.httpListenAddr
-}
-
 // sendBlocking sends the data to vmstorage by executing `send` function and
 // waits until the data is actually sent.
 //

apptest/vmauth.go

@@ -2,6 +2,7 @@ package apptest
 
 import (
 	"io"
+	"os"
 	"regexp"
 	"syscall"
 	"testing"
@@ -10,7 +11,48 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
 )
 
-var httpBuilitinListenAddrRE = regexp.MustCompile(`pprof handlers are exposed at http://(.*:\d{1,5})/debug/pprof/`)
+// StartVmauth starts the latest version of vmauth.
+//
+// The path to the binary can be provided via VMAUTH_PATH environment
+// variable. If the variable is not set, ../../bin/vmauth-race will be
+// used.
+func StartVmauth(instance string, flags []string, cli *Client, configFilePath string, output io.Writer) (*Vmauth, error) {
+	binary := os.Getenv("VMAUTH_PATH")
+	if binary == "" {
+		binary = "../../bin/vmauth-race"
+	}
+	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
+		defaultFlags: map[string]string{
+			"-httpListenAddr": "127.0.0.1:0",
+			"-auth.config":    configFilePath,
+		},
+		extractREs: []*regexp.Regexp{
+			vmauthHttpListenAddrRE,
+		},
+		output: output,
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	return newVmauth(app, cli, configFilePath, vmauthRuntimeValues{
+		httpListenAddr: stderrExtracts[0],
+	}), nil
+}
+
+type vmauthRuntimeValues struct {
+	httpListenAddr string
+}
+
+func newVmauth(app *app, cli *Client, configFilePath string, rt vmauthRuntimeValues) *Vmauth {
+	return &Vmauth{
+		app:            app,
+		metricsClient:  newMetricsClient(cli, rt.httpListenAddr),
+		httpListenAddr: rt.httpListenAddr,
+		configFilePath: configFilePath,
+		cli:            cli,
+	}
+}
 
 // Vmauth holds the state of a vmauth app and provides vmauth-specific
 // functions.
@@ -18,38 +60,14 @@ type Vmauth struct {
 	*app
 	*metricsClient
 
+	cli            *Client
 	httpListenAddr string
 	configFilePath string
-	cli            *Client
 }
 
-// StartVmauth starts an instance of vmauth with the given flags. It also
-// sets the default flags and populates the app instance state with runtime
-// values extracted from the application log (such as httpListenAddr)
-func StartVmauth(instance string, flags []string, cli *Client, configFilePath string, output io.Writer) (*Vmauth, error) {
-	extractREs := []*regexp.Regexp{
-		httpBuilitinListenAddrRE,
-	}
-
-	app, stderrExtracts, err := startApp(instance, "../../bin/vmauth-race", flags, &appOptions{
-		defaultFlags: map[string]string{
-			"-httpListenAddr": "127.0.0.1:0",
-			"-auth.config":    configFilePath,
-		},
-		extractREs: extractREs,
-		output:     output,
-	})
-	if err != nil {
-		return nil, err
-	}
-
-	return &Vmauth{
-		app:            app,
-		metricsClient:  newMetricsClient(cli, stderrExtracts[0]),
-		httpListenAddr: stderrExtracts[0],
-		configFilePath: configFilePath,
-		cli:            cli,
-	}, nil
+// GetHTTPListenAddr returns listen http addr
+func (app *Vmauth) GetHTTPListenAddr() string {
+	return app.httpListenAddr
 }
 
 // UpdateConfiguration updates the vmauth configuration file with the provided YAML content,
@@ -79,8 +97,3 @@ func (app *Vmauth) UpdateConfiguration(t *testing.T, configFileYAML string) {
 
 	t.Fatalf("config were not reloaded after SIGHUP signal; previous total: %d, current total: %d", prevTotal, currTotal)
 }
-
-// GetHTTPListenAddr returns listen http addr
-func (app *Vmauth) GetHTTPListenAddr() string {
-	return app.httpListenAddr
-}

apptest/vmbackup.go

@@ -1,15 +1,26 @@
 package apptest
 
-import "io"
+import (
+	"io"
+	"os"
+)
 
-// StartVmbackup starts an instance of vmbackup with the given flags and waits
-// until it exits.
+// StartVmbackup starts the latest version of vmbackup with the given flags and
+// waits until it exits.
+//
+// The path to the binary can be provided via VMBACKUP_PATH environment
+// variable. If the variable is not set, ../../bin/vmbackup-race will be
+// used.
 func StartVmbackup(instance, storageDataPath, snapshotCreateURL, dst string, output io.Writer) error {
+	binary := os.Getenv("VMBACKUP_PATH")
+	if binary == "" {
+		binary = "../../bin/vmbackup-race"
+	}
 	flags := []string{
 		"-storageDataPath=" + storageDataPath,
 		"-snapshot.createURL=" + snapshotCreateURL,
 		"-dst=" + dst,
 	}
-	_, _, err := startApp(instance, "../../bin/vmbackup-race", flags, &appOptions{wait: true, output: output})
+	_, _, err := startApp(instance, binary, flags, &appOptions{wait: true, output: output})
 	return err
 }

apptest/vmctl.go

@@ -1,9 +1,23 @@
 package apptest
 
-import "io"
+import (
+	"io"
+	"os"
+)
 
 // StartVmctl starts an instance of vmctl cli with the given flags
+
+// StartVmctl starts the latest version of vmctl with the given flags and
+// waits until it exits.
+//
+// The path to the binary can be provided via VMCTL_PATH environment
+// variable. If the variable is not set, ../../bin/vmctl-race will be
+// used.
 func StartVmctl(instance string, flags []string, output io.Writer) error {
-	_, _, err := startApp(instance, "../../bin/vmctl-race", flags, &appOptions{wait: true, output: output})
+	binary := os.Getenv("VMCTL_PATH")
+	if binary == "" {
+		binary = "../../bin/vmctl-race"
+	}
+	_, _, err := startApp(instance, binary, flags, &appOptions{wait: true, output: output})
 	return err
 }

apptest/vminsert.go

@@ -3,23 +3,13 @@ package apptest
 import (
 	"fmt"
 	"io"
+	"os"
 	"regexp"
 	"strings"
 	"testing"
 	"time"
 )
 
-// Vminsert holds the state of a vminsert app and provides vminsert-specific
-// functions.
-type Vminsert struct {
-	*app
-	*metricsClient
-	*vminsertClient
-
-	httpListenAddr          string
-	clusternativeListenAddr string
-}
-
 // storageNodes returns the storage node addresses passed to vminsert via
 // -storageNode command line flag.
 func storageNodes(flags []string) []string {
@@ -31,9 +21,11 @@ func storageNodes(flags []string) []string {
 	return nil
 }
 
-// StartVminsert starts an instance of vminsert with the given flags. It also
-// sets the default flags and populates the app instance state with runtime
-// values extracted from the application log (such as httpListenAddr)
+// StartVminsert starts the latest version of vminsert.
+//
+// The path to the binary can be provided via VMINSERT_PATH environment
+// variable. If the variable is not set, ../../bin/vminsert-race will be
+// used.
 func StartVminsert(instance string, flags []string, cli *Client, output io.Writer) (*Vminsert, error) {
 	extractREs := []*regexp.Regexp{
 		httpListenAddrRE,
@@ -48,11 +40,15 @@ func StartVminsert(instance string, flags []string, cli *Client, output io.Write
 		extractREs = append(extractREs, regexp.MustCompile(logRecord))
 	}
 
-	app, stderrExtracts, err := startApp(instance, "../../bin/vminsert-race", flags, &appOptions{
+	binary := os.Getenv("VMINSERT_PATH")
+	if binary == "" {
+		binary = "../../bin/vminsert-race"
+	}
+	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
 		defaultFlags: map[string]string{
 			"-httpListenAddr":                              "127.0.0.1:0",
 			"-clusternativeListenAddr":                     "127.0.0.1:0",
-			"-graphiteListenAddr":                          ":0",
+			"-graphiteListenAddr":                          "127.0.0.1:0",
 			"-opentsdbListenAddr":                          "127.0.0.1:0",
 			"-clusternative.vminsertConnsShutdownDuration": "1ms",
 		},
@@ -63,27 +59,56 @@ func StartVminsert(instance string, flags []string, cli *Client, output io.Write
 		return nil, err
 	}
 
