Compare commits

..

1 Commits

Author SHA1 Message Date
Andrii Chubatiuk
60941a311f added compressor cleanup 2024-11-18 09:33:43 +02:00
51 changed files with 1243 additions and 1075 deletions

View File

@@ -9,7 +9,6 @@ import (
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
"github.com/VictoriaMetrics/metricsql"
)
@@ -49,7 +48,7 @@ Outer:
}
var expSamples []parsedSample
for _, s := range mt.ExpSamples {
expLb := []prompbmarshal.Label{}
expLb := datasource.Labels{}
if s.Labels != "" {
metricsqlExpr, err := metricsql.Parse(s.Labels)
if err != nil {
@@ -65,7 +64,7 @@ Outer:
}
if len(metricsqlMetricExpr.LabelFilterss) > 0 {
for _, l := range metricsqlMetricExpr.LabelFilterss[0] {
expLb = append(expLb, prompbmarshal.Label{
expLb = append(expLb, datasource.Label{
Name: l.Label,
Value: l.Value,
})

View File

@@ -46,8 +46,8 @@ const (
graphitePrefix = "/graphite"
)
func (c *Client) setGraphiteReqParams(r *http.Request, query string) {
if c.appendTypePrefix {
func (s *Client) setGraphiteReqParams(r *http.Request, query string) {
if s.appendTypePrefix {
r.URL.Path += graphitePrefix
}
r.URL.Path += graphitePath
@@ -58,7 +58,7 @@ func (c *Client) setGraphiteReqParams(r *http.Request, query string) {
q.Set("target", query)
q.Set("until", "now")
for k, vs := range c.extraParams {
for k, vs := range s.extraParams {
if q.Has(k) { // extraParams are prior to params in URL
q.Del(k)
}

View File

@@ -9,7 +9,6 @@ import (
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/valyala/fastjson"
)
@@ -82,14 +81,14 @@ func (pi *promInstant) Unmarshal(b []byte) error {
labels := metric.GetObject()
r := &pi.ms[i]
r.Labels = make([]prompbmarshal.Label, 0, labels.Len())
r.Labels = make([]Label, 0, labels.Len())
labels.Visit(func(key []byte, v *fastjson.Value) {
lv, errLocal := v.StringBytes()
if errLocal != nil {
err = fmt.Errorf("error when parsing label value %q: %s", v, errLocal)
return
}
r.Labels = append(r.Labels, prompbmarshal.Label{
r.Labels = append(r.Labels, Label{
Name: string(key),
Value: string(lv),
})
@@ -219,8 +218,8 @@ func parsePrometheusResponse(req *http.Request, resp *http.Response) (res Result
return res, nil
}
func (c *Client) setPrometheusInstantReqParams(r *http.Request, query string, timestamp time.Time) {
if c.appendTypePrefix {
func (s *Client) setPrometheusInstantReqParams(r *http.Request, query string, timestamp time.Time) {
if s.appendTypePrefix {
r.URL.Path += "/prometheus"
}
if !*disablePathAppend {
@@ -228,22 +227,22 @@ func (c *Client) setPrometheusInstantReqParams(r *http.Request, query string, ti
}
q := r.URL.Query()
q.Set("time", timestamp.Format(time.RFC3339))
if !*disableStepParam && c.evaluationInterval > 0 { // set step as evaluationInterval by default
if !*disableStepParam && s.evaluationInterval > 0 { // set step as evaluationInterval by default
// always convert to seconds to keep compatibility with older
// Prometheus versions. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1943
q.Set("step", fmt.Sprintf("%ds", int(c.evaluationInterval.Seconds())))
q.Set("step", fmt.Sprintf("%ds", int(s.evaluationInterval.Seconds())))
}
if !*disableStepParam && c.queryStep > 0 { // override step with user-specified value
if !*disableStepParam && s.queryStep > 0 { // override step with user-specified value
// always convert to seconds to keep compatibility with older
// Prometheus versions. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1943
q.Set("step", fmt.Sprintf("%ds", int(c.queryStep.Seconds())))
q.Set("step", fmt.Sprintf("%ds", int(s.queryStep.Seconds())))
}
r.URL.RawQuery = q.Encode()
c.setReqParams(r, query)
s.setReqParams(r, query)
}
func (c *Client) setPrometheusRangeReqParams(r *http.Request, query string, start, end time.Time) {
if c.appendTypePrefix {
func (s *Client) setPrometheusRangeReqParams(r *http.Request, query string, start, end time.Time) {
if s.appendTypePrefix {
r.URL.Path += "/prometheus"
}
if !*disablePathAppend {
@@ -252,11 +251,11 @@ func (c *Client) setPrometheusRangeReqParams(r *http.Request, query string, star
q := r.URL.Query()
q.Add("start", start.Format(time.RFC3339))
q.Add("end", end.Format(time.RFC3339))
if c.evaluationInterval > 0 { // set step as evaluationInterval by default
if s.evaluationInterval > 0 { // set step as evaluationInterval by default
// always convert to seconds to keep compatibility with older
// Prometheus versions. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1943
q.Set("step", fmt.Sprintf("%ds", int(c.evaluationInterval.Seconds())))
q.Set("step", fmt.Sprintf("%ds", int(s.evaluationInterval.Seconds())))
}
r.URL.RawQuery = q.Encode()
c.setReqParams(r, query)
s.setReqParams(r, query)
}