-	metricsClient := newMetricsClient(cli, stderrExtracts[0])
-	return &Vminsert{
-		app:           app,
-		metricsClient: metricsClient,
-		vminsertClient: &vminsertClient{
-			vminsertCli: cli,
-			url: func(op, path string, opts QueryOpts) string {
-				return getClusterPath(stderrExtracts[0], op, path, opts)
-			},
-			openTSDBURL: func(op, path string, opts QueryOpts) string {
-				return getClusterPath(stderrExtracts[3], op, path, opts)
-			},
-			graphiteListenAddr: stderrExtracts[2],
-			sendBlocking: func(t *testing.T, numRecordsToSend int, send func()) {
-				t.Helper()
-				sendBlocking(t, metricsClient, numRecordsToSend, send)
-			},
-		},
+	return newVminsert(app, cli, vminsertRuntimeValues{
 		httpListenAddr:          stderrExtracts[0],
 		clusternativeListenAddr: stderrExtracts[1],
-	}, nil
+		graphiteListenAddr:      stderrExtracts[2],
+		openTSDBListenAddr:      stderrExtracts[3],
+	}), nil
+}
+
+type vminsertRuntimeValues struct {
+	httpListenAddr          string
+	clusternativeListenAddr string
+	graphiteListenAddr      string
+	openTSDBListenAddr      string
+}
+
+func newVminsert(app *app, cli *Client, rt vminsertRuntimeValues) *Vminsert {
+	metricsClient := newMetricsClient(cli, rt.httpListenAddr)
+	vminsertClient := &vminsertClient{
+		vminsertCli: cli,
+		url: func(op, path string, opts QueryOpts) string {
+			return getClusterPath(rt.httpListenAddr, op, path, opts)
+		},
+		openTSDBURL: func(op, path string, opts QueryOpts) string {
+			return getClusterPath(rt.openTSDBListenAddr, op, path, opts)
+		},
+		graphiteListenAddr: rt.graphiteListenAddr,
+		sendBlocking: func(t *testing.T, numRecordsToSend int, send func()) {
+			t.Helper()
+			sendBlocking(t, metricsClient, numRecordsToSend, send)
+		},
+	}
+
+	return &Vminsert{
+		app:                     app,
+		metricsClient:           metricsClient,
+		vminsertClient:          vminsertClient,
+		httpListenAddr:          rt.httpListenAddr,
+		clusternativeListenAddr: rt.clusternativeListenAddr,
+	}
+}
+
+// Vminsert holds the state of a vminsert app and provides vminsert-specific
+// functions.
+type Vminsert struct {
+	*app
+	*metricsClient
+	*vminsertClient
+
+	httpListenAddr          string
+	clusternativeListenAddr string
 }
 
 // ClusternativeListenAddr returns the address at which the vminsert process is

apptest/vmrestore.go

@@ -1,14 +1,25 @@
 package apptest
 
-import "io"
+import (
+	"io"
+	"os"
+)
 
-// StartVmrestore starts an instance of vmrestore with the given flags and waits
-// until it exits.
+// StartVmrestore starts the latest version of vmrestore with the given flags
+// and waits until it exits.
+//
+// The path to the binary can be provided via VMRESTORE_PATH environment
+// variable. If the variable is not set, ../../bin/vmrestore-race will be
+// used.
 func StartVmrestore(instance, src, storageDataPath string, output io.Writer) error {
+	binary := os.Getenv("VMRESTORE_PATH")
+	if binary == "" {
+		binary = "../../bin/vmrestore-race"
+	}
 	flags := []string{
 		"-src=" + src,
 		"-storageDataPath=" + storageDataPath,
 	}
-	_, _, err := startApp(instance, "../../bin/vmrestore-race", flags, &appOptions{wait: true, output: output})
+	_, _, err := startApp(instance, binary, flags, &appOptions{wait: true, output: output})
 	return err
 }

apptest/vmselect.go

@@ -3,26 +3,21 @@ package apptest
 import (
 	"fmt"
 	"io"
+	"os"
 	"regexp"
 )
 
-// Vmselect holds the state of a vmselect app and provides vmselect-specific
-// functions.
-type Vmselect struct {
-	*app
-	*metricsClient
-	*vmselectClient
-
-	httpListenAddr          string
-	clusternativeListenAddr string
-	cli                     *Client
-}
-
-// StartVmselect starts an instance of vmselect with the given flags. It also
-// sets the default flags and populates the app instance state with runtime
-// values extracted from the application log (such as httpListenAddr)
+// StartVmselect starts the latest version of vmselect.
+//
+// The path to the binary can be provided via VMSELECT_PATH environment
+// variable. If the variable is not set, ../../bin/vmselect-race will be
+// used.
 func StartVmselect(instance string, flags []string, cli *Client, output io.Writer) (*Vmselect, error) {
-	app, stderrExtracts, err := startApp(instance, "../../bin/vmselect-race", flags, &appOptions{
+	binary := os.Getenv("VMSELECT_PATH")
+	if binary == "" {
+		binary = "../../bin/vmselect-race"
+	}
+	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
 		defaultFlags: map[string]string{
 			"-httpListenAddr":          "127.0.0.1:0",
 			"-clusternativeListenAddr": "127.0.0.1:0",
@@ -37,21 +32,43 @@ func StartVmselect(instance string, flags []string, cli *Client, output io.Write
 		return nil, err
 	}
 
+	return newVmselect(app, cli, vmselectRuntimeValues{
+		httpListenAddr:          stderrExtracts[0],
+		clusternativeListenAddr: stderrExtracts[1],
+	}), nil
+}
+
+type vmselectRuntimeValues struct {
+	httpListenAddr          string
+	clusternativeListenAddr string
+}
+
+func newVmselect(app *app, cli *Client, rt vmselectRuntimeValues) *Vmselect {
 	return &Vmselect{
 		app:           app,
-		metricsClient: newMetricsClient(cli, stderrExtracts[0]),
+		metricsClient: newMetricsClient(cli, rt.httpListenAddr),
 		vmselectClient: &vmselectClient{
 			vmselectCli: cli,
 			url: func(op, path string, opts QueryOpts) string {
-				return getClusterPath(stderrExtracts[0], op, path, opts)
+				return getClusterPath(rt.httpListenAddr, op, path, opts)
 			},
-			metricNamesStatsResetURL: fmt.Sprintf("http://%s/admin/api/v1/admin/status/metric_names_stats/reset", stderrExtracts[0]),
-			tenantsURL:               fmt.Sprintf("http://%s/admin/tenants", stderrExtracts[0]),
+			metricNamesStatsResetURL: fmt.Sprintf("http://%s/admin/api/v1/admin/status/metric_names_stats/reset", rt.httpListenAddr),
+			tenantsURL:               fmt.Sprintf("http://%s/admin/tenants", rt.httpListenAddr),
 		},
-		httpListenAddr:          stderrExtracts[0],
-		clusternativeListenAddr: stderrExtracts[1],
-		cli:                     cli,
-	}, nil
+		httpListenAddr:          rt.httpListenAddr,
+		clusternativeListenAddr: rt.clusternativeListenAddr,
+	}
+}
+
+// Vmselect holds the state of a vmselect app and provides vmselect-specific
+// functions.
+type Vmselect struct {
+	*app
+	*metricsClient
+	*vmselectClient
+
+	httpListenAddr          string
+	clusternativeListenAddr string
 }
 
 // ClusternativeListenAddr returns the address at which the vmselect process is

apptest/vmsingle.go

@@ -9,28 +9,21 @@ import (
 	"time"
 )
 
-// Vmsingle holds the state of a vmsingle app and provides vmsingle-specific
-// functions.
-type Vmsingle struct {
-	*app
-	*metricsClient
-	*vmstorageClient
-	*vmselectClient
-	*vminsertClient
-
-	storageDataPath string
-	httpListenAddr  string
-}
-
-// StartVmsingleAt starts an instance of vmsingle with the given flags. It also
-// sets the default flags and populates the app instance state with runtime
-// values extracted from the application log (such as httpListenAddr).
-func StartVmsingleAt(instance, binary string, flags []string, cli *Client, output io.Writer) (*Vmsingle, error) {
+// StartVmsingle starts the latest version of vmsingle.
+//
+// The path to the binary can be provided via VMSINGLE_PATH environment
+// variable. If the variable is not set, ../../bin/victoria-metrics-race will be
+// used.
+func StartVmsingle(instance string, flags []string, cli *Client, output io.Writer) (*Vmsingle, error) {
+	binary := os.Getenv("VMSINGLE_PATH")
+	if binary == "" {
+		binary = "../../bin/victoria-metrics-race"
+	}
 	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
 		defaultFlags: map[string]string{
 			"-storageDataPath":    fmt.Sprintf("%s/%s-%d", os.TempDir(), instance, time.Now().UnixNano()),
 			"-httpListenAddr":     "127.0.0.1:0",
-			"-graphiteListenAddr": ":0",
+			"-graphiteListenAddr": "127.0.0.1:0",
 			"-opentsdbListenAddr": "127.0.0.1:0",
 		},
 		extractREs: []*regexp.Regexp{
@@ -45,38 +38,67 @@ func StartVmsingleAt(instance, binary string, flags []string, cli *Client, outpu
 		return nil, err
 	}
 