View File

@@ -14,7 +14,6 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/utils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promauth"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
var (
@@ -145,12 +144,12 @@ func TestVMInstantQuery(t *testing.T) {
}
expected := []Metric{
{
Labels: []prompbmarshal.Label{{Value: "vm_rows", Name: "__name__"}, {Value: "bar", Name: "foo"}},
Labels: []Label{{Value: "vm_rows", Name: "__name__"}, {Value: "bar", Name: "foo"}},
Timestamps: []int64{1583786142},
Values: []float64{13763},
},
{
Labels: []prompbmarshal.Label{{Value: "vm_requests", Name: "__name__"}, {Value: "baz", Name: "foo"}},
Labels: []Label{{Value: "vm_requests", Name: "__name__"}, {Value: "baz", Name: "foo"}},
Timestamps: []int64{1583786140},
Values: []float64{2000},
},
@@ -215,7 +214,7 @@ func TestVMInstantQuery(t *testing.T) {
}
exp := []Metric{
{
Labels: []prompbmarshal.Label{{Value: "constantLine(10)", Name: "name"}},
Labels: []Label{{Value: "constantLine(10)", Name: "name"}},
Timestamps: []int64{1611758403},
Values: []float64{10},
},
@@ -237,12 +236,12 @@ func TestVMInstantQuery(t *testing.T) {
}
expected = []Metric{
{
Labels: []prompbmarshal.Label{{Value: "total", Name: "stats_result"}, {Value: "bar", Name: "foo"}},
Labels: []Label{{Value: "total", Name: "stats_result"}, {Value: "bar", Name: "foo"}},
Timestamps: []int64{1583786142},
Values: []float64{13763},
},
{
Labels: []prompbmarshal.Label{{Value: "total", Name: "stats_result"}, {Value: "baz", Name: "foo"}},
Labels: []Label{{Value: "total", Name: "stats_result"}, {Value: "baz", Name: "foo"}},
Timestamps: []int64{1583786140},
Values: []float64{2000},
},
@@ -445,7 +444,7 @@ func TestVMRangeQuery(t *testing.T) {
t.Fatalf("expected 1 metric got %d in %+v", len(m), m)
}
expected := Metric{
Labels: []prompbmarshal.Label{{Value: "vm_rows", Name: "__name__"}},
Labels: []Label{{Value: "vm_rows", Name: "__name__"}},
Timestamps: []int64{1583786142},
Values: []float64{13763},
}
@@ -476,7 +475,7 @@ func TestVMRangeQuery(t *testing.T) {
t.Fatalf("expected 1 metric got %d in %+v", len(m), m)
}
expected = Metric{
Labels: []prompbmarshal.Label{{Value: "total", Name: "stats_result"}},
Labels: []Label{{Value: "total", Name: "stats_result"}},
Timestamps: []int64{1583786142},
Values: []float64{10},
}

View File

@@ -6,7 +6,7 @@ import (
"time"
)
func (c *Client) setVLogsInstantReqParams(r *http.Request, query string, timestamp time.Time) {
func (s *Client) setVLogsInstantReqParams(r *http.Request, query string, timestamp time.Time) {
// there is no type path prefix in victorialogs APIs right now, ignore appendTypePrefix.
if !*disablePathAppend {
r.URL.Path += "/select/logsql/stats_query"
@@ -16,15 +16,15 @@ func (c *Client) setVLogsInstantReqParams(r *http.Request, query string, timesta
q.Set("time", timestamp.Format(time.RFC3339))
// set the `start` and `end` params if applyIntervalAsTimeFilter is enabled(time filter is missing in the rule expr),
// so the query will be executed in time range [timestamp - evaluationInterval, timestamp].
if c.applyIntervalAsTimeFilter && c.evaluationInterval > 0 {
q.Set("start", timestamp.Add(-c.evaluationInterval).Format(time.RFC3339))
if s.applyIntervalAsTimeFilter && s.evaluationInterval > 0 {
q.Set("start", timestamp.Add(-s.evaluationInterval).Format(time.RFC3339))
q.Set("end", timestamp.Format(time.RFC3339))
}
r.URL.RawQuery = q.Encode()
c.setReqParams(r, query)
s.setReqParams(r, query)
}
func (c *Client) setVLogsRangeReqParams(r *http.Request, query string, start, end time.Time) {
func (s *Client) setVLogsRangeReqParams(r *http.Request, query string, start, end time.Time) {
// there is no type path prefix in victorialogs APIs right now, ignore appendTypePrefix.
if !*disablePathAppend {
r.URL.Path += "/select/logsql/stats_query_range"
@@ -33,11 +33,11 @@ func (c *Client) setVLogsRangeReqParams(r *http.Request, query string, start, en
q.Add("start", start.Format(time.RFC3339))
q.Add("end", end.Format(time.RFC3339))
// set step as evaluationInterval by default
if c.evaluationInterval > 0 {
q.Set("step", fmt.Sprintf("%ds", int(c.evaluationInterval.Seconds())))
if s.evaluationInterval > 0 {
q.Set("step", fmt.Sprintf("%ds", int(s.evaluationInterval.Seconds())))
}
r.URL.RawQuery = q.Encode()
c.setReqParams(r, query)
s.setReqParams(r, query)
}
func parseVLogsResponse(req *http.Request, resp *http.Response) (res Result, err error) {

View File

@@ -8,8 +8,6 @@ import (
"sort"
"strconv"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
// Querier interface wraps Query and QueryRange methods
@@ -57,7 +55,7 @@ type QuerierParams struct {
// Metric is the basic entity which should be return by datasource
type Metric struct {
Labels []prompbmarshal.Label
Labels []Label
Timestamps []int64
Values []float64
}
@@ -74,9 +72,22 @@ func (m *Metric) SetLabel(key, value string) {
m.AddLabel(key, value)
}
// SetLabels sets the given map as Metric labels
func (m *Metric) SetLabels(ls map[string]string) {
var i int
m.Labels = make([]Label, len(ls))
for k, v := range ls {
m.Labels[i] = Label{
Name: k,
Value: v,
}
i++
}
}
// AddLabel appends the given label to the label set
func (m *Metric) AddLabel(key, value string) {
m.Labels = append(m.Labels, prompbmarshal.Label{Name: key, Value: value})
m.Labels = append(m.Labels, Label{Name: key, Value: value})
}
// DelLabel deletes the given label from the label set
@@ -99,8 +110,14 @@ func (m *Metric) Label(key string) string {
return ""
}
// Label represents metric's label
type Label struct {
Name string
Value string
}
// Labels is collection of Label
type Labels []prompbmarshal.Label
type Labels []Label
func (ls Labels) Len() int { return len(ls) }
func (ls Labels) Swap(i, j int) { ls[i], ls[j] = ls[j], ls[i] }
@@ -155,7 +172,7 @@ func LabelCompare(a, b Labels) int {
// ConvertToLabels convert map to Labels
func ConvertToLabels(m map[string]string) (labelset Labels) {
for k, v := range m {
labelset = append(labelset, prompbmarshal.Label{
labelset = append(labelset, Label{
Name: k,
Value: v,
})