+	return newVmsingle(app, cli, vmsingleRuntimeValues{
+		storageDataPath:    stderrExtracts[0],
+		httpListenAddr:     stderrExtracts[1],
+		graphiteListenAddr: stderrExtracts[2],
+		openTSDBListenAddr: stderrExtracts[3],
+	}), nil
+}
+
+type vmsingleRuntimeValues struct {
+	storageDataPath    string
+	httpListenAddr     string
+	graphiteListenAddr string
+	openTSDBListenAddr string
+}
+
+func newVmsingle(app *app, cli *Client, rt vmsingleRuntimeValues) *Vmsingle {
 	return &Vmsingle{
 		app:           app,
-		metricsClient: newMetricsClient(cli, stderrExtracts[1]),
+		metricsClient: newMetricsClient(cli, rt.httpListenAddr),
 		vmstorageClient: &vmstorageClient{
 			vmstorageCli:   cli,
-			httpListenAddr: stderrExtracts[1],
+			httpListenAddr: rt.httpListenAddr,
 		},
 		vmselectClient: &vmselectClient{
 			vmselectCli: cli,
 			url: func(op, path string, opts QueryOpts) string {
-				return fmt.Sprintf("http://%s/%s", stderrExtracts[1], path)
+				return fmt.Sprintf("http://%s/%s", rt.httpListenAddr, path)
 			},
-			metricNamesStatsResetURL: fmt.Sprintf("http://%s/api/v1/admin/status/metric_names_stats/reset", stderrExtracts[1]),
+			metricNamesStatsResetURL: fmt.Sprintf("http://%s/api/v1/admin/status/metric_names_stats/reset", rt.httpListenAddr),
 			tenantsURL:               "vmsingle-does-not-serve-tenants",
 		},
 		vminsertClient: &vminsertClient{
 			vminsertCli: cli,
 			url: func(_, path string, _ QueryOpts) string {
-				return fmt.Sprintf("http://%s/%s", stderrExtracts[1], path)
+				return fmt.Sprintf("http://%s/%s", rt.httpListenAddr, path)
 			},
 			openTSDBURL: func(_, path string, _ QueryOpts) string {
-				return fmt.Sprintf("http://%s/%s", stderrExtracts[3], path)
+				return fmt.Sprintf("http://%s/%s", rt.openTSDBListenAddr, path)
 			},
-			graphiteListenAddr: stderrExtracts[2],
+			graphiteListenAddr: rt.graphiteListenAddr,
 			sendBlocking: func(t *testing.T, _ int, send func()) {
 				t.Helper()
 				send()
 			},
 		},
-		storageDataPath: stderrExtracts[0],
-		httpListenAddr:  stderrExtracts[1],
-	}, nil
+		storageDataPath: rt.storageDataPath,
+		httpListenAddr:  rt.httpListenAddr,
+	}
+}
+
+// Vmsingle holds the state of a vmsingle app and provides vmsingle-specific
+// functions.
+type Vmsingle struct {
+	*app
+	*metricsClient
+	*vmstorageClient
+	*vmselectClient
+	*vminsertClient
+
+	storageDataPath string
+	httpListenAddr  string
 }
 
 // HTTPAddr returns the address at which the vminsert process is

apptest/vmsingle_legacy.go

@@ -0,0 +1,43 @@
+package apptest
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"regexp"
+	"time"
+)
+
+// StartVmsingle_v1_132_0 starts vmsingle-v1.132.0 (the last version that uses
+// legacy index).
+//
+// The path to the binary must be provided via VMSINGLE_V1_132_0_PATH
+// environment variable.
+func StartVmsingle_v1_132_0(instance string, flags []string, cli *Client, output io.Writer) (*Vmsingle, error) {
+	binary := os.Getenv("VMSINGLE_V1_132_0_PATH")
+	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
+		defaultFlags: map[string]string{
+			"-storageDataPath":    fmt.Sprintf("%s/%s-%d", os.TempDir(), instance, time.Now().UnixNano()),
+			"-httpListenAddr":     "127.0.0.1:0",
+			"-graphiteListenAddr": "127.0.0.1:0",
+			"-opentsdbListenAddr": "127.0.0.1:0",
+		},
+		extractREs: []*regexp.Regexp{
+			storageDataPathRE,
+			httpListenAddrRE,
+			graphiteListenAddrRE,
+			openTSDBListenAddrRE,
+		},
+		output: output,
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	return newVmsingle(app, cli, vmsingleRuntimeValues{
+		storageDataPath:    stderrExtracts[0],
+		httpListenAddr:     stderrExtracts[1],
+		graphiteListenAddr: stderrExtracts[2],
+		openTSDBListenAddr: stderrExtracts[3],
+	}), nil
+}

apptest/vmstorage.go

@@ -8,23 +8,22 @@ import (
 	"time"
 )
 
-// Vmstorage holds the state of a vmstorage app and provides vmstorage-specific
-// functions.
-type Vmstorage struct {
-	*app
-	*metricsClient
-	*vmstorageClient
-
-	storageDataPath string
-	httpListenAddr  string
-	vminsertAddr    string
-	vmselectAddr    string
+// StartVmstorage starts the latest version of vmstorage.
+//
+// The path to the binary can be provided via VMSTORAGE_PATH environment
+// variable. If the variable is not set, ../../bin/vmstorage-race will be used.
+func StartVmstorage(instance string, flags []string, cli *Client, output io.Writer) (*Vmstorage, error) {
+	binary := os.Getenv("VMSTORAGE_PATH")
+	if binary == "" {
+		binary = "../../bin/vmstorage-race"
+	}
+	return startVmstorage(instance, binary, flags, cli, output)
 }
 
-// StartVmstorageAt starts an instance of vmstorage with the given flags. It also
+// startVmstorage starts an instance of vmstorage with the given flags. It also
 // sets the default flags and populates the app instance state with runtime
 // values extracted from the application log (such as httpListenAddr)
-func StartVmstorageAt(instance, binary string, flags []string, cli *Client, output io.Writer) (*Vmstorage, error) {
+func startVmstorage(instance, binary string, flags []string, cli *Client, output io.Writer) (*Vmstorage, error) {
 	app, stderrExtracts, err := startApp(instance, binary, flags, &appOptions{
 		defaultFlags: map[string]string{
 			"-storageDataPath": fmt.Sprintf("%s/%s-%d", os.TempDir(), instance, time.Now().UnixNano()),
@@ -44,18 +43,47 @@ func StartVmstorageAt(instance, binary string, flags []string, cli *Client, outp
 		return nil, err
 	}
 
-	return &Vmstorage{
-		app:           app,
-		metricsClient: newMetricsClient(cli, stderrExtracts[1]),
-		vmstorageClient: &vmstorageClient{
-			vmstorageCli:   cli,
-			httpListenAddr: stderrExtracts[1],
-		},
+	return newVmstorage(app, cli, vmstorageRuntimeValues{
 		storageDataPath: stderrExtracts[0],
 		httpListenAddr:  stderrExtracts[1],
 		vminsertAddr:    stderrExtracts[2],
 		vmselectAddr:    stderrExtracts[3],
-	}, nil
+	}), nil
+}
+
+type vmstorageRuntimeValues struct {
+	storageDataPath string
+	httpListenAddr  string
+	vminsertAddr    string
+	vmselectAddr    string
+}
+
+func newVmstorage(app *app, cli *Client, rt vmstorageRuntimeValues) *Vmstorage {
+	return &Vmstorage{
+		app:           app,
+		metricsClient: newMetricsClient(cli, rt.httpListenAddr),
+		vmstorageClient: &vmstorageClient{
+			vmstorageCli:   cli,
+			httpListenAddr: rt.httpListenAddr,
+		},
+		storageDataPath: rt.storageDataPath,
+		httpListenAddr:  rt.httpListenAddr,
+		vminsertAddr:    rt.vminsertAddr,
+		vmselectAddr:    rt.vmselectAddr,
+	}
+}
+
+// Vmstorage holds the state of a vmstorage app and provides vmstorage-specific
+// functions.
+type Vmstorage struct {
+	*app
+	*metricsClient
+	*vmstorageClient
+
+	storageDataPath string
+	httpListenAddr  string
+	vminsertAddr    string
+	vmselectAddr    string
 }
 
 // VminsertAddr returns the address at which the vmstorage process is listening

apptest/vmstorage_legacy.go

@@ -0,0 +1,16 @@
+package apptest
+
+import (
+	"io"
+	"os"
+)
+
+// StartVmstorage_v1_132_0 starts vmstorage-v1.132.0 (the last version that uses
+// legacy index).
+//
+// The path to the binary must be provided via VMSTORAGE_V1_132_0_PATH
+// environment variable.
+func StartVmstorage_v1_132_0(instance string, flags []string, cli *Client, output io.Writer) (*Vmstorage, error) {
+	binary := os.Getenv("VMSTORAGE_V1_132_0_PATH")
+	return startVmstorage(instance, binary, flags, cli, output)
+}

dashboards/vm/vmalert.json

@@ -2804,10 +2804,10 @@
             "overrides": []
           },
           "gridPos": {
-            "h": 7,
+            "h": 8,
             "w": 12,
             "x": 0,
-            "y": 11
+            "y": 352
           },
           "id": 63,
           "options": {
@@ -2843,113 +2843,7 @@
           ],
           "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/vmalert.json

@@ -2803,10 +2803,10 @@
             "overrides": []
           },
           "gridPos": {
-            "h": 7,
+            "h": 8,
             "w": 12,
             "x": 0,
-            "y": 11
+            "y": 352
           },
           "id": 63,
           "options": {
@@ -2842,113 +2842,7 @@
           ],
           "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"

deployment/docker/rules/alerts-vmalert.yml

@@ -64,18 +64,6 @@ 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
@@ -120,3 +108,4 @@ 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."
+

docs/ai-tools/README.md

@@ -88,6 +88,7 @@ These skills provide predefined workflows and capabilities such as:
 * Multi-signal investigations 
 * Cardinality optimization 
 * Unused metric detection
+* Stream aggregation configuration
 
 To install the available skills for AI agents, run:
 ```sh

docs/playgrounds/README.md

@@ -150,8 +150,12 @@ You can experiment with your own data during the month‑long trial without depl
 are fast-booting Linux microVMs that run on a fleet of large bare-metal servers. You can start a playground right from your browser. 
 Once up and running, accessing a playground is no different from SSH-ing into a remote server rented from your favorite VPS or Cloud provider.
 
-Iximiuz Labs provides playgrounds for VictoriaMetrics software:
-- [VictoriaMetrics single node (on Ubuntu)](https://labs.iximiuz.com/playgrounds/victoriametrics-e2f9b613)
-- [VictoriaMetrics cluster (on Ubuntu)](https://labs.iximiuz.com/playgrounds/victoriametrics-cluster-8eacb19d)
-- [Getting Started with VictoriaMetrics on Kubernetes](https://labs.iximiuz.com/tutorials/victoriametrics-getting-started-kubernetes-0e9c0993)
-- [VictoriaMetrics Operator](https://labs.iximiuz.com/playgrounds/victoriametrics-kubernetes-9eebc258)
+Iximiuz Labs provides various [learning-by-doing resources for VictoriaMetrics](https://labs.iximiuz.com/v/victoriametrics):
+- Tutorial:
+  - [Getting Started with VictoriaMetrics on Kubernetes](https://labs.iximiuz.com/tutorials/victoriametrics-getting-started-kubernetes)
+- Playgrounds:
+  - [VictoriaMetrics single node](https://labs.iximiuz.com/playgrounds/victoriametrics)
+  - [VictoriaMetrics cluster](https://labs.iximiuz.com/playgrounds/victoriametrics-cluster)
+  - [VictoriaMetrics on Kubernetes](https://labs.iximiuz.com/playgrounds/victoriametrics-kubernetes)
+
+Iximiuz Labs requires a [free account](https://labs.iximiuz.com/signup) to access the materials.

docs/victoriametrics/Quick-Start.md

@@ -43,6 +43,9 @@ Just download VictoriaMetrics and follow [these instructions](https://docs.victo
 See [available integrations](https://docs.victoriametrics.com/victoriametrics/integrations/) with other systems like
 [Prometheus](https://docs.victoriametrics.com/victoriametrics/integrations/prometheus/) or [Grafana](https://docs.victoriametrics.com/victoriametrics/integrations/grafana/).
 
+> Want to see VictoriaMetrics in action, but without installing anything?
+> Try [Playgrounds](https://docs.victoriametrics.com/playgrounds/) - a list of publicly available playgrounds for VictoriaMetrics software.
+
 VictoriaMetrics is developed at a fast pace, so it is recommended to periodically check the [CHANGELOG](https://docs.victoriametrics.com/victoriametrics/changelog/)
 and perform [regular upgrades](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#how-to-upgrade-victoriametrics).
 

docs/victoriametrics/changelog/CHANGELOG.md

@@ -28,8 +28,6 @@ See also [LTS releases](https://docs.victoriametrics.com/victoriametrics/lts-rel
 
 * FEATURE: [vmagent](https://docs.victoriametrics.com/victoriametrics/vmagent/), [vmsingle](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/) and `vminsert` in [VictoriaMetrics cluster](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/): add `-opentelemetry.promoteAllResourceAttributes` and `-opentelemetry.promoteScopeMetadata` command-line flags to allow managing label promotion for resource attributes and OTel scope metadata. See [OpenTelemetry](https://docs.victoriametrics.com/victoriametrics/integrations/opentelemetry/) docs and [#10931](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10931).
 
-* BUGFIX: [vmalert](https://docs.victoriametrics.com/victoriametrics/vmalert/): reset the group evaluation timestamp if it exceeds the current host time. Previously, vmalert could use future timestamps for evaluations if the system clock was shifted backward. See [#10985](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10985).
-
 ## [v1.144.0](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.144.0)
 
 Released at 2026-05-22

docs/victoriametrics/integrations/grafana/_index.md

@@ -59,6 +59,10 @@ Once connected, you can build graphs and dashboards using [PromQL](https://prome
 _Creating a datasource may require [specific permissions](https://grafana.com/docs/grafana/latest/administration/data-source-management/).
 If you don't see an option to create a data source - try contacting system administrator._
 
+If you run [vmalert](https://docs.victoriametrics.com/victoriametrics/vmalert/) and want to see its rules in [Grafana Alerting UI](https://grafana.com/docs/grafana/latest/alerting/),
+then set configure `-vmalert.proxyURL` on VictoriaMetrics [single-node](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#vmalert)
+or [vmselect in cluster version](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#vmalert).
+
 ## Multi-tenant access with vmauth and OIDC
 
 [vmauth](https://docs.victoriametrics.com/victoriametrics/vmauth/) can proxy Grafana datasource requests and enforce

docs/victoriametrics/stream-aggregation/README.md

@@ -26,6 +26,7 @@ Stream aggregation has the following features:
   and/or scraped from [Prometheus-compatible targets](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#how-to-scrape-prometheus-exporters-such-as-node-exporter)
 - It can filter out raw samples matched by aggregation rules, so raw data will never reach the remote destination. See `-streamAggr.keepInput` and `-streamAggr.dropInput` in [aggregation config](https://docs.victoriametrics.com/victoriametrics/stream-aggregation/configuration/);
 - It allows building [flexible processing pipelines](#routing);
+- It is [horizontally scalable](https://docs.victoriametrics.com/victoriametrics/stream-aggregation/#scaling-aggregation-horizontally).
 
 # Limitations
 
@@ -598,6 +599,47 @@ Below is an example of an `aggr.yaml` configuration that drops the `replica` and
   keep_metric_names: true
 ```
 
+## Scaling aggregation horizontally
+
+Aggregation output is only correct when all contributing samples are processed by the same aggregator instance.
+
+To scale the aggregation horizontally, always shard the input samples in a deterministic way. This can be achieved by
+building a two layer topology of vmagents where the first layer is responsible for sharding, and the second layer is responsible for aggregating:
+```mermaid
+flowchart LR
+    V1[vmagent-shard-1] -- requests_total{env=test, pod=foo} --> SV1[vmagent-aggr-1]
+    V1[vmagent-shard-1] -- requests_total{env=prod, pod=bar} --> SV2[vmagent-aggr-1]
+    V2[vmagent-shard-2] -- requests_total{env=prod, pod=baz} --> SV2[vmagent-aggr-2]
+    SV1 -- requests_total:5m_without_pod_total{env=test} --> x(( ))
+    SV2 -- requests_total:5m_without_pod_total{env=prod} --> y(( ))
+style x fill:none,stroke:none
+style y fill:none,stroke:none
+```
+
+The sharding layer of vmagents can be configured via the `-remoteWrite.shardByURL.labels` or `-remoteWrite.shardByURL.ignoreLabels`
+command line flags. See how to [shard data across remote write destinations](https://docs.victoriametrics.com/victoriametrics/vmagent/#sharding-among-remote-storages) for more details.
+
+The following requirements must be met for sharded aggregation to work correctly:
+- All sharding vmagents should have the same deterministic sharding configuration.
+- The sharding configuration must align with the `by` and `without` lists:
+  - Labels listed in `by` setting should be a subset of shard's routing key `-remoteWrite.shardByURL.labels`. 
+    With `-remoteWrite.shardByURL.labels=env,job` aggregator's `by` should include `by: env`, `by: job` or both: `by: [env, job]`.
+    This makes sure that all the samples for the same `env` and `job` are aggregated together and produce the complete output.
+  - Labels listed in `without` setting should be a superset of shard's routing key `--remoteWrite.shardByURL.ignoreLabels`.
+    With `-remoteWrite.shardByURL.ignoreLabels=env,job` aggegator's `without` should include at least both labels `without: [env,job]`.
+    This makes sure that `requests_total{env=test, job=foo}` and `requests_total{env=prod, job=foo}` are routed to the same aggregator
+    and are aggregated together. See also [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5938#issuecomment-2018470324).
+- Aggregating vmagents should not produce collisions: the aggregation output should be unique across all the sharded agents.
+  For example, `requests_total:5m_without_env_pod_total` produced by both `vmagent-aggr-1` and `vmagent-aggr-2` will collide
+  unless they have labels uniquely identifying them. These labels should be either preserved during sharding and aggregation config,
+  or enforced on the output via `-remoteWrite.label` - see [these docs](https://docs.victoriametrics.com/victoriametrics/stream-aggregation/#cluster-mode) for more details.
+
+> Never shard histograms by `le` (or `vmrange` in case of VM histograms) label. A histogram is a logical group of series differing
+only in the bucket label. All of those buckets must land on the same aggregator at the same time so it can produce a
+coherent bucket set. See more about [aggregating histograms](https://docs.victoriametrics.com/stream-aggregation/#aggregating-histograms).
+
+See also [why you shouldn't put an aggregator behind a load balancer](https://docs.victoriametrics.com/stream-aggregation/#put-aggregator-behind-load-balancer).
+
 # Troubleshooting
 
 - [Unexpected spikes for `total` or `increase` outputs](#staleness).

docs/victoriametrics/vmagent.md

@@ -182,15 +182,15 @@ among remote storage systems specified in `-remoteWrite.url`.
 > For example, if you set `-remoteWrite.url=srv+foo` and it's resolved to three addresses (`192.168.1.1`, `192.168.1.2`, `192.168.1.3`),
 > vmagent will only choose **one** randomly every time it (re-)creates the connection. In contrast, specifying the addresses manually (`-remoteWrite.url=192.168.1.1 -remoteWrite.url=192.168.1.2 -remoteWrite.url=192.168.1.3`) will shard samples across all three URLs.
 