View File

@@ -3,8 +3,6 @@ package datasource
import (
"reflect"
"testing"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
func TestPromInstant_UnmarshalPositive(t *testing.T) {
@@ -23,7 +21,7 @@ func TestPromInstant_UnmarshalPositive(t *testing.T) {
f(`[{"metric":{"__name__":"up"},"value":[1583780000,"42"]}]`, []Metric{
{
Labels: []prompbmarshal.Label{{Name: "__name__", Value: "up"}},
Labels: []Label{{Name: "__name__", Value: "up"}},
Timestamps: []int64{1583780000},
Values: []float64{42},
},
@@ -33,17 +31,17 @@ func TestPromInstant_UnmarshalPositive(t *testing.T) {
{"metric":{"__name__":"foo"},"value":[1583780001,"7"]},
{"metric":{"__name__":"baz", "instance":"bar"},"value":[1583780002,"8"]}]`, []Metric{
{
Labels: []prompbmarshal.Label{{Name: "__name__", Value: "up"}},
Labels: []Label{{Name: "__name__", Value: "up"}},
Timestamps: []int64{1583780000},
Values: []float64{42},
},
{
Labels: []prompbmarshal.Label{{Name: "__name__", Value: "foo"}},
Labels: []Label{{Name: "__name__", Value: "foo"}},
Timestamps: []int64{1583780001},
Values: []float64{7},
},
{
Labels: []prompbmarshal.Label{{Name: "__name__", Value: "baz"}, {Name: "instance", Value: "bar"}},
Labels: []Label{{Name: "__name__", Value: "baz"}, {Name: "instance", Value: "bar"}},
Timestamps: []int64{1583780002},
Values: []float64{8},
},

View File

@@ -167,8 +167,14 @@ type tplData struct {
ExternalURL string
}
func templateAnnotation(dst io.Writer, text string, data tplData, tpl *textTpl.Template, execute bool) error {
tpl, err := tpl.Parse(text)
func templateAnnotation(dst io.Writer, text string, data tplData, tmpl *textTpl.Template, execute bool) error {
tpl, err := tmpl.Clone()
if err != nil {
return fmt.Errorf("error cloning template before parse annotation: %w", err)
}
// Clone() doesn't copy tpl Options, so we set them manually
tpl = tpl.Option("missingkey=zero")
tpl, err = tpl.Parse(text)
if err != nil {
return fmt.Errorf("error parsing annotation template: %w", err)
}

View File

@@ -33,7 +33,7 @@ func TestAlertExecTemplate(t *testing.T) {
qFn := func(_ string) ([]datasource.Metric, error) {
return []datasource.Metric{
{
Labels: []prompbmarshal.Label{
Labels: []datasource.Label{
{Name: "foo", Value: "bar"},
{Name: "baz", Value: "qux"},
},
@@ -41,7 +41,7 @@ func TestAlertExecTemplate(t *testing.T) {
Timestamps: []int64{1},
},
{
Labels: []prompbmarshal.Label{
Labels: []datasource.Label{
{Name: "foo", Value: "garply"},
{Name: "baz", Value: "fred"},
},

View File

@@ -14,10 +14,8 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/templates"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/utils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
)
// AlertingRule is basic alert entity
@@ -456,16 +454,13 @@ func (ar *AlertingRule) exec(ctx context.Context, ts time.Time, limit int) ([]pr
ar.logDebugf(ts, a, "created in state PENDING")
}
var numActivePending int
var tss []prompbmarshal.TimeSeries
for h, a := range ar.alerts {
// if alert wasn't updated in this iteration
// means it is resolved already
if _, ok := updated[h]; !ok {
if a.State == notifier.StatePending {
// alert was in Pending state - it is not active anymore
// add stale time series
tss = append(tss, pendingAlertStaleTimeSeries(a.Labels, ts.Unix(), true)...)
// alert was in Pending state - it is not
// active anymore
delete(ar.alerts, h)
ar.logDebugf(ts, a, "PENDING => DELETED: is absent in current evaluation round")
continue
@@ -483,9 +478,6 @@ func (ar *AlertingRule) exec(ctx context.Context, ts time.Time, limit int) ([]pr
if ts.Sub(a.KeepFiringSince) >= ar.KeepFiringFor {
a.State = notifier.StateInactive
a.ResolvedAt = ts
// add stale time series
tss = append(tss, firingAlertStaleTimeSeries(a.Labels, ts.Unix())...)
ar.logDebugf(ts, a, "FIRING => INACTIVE: is absent in current evaluation round")
continue
}
@@ -497,10 +489,6 @@ func (ar *AlertingRule) exec(ctx context.Context, ts time.Time, limit int) ([]pr
a.State = notifier.StateFiring
a.Start = ts
alertsFired.Inc()
if ar.For > 0 {
// add stale time series
tss = append(tss, pendingAlertStaleTimeSeries(a.Labels, ts.Unix(), false)...)
}
ar.logDebugf(ts, a, "PENDING => FIRING: %s since becoming active at %v", ts.Sub(a.ActiveAt), a.ActiveAt)
}
}
@@ -509,7 +497,7 @@ func (ar *AlertingRule) exec(ctx context.Context, ts time.Time, limit int) ([]pr
curState.Err = fmt.Errorf("exec exceeded limit of %d with %d alerts", limit, numActivePending)
return nil, curState.Err
}
return append(tss, ar.toTimeSeries(ts.Unix())...), nil
return ar.toTimeSeries(ts.Unix()), nil
}
func (ar *AlertingRule) expandTemplates(m datasource.Metric, qFn templates.QueryFn, ts time.Time) (*labelSet, map[string]string, error) {
@@ -534,7 +522,6 @@ func (ar *AlertingRule) expandTemplates(m datasource.Metric, qFn templates.Query
return ls, as, nil
}
// toTimeSeries creates `ALERTS` and `ALERTS_FOR_STATE` for active alerts
func (ar *AlertingRule) toTimeSeries(timestamp int64) []prompbmarshal.TimeSeries {
var tss []prompbmarshal.TimeSeries
for _, a := range ar.alerts {
@@ -614,83 +601,26 @@ func (ar *AlertingRule) alertToTimeSeries(a *notifier.Alert, timestamp int64) []
}
func alertToTimeSeries(a *notifier.Alert, timestamp int64) prompbmarshal.TimeSeries {
var labels []prompbmarshal.Label
labels := make(map[string]string)
for k, v := range a.Labels {
labels = append(labels, prompbmarshal.Label{
Name: k,
Value: v,
})
}
// __name__ already been dropped, no need to check duplication
labels = append(labels, prompbmarshal.Label{Name: "__name__", Value: alertMetricName})
if ol := promrelabel.GetLabelByName(labels, alertStateLabel); ol != nil {
ol.Value = a.State.String()
} else {
labels = append(labels, prompbmarshal.Label{Name: alertStateLabel, Value: a.State.String()})
labels[k] = v
}
labels["__name__"] = alertMetricName
labels[alertStateLabel] = a.State.String()
return newTimeSeries([]float64{1}, []int64{timestamp}, labels)
}
// alertForToTimeSeries returns a time series that represents
// alertForToTimeSeries returns a timeseries that represents
// state of active alerts, where value is time when alert become active
func alertForToTimeSeries(a *notifier.Alert, timestamp int64) prompbmarshal.TimeSeries {
var labels []prompbmarshal.Label
labels := make(map[string]string)
for k, v := range a.Labels {
labels = append(labels, prompbmarshal.Label{
Name: k,
Value: v,
})
labels[k] = v
}
// __name__ already been dropped, no need to check duplication
labels = append(labels, prompbmarshal.Label{Name: "__name__", Value: alertForStateMetricName})
labels["__name__"] = alertForStateMetricName
return newTimeSeries([]float64{float64(a.ActiveAt.Unix())}, []int64{timestamp}, labels)
}
// pendingAlertStaleTimeSeries returns stale `ALERTS` and `ALERTS_FOR_STATE` time series
// for alerts which changed their state from Pending to Inactive or Firing.
func pendingAlertStaleTimeSeries(ls map[string]string, timestamp int64, includeAlertForState bool) []prompbmarshal.TimeSeries {
var result []prompbmarshal.TimeSeries
var baseLabels []prompbmarshal.Label
for k, v := range ls {
baseLabels = append(baseLabels, prompbmarshal.Label{
Name: k,
Value: v,
})
}
// __name__ already been dropped, no need to check duplication
alertsLabels := append(baseLabels, prompbmarshal.Label{Name: "__name__", Value: alertMetricName})
alertsLabels = append(alertsLabels, prompbmarshal.Label{Name: alertStateLabel, Value: notifier.StatePending.String()})
result = append(result, newTimeSeries([]float64{decimal.StaleNaN}, []int64{timestamp}, alertsLabels))
if includeAlertForState {
alertsForStateLabels := append(baseLabels, prompbmarshal.Label{Name: "__name__", Value: alertForStateMetricName})
result = append(result, newTimeSeries([]float64{decimal.StaleNaN}, []int64{timestamp}, alertsForStateLabels))
}
return result
}
// firingAlertStaleTimeSeries returns stale `ALERTS` and `ALERTS_FOR_STATE` time series
// for alerts which changed their state from Firing to Inactive.
func firingAlertStaleTimeSeries(ls map[string]string, timestamp int64) []prompbmarshal.TimeSeries {
var baseLabels []prompbmarshal.Label
for k, v := range ls {
baseLabels = append(baseLabels, prompbmarshal.Label{
Name: k,
Value: v,
})
}
// __name__ already been dropped, no need to check duplication
alertsLabels := append(baseLabels, prompbmarshal.Label{Name: "__name__", Value: alertMetricName})
alertsLabels = append(alertsLabels, prompbmarshal.Label{Name: alertStateLabel, Value: notifier.StateFiring.String()})
alertsForStateLabels := append(baseLabels, prompbmarshal.Label{Name: "__name__", Value: alertForStateMetricName})
return []prompbmarshal.TimeSeries{
newTimeSeries([]float64{decimal.StaleNaN}, []int64{timestamp}, alertsLabels),
newTimeSeries([]float64{decimal.StaleNaN}, []int64{timestamp}, alertsForStateLabels),
}
}
// restore restores the value of ActiveAt field for active alerts,
// based on previously written time series `alertForStateMetricName`.
// Only rules with For > 0 can be restored.