-Sometimes, it may be necessary to use only a particular set of labels for sharding. For example, it may be necessary to route all the metrics with the same `instance` label
-to the same `-remoteWrite.url`. In this case, you can specify a comma-separated list of these labels in the `-remoteWrite.shardByURL.labels`
-command-line flag. For example, `-remoteWrite.shardByURL.labels=instance,__name__` would shard metrics with the same name and `instance`
-label to the same `-remoteWrite.url`.
+Use `-remoteWrite.shardByURL.labels` to route metrics among `-remoteWrite.url` based on their label values. 
+For example, `-remoteWrite.shardByURL.labels=instance,__name__` would shard metrics with the same name and `instance`
+label to the same `-remoteWrite.url`. This command-line flag allows specifying a comma-separated list of labels.
 
-Sometimes, it may be necessary to ignore some labels when sharding samples across multiple `-remoteWrite.url` backends.
-For example, if all the [raw samples](https://docs.victoriametrics.com/victoriametrics/keyconcepts/#raw-samples) with the same set of labels
-except for the labels `instance` and `pod` must be routed to the same backend. In this case the list of ignored labels must be passed to
-`-remoteWrite.shardByURL.ignoreLabels` command-line flag: `-remoteWrite.shardByURL.ignoreLabels=instance,pod`.
+Alternatively, you can use `-remoteWrite.shardByURL.ignoreLabels` to route metrics among `-remoteWrite.url` based on their label values, excluding the specified labels.
+For example, `-remoteWrite.shardByURL.ignoreLabels=pod` would shard metrics `metric{pod="foo"}` and `metric{pod="bar"}` to the same `-remoteWrite.url`
+by ignoring the `pod` label. This command-line flag allows specifying a comma-separated list of labels.
+
+> Command-line flags `-remoteWrite.shardByURL.labels` and `-remoteWrite.shardByURL.ignoreLabels` are mutually exclusive.
 
 See also [how to scrape a large number of targets](#scraping-big-number-of-targets).
 

docs/victoriametrics/vmalert.md

@@ -21,23 +21,9 @@ Recording rules results are persisted via remote write protocols and require `-r
 `vmalert` is heavily inspired by [Prometheus](https://prometheus.io/docs/alerting/latest/overview/)
 implementation and aims to be compatible with its syntax.
 
-Configure `-vmalert.proxyURL` on VictoriaMetrics [single-node](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#vmalert)
-or [vmselect in cluster version](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#vmalert)
-to proxy requests to `vmalert`. Proxying is needed for the following cases:
-
-* to proxy requests from [Grafana Alerting UI](https://grafana.com/docs/grafana/latest/alerting/);
-* to access `vmalert`'s UI through [vmui](https://docs.victoriametrics.com/victoriametrics/#vmui).
-
-[VictoriaMetrics Cloud](https://console.victoriametrics.cloud/signUp?utm_source=website&utm_campaign=docs_vm_vmalert_intro)
-provides out-of-the-box alerting functionality based on `vmalert`. This service simplifies the setup
-and management of alerting and recording rules as well as the integration with Alertmanager. For more details,
-please refer to the [VictoriaMetrics Cloud documentation](https://docs.victoriametrics.com/victoriametrics-cloud/alertmanager-setup-for-deployment/).
-
 ## Features
 
-* Integration with [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics) and [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/);
-* Integration with [VictoriaLogs](https://docs.victoriametrics.com/victorialogs/) and [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/). See [this doc](https://docs.victoriametrics.com/victorialogs/vmalert/);
-* Integration with [VictoriaTraces](https://docs.victoriametrics.com/victoriatraces/) which also uses [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/). See [this doc](https://docs.victoriametrics.com/victoriatraces/vmalert/);
+* Integration with VictoriaMetrics, VictoriaLogs, VictoriaTraces, Graphite and Prometheus compatible storages. See [Integrations](https://docs.victoriametrics.com/victoriametrics/vmalert/#integrations) for details;
 * Prometheus [alerting rules definition format](https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/#defining-alerting-rules)
   support;
 * Integration with [Alertmanager](https://github.com/prometheus/alertmanager) starting from [Alertmanager v0.16.0-alpha](https://github.com/prometheus/alertmanager/releases/tag/v0.16.0-alpha.0);
@@ -62,8 +48,8 @@ please refer to the [VictoriaMetrics Cloud documentation](https://docs.victoriam
 
 To start using `vmalert` you will need the following things:
 
-* list of rules - PromQL/MetricsQL expressions to execute;
-* datasource address - reachable endpoint with [Prometheus HTTP API](https://prometheus.io/docs/prometheus/latest/querying/api/#http-api) support for running queries against;
+* list of rules - PromQL/MetricsQL/LogsQL/GraphiteQL expressions to execute;
+* datasource address - a storage that [vmalert integrates with](https://docs.victoriametrics.com/victoriametrics/vmalert/#integrations) for executing queries;
 * notifier address [optional] - reachable [Alert Manager](https://github.com/prometheus/alertmanager) instance for processing,
   aggregating alerts, and sending notifications. Please note, notifier address also supports Consul and DNS Service Discovery via
   [config file](https://docs.victoriametrics.com/victoriametrics/vmalert/#notifier-configuration-file).
@@ -73,7 +59,7 @@ To start using `vmalert` you will need the following things:
 * remote read address [optional] - MetricsQL compatible datasource to restore alerts state from.
 
 You can use the existing [docker-compose environment](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/master/deployment/docker#victoriametrics-single-server)
-as example. It already contains vmalert configured with list of alerting rules and integrated with Alert Manager and VictoriaMetrics.
+as an example. It already contains vmalert configured with the list of alerting rules and integrated with Alert Manager and VictoriaMetrics.
 
 Alternatively, build `vmalert` from sources:
 
@@ -87,7 +73,7 @@ Then run `vmalert`:
 
 ```sh
 ./bin/vmalert -rule=alert.rules \            # Path to the file with rules configuration. Supports wildcard and HTTP URL (S3/GCS are available in Enterprise).
-    -datasource.url=http://localhost:8428 \  # Prometheus HTTP API compatible datasource
+    -datasource.url=http://localhost:8428 \  # VictoriaMetrics URL to query for rules evaluation. See other available Integrations above.
     -notifier.url=http://localhost:9093 \    # AlertManager URL (required if alerting rules are used)
     -notifier.url=http://127.0.0.1:9093 \    # AlertManager replica URL
     -remoteWrite.url=http://localhost:8428 \ # Remote write compatible storage to persist rules and alerts state info (required if recording rules are used)
@@ -107,7 +93,7 @@ See also [stream aggregation](https://docs.victoriametrics.com/victoriametrics/s
 
 See the full list of configuration flags in [configuration](#configuration) section.
 
-If you run multiple `vmalert` services on the same datastore or AlertManager and need to distinguish the results or alerts, 
+If you run multiple `vmalert` services for the same datasource or AlertManager and need to distinguish the results or alerts, 
 specify different `-external.label` command-line flags to indicate which `vmalert` generated them. 
 If rule result metrics have label that conflict with `-external.label`, `vmalert` will automatically rename
 it with prefix `exported_`.
@@ -116,6 +102,7 @@ Configuration for [recording](https://prometheus.io/docs/prometheus/latest/confi
 and [alerting](https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/) rules is very
 similar to Prometheus rules and configured using YAML. Configuration examples may be found
 in [testdata](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/config/testdata) folder.
+
 Every `rule` belongs to a `group` and every configuration file may contain arbitrary number of groups:
 
 ```yaml
@@ -123,13 +110,7 @@ groups:
   [ - <rule_group> ]
 ```
 
-> Explore how to integrate `vmalert` with [VictoriaMetrics Anomaly Detection](https://docs.victoriametrics.com/anomaly-detection/) in the following [guide](https://docs.victoriametrics.com/anomaly-detection/guides/guide-vmanomaly-vmalert/).
-
-> For users of [VictoriaMetrics Cloud](https://console.victoriametrics.cloud/signUp?utm_source=website&utm_campaign=docs_vm_vmalert_config),
-> many of the configuration steps (including highly available setup of `vmalert` for cluster deployments) are handled automatically.
-> Please, refer to the [VictoriaMetrics Cloud documentation](https://docs.victoriametrics.com/victoriametrics-cloud/alertmanager-setup-for-deployment/) for more details.
-
-### Groups
+## Groups
 
 Each group has the following attributes:
 
@@ -230,9 +211,11 @@ rules:
 
 ### Rules
 
-Every rule contains `expr` field for [PromQL](https://prometheus.io/docs/prometheus/latest/querying/basics/)
-or [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/) expression. `vmalert` will execute the configured
-expression and then act according to the Rule type.
+Every rule contains an `expr` field for the expression to evaluate against the configured datasource.
+Depending on `group.type` value or `-rule.defaultRuleType` cmd-line flag expression can be one of the following types:
+- `prometheus` (default) - [PromQL](https://prometheus.io/docs/prometheus/latest/querying/basics/) or [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/) expression.
+- `vlogs` - [VictoriaLogs](https://docs.victoriametrics.com/victorialogs/vmalert/) expression.
+- `graphite` - [Graphite](https://graphite.readthedocs.io/en/stable/render_api.html) expression.
 
 There are two types of Rules:
 
@@ -244,8 +227,7 @@ There are two types of Rules:
   `-remoteWrite.url`. Recording rules are used to precompute frequently needed or computationally
   expensive expressions and save their result as a new set of time series ([Prometheus recording rules docs](https://prometheus.io/docs/prometheus/latest/configuration/recording_rules/)).
 
-`vmalert` forbids defining duplicates - rules with the same combination of name, expression, and labels
-within one group.
+> `vmalert` forbids defining duplicates - rules with the same combination of name, expression and labels within one group.
 
 #### Alerting rules
 
@@ -256,8 +238,8 @@ The syntax for alerting rule is the following:
 alert: <string>
 
 # The expression to evaluate. The expression language depends on the type value.
-# By default, PromQL/MetricsQL expression is used. If group.type="graphite", then the expression
-# must contain valid Graphite expression.
+# By default, PromQL/MetricsQL expression is used. Other available types are "graphite" and "vlogs".
+# See https://docs.victoriametrics.com/victoriametrics/vmalert/#integrations
 expr: <string>
 
 # Alerts are considered firing once they have been returned for this long.
@@ -303,7 +285,44 @@ annotations:
   [ <labelname>: <tmpl_string> ]
 ```
 
-#### Templating
+#### Recording rules
+
+The syntax for recording rules is the following:
+
+```yaml
+# The name of the time series to output to. Must be a valid metric name.
+record: <string>
+
+# The expression to evaluate. The expression language depends on the type value.
+# By default, PromQL/MetricsQL expression is used. Other available types are "graphite" and "vlogs".
+# See https://docs.victoriametrics.com/victoriametrics/vmalert/#integrations
+expr: <string>
+
+# Labels to add or overwrite labels from other external label sources, such as group labels, before storing the result.
+#
+# In case of conflicts, original labels are kept with prefix `exported_`.
+# As a special case, specifying a label with an empty string value removes the label from the result if it exists 
+# in the original query result; otherwise, it is ignored.
+#
+# Labels do not support templating in https://docs.victoriametrics.com/victoriametrics/vmalert/#templating due to cardinality concerns. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8171.
+labels:
+  [ <labelname>: <labelvalue> ]
+
+# Whether to print debug information into logs.
+# Information includes requests sent to the datasource.
+# information - it will be printed to logs.
+# Logs are printed with INFO level, so make sure that -loggerLevel=INFO to see the output.
+[ debug: <bool> | default = false ]
+
+# Defines the number of rule updates entries stored in memory
+# and available for view on rule Details page.
+# Overrides `rule.updateEntriesLimit` value for this specific rule.
+[ update_entries_limit: <integer> | default 0 ]
+```
+
+For recording rules to work `-remoteWrite.url` must be specified.
+
+## Templating
 
 It is allowed to use [Go templating](https://golang.org/pkg/text/template/) in annotations and labels(with limited support) to format data, iterate over
 or execute expressions.
@@ -325,7 +344,7 @@ The following variables are available in templating:
 
 Additionally, `vmalert` provides some extra templating functions listed in [template functions](#template-functions) and [reusable templates](#reusable-templates).
 
-#### Template functions
+### Template functions
 
 `vmalert` provides the following template functions, which can be used during [templating](#templating):
 
@@ -349,7 +368,7 @@ Additionally, `vmalert` provides some extra templating functions listed in [temp
 * `parseDurationTime` - parses the input string into [time.Duration](https://pkg.go.dev/time#Duration).
 * `pathEscape` - escapes the input string, so it can be safely put inside path part of URL.
 * `pathPrefix` - returns the path part of the `-external.url` command-line flag.
-* `query` - executes the [MetricsQL](https://docs.victoriametrics.com/victoriametrics/metricsql/) query against `-datasource.url` and returns the query result.
+* `query` - executes query against `-datasource.url` and returns the query result.
   For example, `{{ query "sort_desc(process_resident_memory_bytes)" | first | value }}` executes the `sort_desc(process_resident_memory_bytes)`
   query at `-datasource.url` and returns the first result.
 * `queryEscape` - escapes the input string, so it can be safely put inside [query arg](https://en.wikipedia.org/wiki/Percent-encoding) part of URL.
@@ -369,7 +388,7 @@ Additionally, `vmalert` provides some extra templating functions listed in [temp
 * `toUpper` - converts all the chars in the input string to uppercase.
 * `value` - returns the numeric value from the input query result.
 
-#### Reusable templates
+### Reusable templates
 
 Like in Alertmanager you can define [reusable templates](https://prometheus.io/docs/prometheus/latest/configuration/template_examples/#defining-reusable-templates)
 to share same templates across annotations. Just define the templates in a file and
@@ -407,44 +426,8 @@ groups:
 The `-rule.templates` flag supports wildcards so multiple files with templates can be loaded.
 The content of `-rule.templates` can be also [hot reloaded](#hot-config-reload).
 
-#### Recording rules
 
-The syntax for recording rules is following:
-
-```yaml
-# The name of the time series to output to. Must be a valid metric name.
-record: <string>
-
-# The expression to evaluate. The expression language depends on the type value.
-# By default, MetricsQL expression is used. If group.type="graphite", then the expression
-# must contain valid Graphite expression.
-expr: <string>
-
-# Labels to add or overwrite labels from other external label sources, such as group labels, before storing the result.
-#
-# In case of conflicts, original labels are kept with prefix `exported_`.
-# As a special case, specifying a label with an empty string value removes the label from the result if it exists 
-# in the original query result; otherwise, it is ignored.
-#
-# Labels do not support templating in https://docs.victoriametrics.com/victoriametrics/vmalert/#templating due to cardinality concerns. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8171.
-labels:
-  [ <labelname>: <labelvalue> ]
-
-# Whether to print debug information into logs.
-# Information includes requests sent to the datasource.
-# information - it will be printed to logs.
-# Logs are printed with INFO level, so make sure that -loggerLevel=INFO to see the output.
-[ debug: <bool> | default = false ]
-
-# Defines the number of rule's updates entries stored in memory
-# and available for view on rule's Details page.
-# Overrides `rule.updateEntriesLimit` value for this specific rule.
-[ update_entries_limit: <integer> | default 0 ]
-```
-
-For recording rules to work `-remoteWrite.url` must be specified.
-
-### Alerts state on restarts
+## Alerts state on restarts
 
 `vmalert` holds alerts state in the memory. Restart of the `vmalert` process will reset the state of all active alerts
 in the memory. To prevent `vmalert` from losing the state on restarts configure it to persist the state
@@ -465,7 +448,7 @@ in configured `-remoteRead.url`, weren't updated in the last `1h` (controlled by
 or received state doesn't match current `vmalert` rules configuration. `vmalert` marks successfully restored rules
 with `restored` label in [web UI](#web).
 
-### Link to alert source
+## Link to alert source
 
 Alerting notifications sent by vmalert always contain a `source` link. By default, the link format
 is the following `http://<vmalert-addr>/vmalert/alert?group_id=<group_id>&alert_id=<alert_id>`. On click, it opens
@@ -501,7 +484,9 @@ In addition to `source` link, some extra links could be added to alert's [annota
 field. See [how we use them](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/9751ea10983d42068487624849cac7ad6fd7e1d8/deployment/docker/rules/alerts-cluster.yml#L44)
 to link alerting rule and the corresponding panel on Grafana dashboard.
 
-### Multitenancy
+## Multitenancy
+
+> See how to use [multitenancy in rules for VictoriaLogs](https://docs.victoriametrics.com/victorialogs/vmalert/#how-to-use-multitenancy-in-rules).
 
 There are the following approaches exist for alerting and recording rules across
 [multiple tenants](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#multitenancy):
@@ -559,7 +544,7 @@ The enterprise version of vmalert is available in `vmutils-*-enterprise.tar.gz`
 at [release page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/latest) and in `*-enterprise`
 tags at [Docker Hub](https://hub.docker.com/r/victoriametrics/vmalert/tags) and [Quay](https://quay.io/repository/victoriametrics/vmalert?tab=tags).
 