View File

@@ -15,7 +15,6 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/utils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
)
@@ -29,7 +28,7 @@ func TestAlertingRuleToTimeSeries(t *testing.T) {
rule.alerts[alert.ID] = alert
tss := rule.toTimeSeries(timestamp.Unix())
if err := compareTimeSeries(t, tssExpected, tss); err != nil {
t.Fatalf("timeseries mismatch for rule %q: %s", rule.Name, err)
t.Fatalf("timeseries mismatch: %s", err)
}
}
@@ -37,23 +36,14 @@ func TestAlertingRuleToTimeSeries(t *testing.T) {
State: notifier.StateFiring,
ActiveAt: timestamp.Add(time.Second),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: alertMetricName,
},
{
Name: alertStateLabel,
Value: notifier.StateFiring.String(),
},
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": alertMetricName,
alertStateLabel: notifier.StateFiring.String(),
}),
newTimeSeries([]float64{float64(timestamp.Add(time.Second).Unix())},
[]int64{timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: alertForStateMetricName,
},
map[string]string{
"__name__": alertForStateMetricName,
}),
})
@@ -64,40 +54,18 @@ func TestAlertingRuleToTimeSeries(t *testing.T) {
"instance": "bar",
},
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: alertMetricName,
},
{
Name: alertStateLabel,
Value: notifier.StateFiring.String(),
},
{
Name: "job",
Value: "foo",
},
{
Name: "instance",
Value: "bar",
},
}),
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": alertMetricName,
alertStateLabel: notifier.StateFiring.String(),
"job": "foo",
"instance": "bar",
}),
newTimeSeries([]float64{float64(timestamp.Add(time.Second).Unix())},
[]int64{timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: alertForStateMetricName,
},
{
Name: "job",
Value: "foo",
},
{
Name: "instance",
Value: "bar",
},
map[string]string{
"__name__": alertForStateMetricName,
"job": "foo",
"instance": "bar",
}),
})
@@ -105,29 +73,18 @@ func TestAlertingRuleToTimeSeries(t *testing.T) {
State: notifier.StateFiring, ActiveAt: timestamp.Add(time.Second),
Labels: map[string]string{
alertStateLabel: "foo",
"__name__": "bar",
},
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: alertMetricName,
},
{
Name: alertStateLabel,
Value: notifier.StateFiring.String(),
},
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": alertMetricName,
alertStateLabel: notifier.StateFiring.String(),
}),
newTimeSeries([]float64{float64(timestamp.Add(time.Second).Unix())},
[]int64{timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: alertForStateMetricName,
},
{
Name: alertStateLabel,
Value: "foo",
},
map[string]string{
"__name__": alertForStateMetricName,
alertStateLabel: "foo",
}),
})
@@ -135,23 +92,14 @@ func TestAlertingRuleToTimeSeries(t *testing.T) {
State: notifier.StateFiring,
ActiveAt: timestamp.Add(time.Second),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: alertMetricName,
},
{
Name: alertStateLabel,
Value: notifier.StateFiring.String(),
},
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": alertMetricName,
alertStateLabel: notifier.StateFiring.String(),
}),
newTimeSeries([]float64{float64(timestamp.Add(time.Second).Unix())},
[]int64{timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: alertForStateMetricName,
},
map[string]string{
"__name__": alertForStateMetricName,
}),
})
@@ -159,21 +107,12 @@ func TestAlertingRuleToTimeSeries(t *testing.T) {
State: notifier.StatePending,
ActiveAt: timestamp.Add(time.Second),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: alertMetricName,
},
{
Name: alertStateLabel,
Value: notifier.StatePending.String(),
},
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": alertMetricName,
alertStateLabel: notifier.StatePending.String(),
}),
newTimeSeries([]float64{float64(timestamp.Add(time.Second).Unix())}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: alertForStateMetricName,
},
newTimeSeries([]float64{float64(timestamp.Add(time.Second).Unix())}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": alertForStateMetricName,
}),
})
}
@@ -185,9 +124,7 @@ func TestAlertingRule_Exec(t *testing.T) {
alert *notifier.Alert
}
ts, _ := time.Parse(time.RFC3339, "2024-10-29T00:00:00Z")
f := func(rule *AlertingRule, steps [][]datasource.Metric, alertsExpected map[int][]testAlert, tssExpected map[int][]prompbmarshal.TimeSeries) {
f := func(rule *AlertingRule, steps [][]datasource.Metric, alertsExpected map[int][]testAlert) {
t.Helper()
fq := &datasource.FakeQuerier{}
@@ -197,19 +134,13 @@ func TestAlertingRule_Exec(t *testing.T) {
Name: "TestRule_Exec",
}
rule.GroupID = fakeGroup.ID()
ts := time.Now()
for i, step := range steps {
fq.Reset()
fq.Add(step...)
tss, err := rule.exec(context.TODO(), ts, 0)
if err != nil {
if _, err := rule.exec(context.TODO(), ts, 0); err != nil {
t.Fatalf("unexpected error: %s", err)
}
// check generate time series
if _, ok := tssExpected[i]; ok {
if err := compareTimeSeries(t, tssExpected[i], tss); err != nil {
t.Fatalf("generated time series mismatch for rule %q in step %d: %s", rule.Name, i, err)
}
}
// shift the execution timestamp before the next iteration
ts = ts.Add(defaultStep)
@@ -243,21 +174,13 @@ func TestAlertingRule_Exec(t *testing.T) {
}
}
f(newTestAlertingRule("empty", 0), [][]datasource.Metric{}, nil, nil)
f(newTestAlertingRule("empty", 0), [][]datasource.Metric{}, nil)
f(newTestAlertingRule("empty_labels", 0), [][]datasource.Metric{
f(newTestAlertingRule("empty labels", 0), [][]datasource.Metric{
{datasource.Metric{Values: []float64{1}, Timestamps: []int64{1}}},
}, map[int][]testAlert{
0: {{alert: &notifier.Alert{State: notifier.StateFiring}}},
},
map[int][]prompbmarshal.TimeSeries{
0: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "empty_labels"}, {Name: "alertstate", Value: "firing"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "empty_labels"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Unix()), Timestamp: ts.UnixNano() / 1e6}}},
},
})
})
f(newTestAlertingRule("single-firing=>inactive=>firing=>inactive=>inactive", 0), [][]datasource.Metric{
{metricWithLabels(t, "name", "foo")},
@@ -271,25 +194,6 @@ func TestAlertingRule_Exec(t *testing.T) {
2: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
3: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}}},
4: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}}},
}, map[int][]prompbmarshal.TimeSeries{
0: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "single-firing=>inactive=>firing=>inactive=>inactive"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "single-firing=>inactive=>firing=>inactive=>inactive"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Unix()), Timestamp: ts.UnixNano() / 1e6}}},
},
1: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "single-firing=>inactive=>firing=>inactive=>inactive"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "single-firing=>inactive=>firing=>inactive=>inactive"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
},
2: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "single-firing=>inactive=>firing=>inactive=>inactive"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "single-firing=>inactive=>firing=>inactive=>inactive"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Add(2 * defaultStep).Unix()), Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
},
})
f(newTestAlertingRule("single-firing=>inactive=>firing=>inactive=>inactive=>firing", 0), [][]datasource.Metric{
@@ -306,7 +210,7 @@ func TestAlertingRule_Exec(t *testing.T) {
3: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}}},
4: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}}},
5: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
}, nil)
})
f(newTestAlertingRule("multiple-firing", 0), [][]datasource.Metric{
{
@@ -320,7 +224,7 @@ func TestAlertingRule_Exec(t *testing.T) {
{labels: []string{"name", "foo1"}, alert: &notifier.Alert{State: notifier.StateFiring}},
{labels: []string{"name", "foo2"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
}, nil)
})
// 1: fire first alert
// 2: fire second alert, set first inactive
@@ -329,57 +233,27 @@ func TestAlertingRule_Exec(t *testing.T) {
{metricWithLabels(t, "name", "foo")},
{metricWithLabels(t, "name", "foo1")},
{metricWithLabels(t, "name", "foo2")},
}, map[int][]testAlert{
0: {
{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
1: {
{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}},
{labels: []string{"name", "foo1"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
2: {
{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}},
{labels: []string{"name", "foo1"}, alert: &notifier.Alert{State: notifier.StateInactive}},
{labels: []string{"name", "foo2"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
}, map[int][]prompbmarshal.TimeSeries{
0: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Unix()), Timestamp: ts.UnixNano() / 1e6}}},
},
1: {
// stale time series for foo, `firing -> inactive`
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
// new time series for foo1
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo1"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "name", Value: "foo1"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Add(defaultStep).Unix()), Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
},
2: {
// stale time series for foo1
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo1"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "name", Value: "foo1"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
// new time series for foo2
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo2"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "multiple-steps-firing"}, {Name: "name", Value: "foo2"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Add(2 * defaultStep).Unix()), Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
},
})
},
map[int][]testAlert{
0: {
{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
1: {
{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}},
{labels: []string{"name", "foo1"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
2: {
{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}},
{labels: []string{"name", "foo1"}, alert: &notifier.Alert{State: notifier.StateInactive}},
{labels: []string{"name", "foo2"}, alert: &notifier.Alert{State: notifier.StateFiring}},
},
})
f(newTestAlertingRule("for-pending", time.Minute), [][]datasource.Metric{
{metricWithLabels(t, "name", "foo")},
}, map[int][]testAlert{
0: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StatePending}}},
}, nil)
})
f(newTestAlertingRule("for-fired", defaultStep), [][]datasource.Metric{
{metricWithLabels(t, "name", "foo")},
@@ -387,22 +261,6 @@ func TestAlertingRule_Exec(t *testing.T) {
}, map[int][]testAlert{
0: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StatePending}}},
1: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
}, map[int][]prompbmarshal.TimeSeries{
0: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "for-fired"}, {Name: "alertstate", Value: "pending"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "for-fired"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Unix()), Timestamp: ts.UnixNano() / 1e6}}},
},
1: {
// stale time series for `pending -> firing`
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "for-fired"}, {Name: "alertstate", Value: "pending"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "for-fired"}, {Name: "alertstate", Value: "firing"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "for-fired"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Add(defaultStep).Unix()), Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
},
})
f(newTestAlertingRule("for-pending=>empty", time.Second), [][]datasource.Metric{
@@ -414,26 +272,6 @@ func TestAlertingRule_Exec(t *testing.T) {
0: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StatePending}}},
1: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StatePending}}},
2: {},
}, map[int][]prompbmarshal.TimeSeries{
0: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "for-pending=>empty"}, {Name: "alertstate", Value: "pending"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "for-pending=>empty"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Unix()), Timestamp: ts.UnixNano() / 1e6}}},
},
1: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "for-pending=>empty"}, {Name: "alertstate", Value: "pending"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: 1, Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "for-pending=>empty"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: float64(ts.Unix()), Timestamp: ts.Add(defaultStep).UnixNano() / 1e6}}},
},
// stale time series for `pending -> inactive`
2: {
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertMetricName}, {Name: "alertname", Value: "for-pending=>empty"}, {Name: "alertstate", Value: "pending"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
{Labels: []prompbmarshal.Label{{Name: "__name__", Value: alertForStateMetricName}, {Name: "alertname", Value: "for-pending=>empty"}, {Name: "name", Value: "foo"}},
Samples: []prompbmarshal.Sample{{Value: decimal.StaleNaN, Timestamp: ts.Add(2*defaultStep).UnixNano() / 1e6}}},
},
})
f(newTestAlertingRule("for-pending=>firing=>inactive=>pending=>firing", defaultStep), [][]datasource.Metric{
@@ -449,7 +287,7 @@ func TestAlertingRule_Exec(t *testing.T) {
2: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}}},
3: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StatePending}}},
4: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
}, nil)
})
f(newTestAlertingRuleWithCustomFields("for-pending=>firing=>keepfiring=>firing", defaultStep, 0, defaultStep, nil), [][]datasource.Metric{
{metricWithLabels(t, "name", "foo")},
@@ -462,7 +300,7 @@ func TestAlertingRule_Exec(t *testing.T) {
1: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
2: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
3: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
}, nil)
})
f(newTestAlertingRuleWithCustomFields("for-pending=>firing=>keepfiring=>keepfiring=>inactive=>pending=>firing", defaultStep, 0, 2*defaultStep, nil), [][]datasource.Metric{
{metricWithLabels(t, "name", "foo")},
@@ -483,7 +321,7 @@ func TestAlertingRule_Exec(t *testing.T) {
4: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateInactive}}},
5: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StatePending}}},
6: {{labels: []string{"name", "foo"}, alert: &notifier.Alert{State: notifier.StateFiring}}},
}, nil)
})
}
func TestAlertingRuleExecRange(t *testing.T) {
@@ -639,7 +477,7 @@ func TestAlertingRuleExecRange(t *testing.T) {
{Values: []float64{1, 1, 1}, Timestamps: []int64{1, 3, 5}},
{
Values: []float64{1, 1}, Timestamps: []int64{1, 5},
Labels: []prompbmarshal.Label{{Name: "foo", Value: "bar"}},
Labels: []datasource.Label{{Name: "foo", Value: "bar"}},
},
}, []*notifier.Alert{
{State: notifier.StatePending, ActiveAt: time.Unix(1, 0)},
@@ -685,7 +523,7 @@ func TestAlertingRuleExecRange(t *testing.T) {
{Values: []float64{1, 1}, Timestamps: []int64{1, 100}},
{
Values: []float64{1, 1}, Timestamps: []int64{1, 5},
Labels: []prompbmarshal.Label{{Name: "foo", Value: "bar"}},
Labels: []datasource.Label{{Name: "foo", Value: "bar"}},
},
}, []*notifier.Alert{
{
@@ -1209,7 +1047,7 @@ func newTestAlertingRuleWithCustomFields(name string, waitFor, evalInterval, kee
func TestAlertingRule_ToLabels(t *testing.T) {
metric := datasource.Metric{
Labels: []prompbmarshal.Label{
Labels: []datasource.Label{
{Name: "instance", Value: "0.0.0.0:8800"},
{Name: "group", Value: "vmalert"},
{Name: "alertname", Value: "ConfigurationReloadFailure"},