-### Reading rules from object storage
+## Reading rules from object storage
 
 [Enterprise version](https://docs.victoriametrics.com/victoriametrics/enterprise/) of `vmalert` may read alerting and recording rules
 from object storage:
@@ -578,7 +563,7 @@ The following [command-line flags](#flags) can be used for fine-tuning access to
 * `-s3.customEndpoint` - custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set.
 * `-s3.forcePathStyle` - prefixing endpoint with bucket name when set false, true by default.
 
-### Topology examples
+## Topology examples
 
 The following sections are showing how `vmalert` may be used and configured
 for different scenarios.
@@ -589,7 +574,7 @@ Please note, not all flags in examples are required:
   you have recording rules or want to store [alerts state](#alerts-state-on-restarts) on `vmalert` restarts;
 * `-notifier.url` is optional and is needed only if you have alerting rules.
 
-#### Single-node VictoriaMetrics
+### Single-node VictoriaMetrics
 
 The simplest configuration where one single-node VM server is used for
 rules execution, storing recording rules results and alerts state.
@@ -607,7 +592,7 @@ rules execution, storing recording rules results and alerts state.
 ![vmalert single](vmalert_single.webp)
 {width="500"}
 
-#### Cluster VictoriaMetrics
+### Cluster VictoriaMetrics
 
 In [cluster mode](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/)
 VictoriaMetrics has separate components for writing and reading path:
@@ -630,7 +615,7 @@ Cluster mode could have multiple `vminsert` and `vmselect` components.
 In case when you want to spread the load on these components - add balancers before them and configure
 `vmalert` with balancer addresses. Please, see more about [VictoriaMetrics cluster architecture](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#architecture-overview).
 
-#### HA vmalert
+### HA vmalert
 
 For High Availability(HA) user can run multiple identically configured `vmalert` instances.
 It means all of them will execute the same rules, write state and results to
@@ -674,12 +659,12 @@ to ensure [high availability](https://github.com/prometheus/alertmanager#high-av
 This example uses single-node VM server for the sake of simplicity.
 Check how to replace it with [cluster VictoriaMetrics](#cluster-victoriametrics) if needed.
 
-#### Downsampling and aggregation via vmalert
+### Downsampling and aggregation via vmalert
 
 _Please note, [stream aggregation](https://docs.victoriametrics.com/victoriametrics/stream-aggregation/) might be more efficient
 for cases when downsampling or aggregation need to be applied **before data gets into the TSDB.**_
 
-`vmalert` can't modify existing data. But it can run arbitrary PromQL/MetricsQL queries
+`vmalert` can't modify existing data. But it can run arbitrary queries
 via [recording rules](#recording-rules) and backfill results to the configured `-remoteWrite.url`.
 This ability allows to aggregate data. For example, the following rule will calculate the average value for
 metric `http_requests` on the `5m` interval:
@@ -732,7 +717,7 @@ Flags `-remoteRead.url` and `-notifier.url` are omitted since we assume only rec
 
 See also [stream aggregation](https://docs.victoriametrics.com/victoriametrics/stream-aggregation/) and [downsampling](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#downsampling).
 
-#### Multiple remote writes
+### Multiple remote writes
 
 For persisting recording or alerting rule results `vmalert` requires `-remoteWrite.url` to be set.
 But this flag supports only one destination. To persist rule results to multiple destinations
@@ -746,7 +731,72 @@ Using `vmagent` as a proxy provides additional benefits such as
 [data persisting when storage is unreachable](https://docs.victoriametrics.com/victoriametrics/vmagent/#replication-and-high-availability),
 or time series modification via [relabeling](https://docs.victoriametrics.com/victoriametrics/relabeling/).
 
-### Web
+## Integrations
+
+vmalert can be integrated with different data sources for alerting and recording rules. But it deliberately allows
+configuring only one `datasource.url`. We recommend running separate instances of vmalert for each datasource type
+with the specified `-rule.defaultRuleType=<datasource_type>` command-line flag.
+
+###### VictoriaMetrics
+
+vmalert natively integrates with [VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/) for alerting and
+recording rules.
+
+###### VictoriaLogs
+
+vmalert integrates with [VictoriaLogs](https://docs.victoriametrics.com/victorialogs/) and allows configuring alerting and recording rules using [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/).
+Results of recording rules and alerting state should be persisted to the remote-write compatible storage, such as VictoriaMetrics.
+To enable VictoriaLogs compatibility set the `-rule.defaultRuleType=vlogs` command-line flag.
+
+See [this doc](https://docs.victoriametrics.com/victorialogs/vmalert/) for details.
+
+###### VictoriaTraces
+
+vmalert integrates with [VictoriaTraces](https://docs.victoriametrics.com/victoriatraces/) in exactly the same way as
+with [VictoriaLogs](https://docs.victoriametrics.com/victoriametrics/vmalert/#victorialogs).
+
+###### Graphite
+
+vmalert integrates with [Graphite Render API](https://graphite.readthedocs.io/en/stable/render_api.html) and allows configuring alerting and recording rules.
+During evaluation, vmalert will send requests to `<-datasource.url>/render?format=json`.
+To enable Graphite compatibility set the `-rule.defaultRuleType=graphite` command-line flag.
+
+Since VictoriaMetrics supports both Graphite and Prometheus APIs, it is possible to mix Graphite and VictoriaMetrics rules.
+On the group level, set the `type` field to specify to which datasource type it should belong: `prometheus` (MetricsQL) or `graphite` (GraphiteQL).
+When using vmalert with both `graphite` and `prometheus` rules configured against the cluster version of VictoriaMetrics, don't forget
+to set the `-datasource.appendTypePrefix` flag to `true`, so vmalert can adjust URL prefix automatically based on the query type.
+
+###### Prometheus
+
+vmalert uses [Prometheus HTTP API](https://prometheus.io/docs/prometheus/latest/querying/api/#http-api) for querying
+and [Prometheus Remote Write v1 protocol](https://prometheus.io/docs/specs/prw/remote_write_spec/) for persisting
+recording rules results and alerting state. Hence, it can be integrated with any Prometheus-compatible storage
+that supports these protocols.
+
+###### Grafana
+
+To proxy requests from [Grafana Alerting UI](https://grafana.com/docs/grafana/latest/alerting/) configure `-vmalert.proxyURL`
+on VictoriaMetrics [single-node](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#vmalert)
+or [vmselect in cluster version](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#vmalert).
+
+###### vmui
+
+To access rules UI through [vmui](https://docs.victoriametrics.com/victoriametrics/#vmui) configure `-vmalert.proxyURL`
+on VictoriaMetrics [single-node](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#vmalert)
+or [vmselect in cluster version](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#vmalert).
+
+###### vmanomaly
+
+See how to integrate vmalert with [VictoriaMetrics Anomaly Detection](https://docs.victoriametrics.com/anomaly-detection/)
+in the following [guide](https://docs.victoriametrics.com/anomaly-detection/guides/guide-vmanomaly-vmalert/).
+
+###### VictoriaMetrics Cloud
+
+For users of [VictoriaMetrics Cloud](https://console.victoriametrics.cloud/signUp?utm_source=website&utm_campaign=docs_vm_vmalert_config),
+many of the configuration steps (including highly available setup of `vmalert` for cluster deployments) are handled automatically.
+Please refer to the [VictoriaMetrics Cloud documentation](https://docs.victoriametrics.com/victoriametrics-cloud/alertmanager-setup-for-deployment/) for more details.
+
+## Web
 
 `vmalert` runs a web-server (`-httpListenAddr`) for serving metrics and alerts endpoints:
 
@@ -771,28 +821,11 @@ This may be used for better integration with Grafana unified alerting system. Se
 * [How to query vmalert from single-node VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#vmalert)
 * [How to query vmalert from VictoriaMetrics cluster](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#vmalert)
 
-## Graphite
-
-vmalert sends requests to `<-datasource.url>/render?format=json` during evaluation of alerting and recording rules
-if the corresponding group or rule contains `type: "graphite"` config option. It is expected that the `<-datasource.url>/render`
-implements [Graphite Render API](https://graphite.readthedocs.io/en/stable/render_api.html) for `format=json`.
-When using vmalert with both `graphite` and `prometheus` rules configured against cluster version of VM do not forget
-to set `-datasource.appendTypePrefix` flag to `true`, so vmalert can adjust URL prefix automatically based on the query type.
-
-## VictoriaLogs
-
-vmalert supports [VictoriaLogs](https://docs.victoriametrics.com/victorialogs/) as a datasource for writing alerting and recording rules using [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/). See [this doc](https://docs.victoriametrics.com/victorialogs/vmalert/) for details.
-
-## VictoriaTraces
-
-vmalert supports [VictoriaTraces](https://docs.victoriametrics.com/victoriatraces/) as a (`vlogs`) datasource for writing alerting and recording rules using [LogsQL](https://docs.victoriametrics.com/victorialogs/logsql/). See [this doc](https://docs.victoriametrics.com/victoriatraces/vmalert/) for details.
-
 ## Rules backfilling
 
 vmalert supports alerting and recording rules backfilling (aka `replay`). In replay mode vmalert
 can read the same rules configuration as normal, evaluate them on the given time range and backfill
-results via remote write to the configured storage. vmalert supports any PromQL/MetricsQL compatible
-data source for backfilling.
+results via remote write to the configured storage. vmalert supports only the `prometheus` datasource type for backfilling.
 