View File

@@ -8,9 +8,12 @@ import (
"fmt"
"hash/fnv"
"net/url"
"strconv"
"sync"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/cheggaaa/pb/v3"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
@@ -18,6 +21,7 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/remotewrite"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/utils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/metrics"
@@ -346,9 +350,10 @@ func (g *Group) Start(ctx context.Context, nts func() []notifier.Notifier, rw re
}
e := &executor{
Rw: rw,
Notifiers: nts,
notifierHeaders: g.NotifierHeaders,
Rw: rw,
Notifiers: nts,
notifierHeaders: g.NotifierHeaders,
previouslySentSeriesToRW: make(map[uint64]map[string][]prompbmarshal.Label),
}
g.infof("started")
@@ -421,6 +426,8 @@ func (g *Group) Start(ctx context.Context, nts func() []notifier.Notifier, rw re
continue
}
// ensure that staleness is tracked for existing rules only
e.purgeStaleSeries(g.Rules)
e.notifierHeaders = g.NotifierHeaders
g.mu.Unlock()
@@ -532,9 +539,10 @@ func (g *Group) Replay(start, end time.Time, rw remotewrite.RWClient, maxDataPoi
// ExecOnce evaluates all the rules under group for once with given timestamp.
func (g *Group) ExecOnce(ctx context.Context, nts func() []notifier.Notifier, rw remotewrite.RWClient, evalTS time.Time) chan error {
e := &executor{
Rw: rw,
Notifiers: nts,
notifierHeaders: g.NotifierHeaders,
Rw: rw,
Notifiers: nts,
notifierHeaders: g.NotifierHeaders,
previouslySentSeriesToRW: make(map[uint64]map[string][]prompbmarshal.Label),
}
if len(g.Rules) < 1 {
return nil
@@ -625,6 +633,13 @@ type executor struct {
notifierHeaders map[string]string
Rw remotewrite.RWClient
previouslySentSeriesToRWMu sync.Mutex
// previouslySentSeriesToRW stores series sent to RW on previous iteration
// map[ruleID]map[ruleLabels][]prompb.Label
// where `ruleID` is ID of the Rule within a Group
// and `ruleLabels` is []prompb.Label marshalled to a string
previouslySentSeriesToRW map[uint64]map[string][]prompbmarshal.Label
}
// execConcurrently executes rules concurrently if concurrency>1
@@ -691,6 +706,11 @@ func (e *executor) exec(ctx context.Context, r Rule, ts time.Time, resolveDurati
if err := pushToRW(tss); err != nil {
return err
}
staleSeries := e.getStaleSeries(r, tss, ts)
if err := pushToRW(staleSeries); err != nil {
return err
}
}
ar, ok := r.(*AlertingRule)
@@ -717,3 +737,79 @@ func (e *executor) exec(ctx context.Context, r Rule, ts time.Time, resolveDurati
wg.Wait()
return errGr.Err()
}
var bbPool bytesutil.ByteBufferPool
// getStaleSeries checks whether there are stale series from previously sent ones.
func (e *executor) getStaleSeries(r Rule, tss []prompbmarshal.TimeSeries, timestamp time.Time) []prompbmarshal.TimeSeries {
bb := bbPool.Get()
defer bbPool.Put(bb)
ruleLabels := make(map[string][]prompbmarshal.Label, len(tss))
for _, ts := range tss {
// convert labels to strings, so we can compare with previously sent series
bb.B = labelsToString(bb.B, ts.Labels)
ruleLabels[string(bb.B)] = ts.Labels
bb.Reset()
}
rID := r.ID()
var staleS []prompbmarshal.TimeSeries
// check whether there are series which disappeared and need to be marked as stale
e.previouslySentSeriesToRWMu.Lock()
for key, labels := range e.previouslySentSeriesToRW[rID] {
if _, ok := ruleLabels[key]; ok {
continue
}
// previously sent series are missing in current series, so we mark them as stale
ss := newTimeSeriesPB([]float64{decimal.StaleNaN}, []int64{timestamp.Unix()}, labels)
staleS = append(staleS, ss)
}
// set previous series to current
e.previouslySentSeriesToRW[rID] = ruleLabels
e.previouslySentSeriesToRWMu.Unlock()
return staleS
}
// purgeStaleSeries deletes references in tracked
// previouslySentSeriesToRW list to Rules which aren't present
// in the given activeRules list. The method is used when the list
// of loaded rules has changed and executor has to remove
// references to non-existing rules.
func (e *executor) purgeStaleSeries(activeRules []Rule) {
newPreviouslySentSeriesToRW := make(map[uint64]map[string][]prompbmarshal.Label)
e.previouslySentSeriesToRWMu.Lock()
for _, rule := range activeRules {
id := rule.ID()
prev, ok := e.previouslySentSeriesToRW[id]
if ok {
// keep previous series for staleness detection
newPreviouslySentSeriesToRW[id] = prev
}
}
e.previouslySentSeriesToRW = nil
e.previouslySentSeriesToRW = newPreviouslySentSeriesToRW
e.previouslySentSeriesToRWMu.Unlock()
}
func labelsToString(dst []byte, labels []prompbmarshal.Label) []byte {
dst = append(dst, '{')
for i, label := range labels {
if len(label.Name) == 0 {
dst = append(dst, "__name__"...)
} else {
dst = append(dst, label.Name...)
}
dst = append(dst, '=')
dst = strconv.AppendQuote(dst, label.Value)
if i < len(labels)-1 {
dst = append(dst, ',')
}
}
dst = append(dst, '}')
return dst
}