 Please note, that response caching may lead to unexpected results during and after backfilling process.
 In order to avoid this you need to reset cache contents or disable caching when using backfilling
@@ -852,13 +885,9 @@ vmalert respects `evaluationInterval` value set by flag or per-group during the
 vmalert automatically disables caching on VictoriaMetrics side by sending `nocache=1` param. It allows
 to prevent cache pollution and unwanted time range boundaries adjustment during backfilling.
 
-#### Recording rules
+Results of recording rules `replay` should match the results of normal rules evaluation.
 
-The result of recording rules `replay` should match with results of normal rules evaluation.
-
-#### Alerting rules
-
-The result of alerting rules `replay` is time series reflecting [alert's state](#alerts-state-on-restarts).
+Results of alerting rules `replay` are the time series reflecting the [state of the alert](#alerts-state-on-restarts).
 To see if `replayed` alert has fired in the past use the following PromQL/MetricsQL expression:
 
 ```

lib/storage/search.go

@@ -90,6 +90,77 @@ type MetricBlockRef struct {
 	BlockRef *BlockRef
 }
 
+// MetricBlock is a time series block for a single metric.
+type MetricBlock struct {
+	// MetricName is metric name for the given Block.
+	MetricName []byte
+
+	// Block is a block for the given MetricName
+	Block Block
+}
+
+// Marshal marshals MetricBlock to dst
+func (mb *MetricBlock) Marshal(dst []byte) []byte {
+	dst = encoding.MarshalBytes(dst, mb.MetricName)
+	return MarshalBlock(dst, &mb.Block)
+}
+
+// CopyFrom copies src to mb.
+func (mb *MetricBlock) CopyFrom(src *MetricBlock) {
+	mb.MetricName = append(mb.MetricName[:0], src.MetricName...)
+	mb.Block.CopyFrom(&src.Block)
+}
+
+// MarshalBlock marshals b to dst.
+//
+// b.MarshalData must be called on b before calling MarshalBlock.
+func MarshalBlock(dst []byte, b *Block) []byte {
+	dst = b.bh.Marshal(dst)
+	dst = encoding.MarshalBytes(dst, b.timestampsData)
+	dst = encoding.MarshalBytes(dst, b.valuesData)
+	return dst
+}
+
+// Unmarshal unmarshals MetricBlock from src
+func (mb *MetricBlock) Unmarshal(src []byte) ([]byte, error) {
+	mb.Block.Reset()
+	mn, nSize := encoding.UnmarshalBytes(src)
+	if nSize <= 0 {
+		return src, fmt.Errorf("cannot unmarshal MetricName")
+	}
+	src = src[nSize:]
+	mb.MetricName = append(mb.MetricName[:0], mn...)
+
+	return UnmarshalBlock(&mb.Block, src)
+}
+
+// UnmarshalBlock unmarshal Block from src to dst.
+//
+// dst.UnmarshalData isn't called on the block.
+func UnmarshalBlock(dst *Block, src []byte) ([]byte, error) {
+	tail, err := dst.bh.Unmarshal(src)
+	if err != nil {
+		return tail, fmt.Errorf("cannot unmarshal blockHeader: %w", err)
+	}
+	src = tail
+
+	tds, nSize := encoding.UnmarshalBytes(src)
+	if nSize <= 0 {
+		return tail, fmt.Errorf("cannot unmarshal timestampsData")
+	}
+	src = src[nSize:]
+	dst.timestampsData = append(dst.timestampsData[:0], tds...)
+
+	vd, nSize := encoding.UnmarshalBytes(src)
+	if nSize <= 0 {
+		return tail, fmt.Errorf("cannot unmarshal valuesData")
+	}
+	src = src[nSize:]
+	dst.valuesData = append(dst.valuesData[:0], vd...)
+
+	return src, nil
+}
+
 // Search is a search for time series.
 type Search struct {
 	// MetricBlockRef is updated with each Search.NextMetricBlock call.
@@ -290,6 +361,24 @@ func NewSearchQuery(start, end int64, tagFilterss [][]TagFilter, maxMetrics int)
 	}
 }
 
+// TenantToken represents a tenant (accountID, projectID) pair.
+type TenantToken struct {
+	AccountID uint32
+	ProjectID uint32
+}
+
+// String returns string representation of t.
+func (t *TenantToken) String() string {
+	return fmt.Sprintf("{accountID=%d, projectID=%d}", t.AccountID, t.ProjectID)
+}
+
+// Marshal appends marshaled t to dst and returns the result.
+func (t *TenantToken) Marshal(dst []byte) []byte {
+	dst = encoding.MarshalUint32(dst, t.AccountID)
+	dst = encoding.MarshalUint32(dst, t.ProjectID)
+	return dst
+}
+
 // TagFilter represents a single tag filter from SearchQuery.
 type TagFilter struct {
 	Key        []byte

lib/vminsertapi/api.go

@@ -0,0 +1,30 @@
+package vminsertapi
+
+import (
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage/metricsmetadata"
+)
+
+// RPCCall defines rpc call from vminsert to vmstorage
+type RPCCall struct {
+	Name          string
+	VersionedName string
+}
+
+var (
+	MetricRowsRpcCall = RPCCall{
+		Name:          "metric_rows",
+		VersionedName: "writeRows_v1",
+	}
+	MetricMetadataRpcCall = RPCCall{
+		Name:          "metricmetadata_rows",
+		VersionedName: "writeMetadata_v1",
+	}
+)
+
+// API must implement vminsert API.
+type API interface {
+	WriteRows(rows []storage.MetricRow) error
+	WriteMetadata(mrs []metricsmetadata.Row) error
+	IsReadOnly() bool
+}

lib/vmselectapi/api.go

@@ -0,0 +1,68 @@
+package vmselectapi
+
+import (
+	"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"
+)
+
+// API must implement vmselect API.
+type API interface {
+	// InitSearch initialize series search for the given sq.
+	//
+	// The returned BlockIterator must be closed with MustClose to free up resources when it is no longer needed.
+	InitSearch(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (BlockIterator, error)
+
+	// SearchMetricNames returns metric names matching the given sq.
+	SearchMetricNames(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) ([]string, error)
+
+	// LabelValues returns values for labelName label acorss series matching the given sq.
+	LabelValues(qt *querytracer.Tracer, sq *storage.SearchQuery, labelName string, maxLabelValues int, deadline uint64) ([]string, error)
+
+	// TagValueSuffixes returns tag value suffixes for the given args.
+	TagValueSuffixes(qt *querytracer.Tracer, accountID, projectID uint32, tr storage.TimeRange, tagKey, tagValuePrefix string, delimiter byte, maxSuffixes int, deadline uint64) ([]string, error)
+
+	// LabelNames returns lable names for series matching the given sq.
+	LabelNames(qt *querytracer.Tracer, sq *storage.SearchQuery, maxLableNames int, deadline uint64) ([]string, error)
+
+	// SeriesCount returns the number of series for the given (accountID, projectID).
+	SeriesCount(qt *querytracer.Tracer, accountID, projectID uint32, deadline uint64) (uint64, error)
+
+	// TSDBStatus returns tsdb status for the given sq.
+	TSDBStatus(qt *querytracer.Tracer, sq *storage.SearchQuery, focusLabel string, topN int, deadline uint64) (*storage.TSDBStatus, error)
+
+	// DeleteSeries deletes series matching the given sq.
+	DeleteSeries(qt *querytracer.Tracer, sq *storage.SearchQuery, deadline uint64) (int, error)
+
+	// RegisterMetricNames registers the given mrs in the storage.
+	RegisterMetricNames(qt *querytracer.Tracer, mrs []storage.MetricRow, deadline uint64) error
+
+	// Tenants returns list of tenants in the storage on the given tr.
+	Tenants(qt *querytracer.Tracer, tr storage.TimeRange, deadline uint64) ([]string, error)
+
+	// GetMetricNamesUsageStats returns statistics for metric names
+	GetMetricNamesUsageStats(qt *querytracer.Tracer, tt *storage.TenantToken, limit, le int, matchPattern string, deadline uint64) (metricnamestats.StatsResult, error)
+
+	// ResetMetricNamesUsageStats resets internal state of metric names tracker
+	ResetMetricNamesUsageStats(qt *querytracer.Tracer, deadline uint64) error
+
+	// GetMetadataRecords returns metrics metadata.
+	GetMetadataRecords(qt *querytracer.Tracer, tt *storage.TenantToken, limit int, metricName string, deadline uint64) ([]*metricsmetadata.Row, error)
+}
+
+// BlockIterator must iterate through series blocks found by VMSelect.InitSearch.
+//
+// MustClose must be called in order to free up allocated resources when BlockIterator is no longer needed.
+type BlockIterator interface {
+	// NextBlock marshals next storage.MetricBlock into dst.
+	//
+	// It returns true on success, false on error or if no blocks to read.
+	NextBlock(dst []byte) ([]byte, bool)
+
+	// MustClose frees up resources allocated by BlockIterator.
+	MustClose()
+
+	// Error returns the last error occurred in NextBlock(), which returns false.
+	Error() error
+}