View File

@@ -5,6 +5,7 @@ import (
"fmt"
"math"
"os"
"reflect"
"sort"
"testing"
"time"
@@ -16,6 +17,8 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/remotewrite"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/templates"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
)
@@ -380,6 +383,153 @@ func TestGetResolveDuration(t *testing.T) {
f(2*time.Minute, 0, 1*time.Minute, 8*time.Minute)
}
func TestGetStaleSeries(t *testing.T) {
ts := time.Now()
e := &executor{
previouslySentSeriesToRW: make(map[uint64]map[string][]prompbmarshal.Label),
}
f := func(r Rule, labels, expLabels [][]prompbmarshal.Label) {
t.Helper()
var tss []prompbmarshal.TimeSeries
for _, l := range labels {
tss = append(tss, newTimeSeriesPB([]float64{1}, []int64{ts.Unix()}, l))
}
staleS := e.getStaleSeries(r, tss, ts)
if staleS == nil && expLabels == nil {
return
}
if len(staleS) != len(expLabels) {
t.Fatalf("expected to get %d stale series, got %d",
len(expLabels), len(staleS))
}
for i, exp := range expLabels {
got := staleS[i]
if !reflect.DeepEqual(exp, got.Labels) {
t.Fatalf("expected to get labels: \n%v;\ngot instead: \n%v",
exp, got.Labels)
}
if len(got.Samples) != 1 {
t.Fatalf("expected to have 1 sample; got %d", len(got.Samples))
}
if !decimal.IsStaleNaN(got.Samples[0].Value) {
t.Fatalf("expected sample value to be %v; got %v", decimal.StaleNaN, got.Samples[0].Value)
}
}
}
// warn: keep in mind, that executor holds the state, so sequence of f calls matters
// single series
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "foo")},
nil)
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "foo")},
nil)
f(&AlertingRule{RuleID: 1},
nil,
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "foo")})
f(&AlertingRule{RuleID: 1},
nil,
nil)
// multiple series
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{
toPromLabels(t, "__name__", "job:foo", "job", "foo"),
toPromLabels(t, "__name__", "job:foo", "job", "bar"),
},
nil)
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "bar")},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "foo")})
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "bar")},
nil)
f(&AlertingRule{RuleID: 1},
nil,
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "bar")})
// multiple rules and series
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{
toPromLabels(t, "__name__", "job:foo", "job", "foo"),
toPromLabels(t, "__name__", "job:foo", "job", "bar"),
},
nil)
f(&AlertingRule{RuleID: 2},
[][]prompbmarshal.Label{
toPromLabels(t, "__name__", "job:foo", "job", "foo"),
toPromLabels(t, "__name__", "job:foo", "job", "bar"),
},
nil)
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "bar")},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "foo")})
f(&AlertingRule{RuleID: 1},
[][]prompbmarshal.Label{toPromLabels(t, "__name__", "job:foo", "job", "bar")},
nil)
}
func TestPurgeStaleSeries(t *testing.T) {
ts := time.Now()
labels := toPromLabels(t, "__name__", "job:foo", "job", "foo")
tss := []prompbmarshal.TimeSeries{newTimeSeriesPB([]float64{1}, []int64{ts.Unix()}, labels)}
f := func(curRules, newRules, expStaleRules []Rule) {
t.Helper()
e := &executor{
previouslySentSeriesToRW: make(map[uint64]map[string][]prompbmarshal.Label),
}
// seed executor with series for
// current rules
for _, rule := range curRules {
e.getStaleSeries(rule, tss, ts)
}
e.purgeStaleSeries(newRules)
if len(e.previouslySentSeriesToRW) != len(expStaleRules) {
t.Fatalf("expected to get %d stale series, got %d",
len(expStaleRules), len(e.previouslySentSeriesToRW))
}
for _, exp := range expStaleRules {
if _, ok := e.previouslySentSeriesToRW[exp.ID()]; !ok {
t.Fatalf("expected to have rule %d; got nil instead", exp.ID())
}
}
}
f(nil, nil, nil)
f(
nil,
[]Rule{&AlertingRule{RuleID: 1}},
nil,
)
f(
[]Rule{&AlertingRule{RuleID: 1}},
nil,
nil,
)
f(
[]Rule{&AlertingRule{RuleID: 1}},
[]Rule{&AlertingRule{RuleID: 2}},
nil,
)
f(
[]Rule{&AlertingRule{RuleID: 1}, &AlertingRule{RuleID: 2}},
[]Rule{&AlertingRule{RuleID: 2}},
[]Rule{&AlertingRule{RuleID: 2}},
)
f(
[]Rule{&AlertingRule{RuleID: 1}, &AlertingRule{RuleID: 2}},
[]Rule{&AlertingRule{RuleID: 1}, &AlertingRule{RuleID: 2}},
[]Rule{&AlertingRule{RuleID: 1}, &AlertingRule{RuleID: 2}},
)
}
func TestFaultyNotifier(t *testing.T) {
fq := &datasource.FakeQuerier{}
fq.Add(metricWithValueAndLabels(t, 1, "__name__", "foo", "job", "bar"))
@@ -430,7 +580,8 @@ func TestFaultyRW(t *testing.T) {
}
e := &executor{
Rw: &remotewrite.Client{},
Rw: &remotewrite.Client{},
previouslySentSeriesToRW: make(map[uint64]map[string][]prompbmarshal.Label),
}
err := e.exec(context.Background(), r, time.Now(), 0, 10)

View File

@@ -0,0 +1,36 @@
package rule
import (
"fmt"
"testing"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
func BenchmarkGetStaleSeries(b *testing.B) {
ts := time.Now()
n := 100
payload := make([]prompbmarshal.TimeSeries, 0, n)
for i := 0; i < n; i++ {
s := fmt.Sprintf("%d", i)
labels := toPromLabels(b,
"__name__", "foo", ""+
"instance", s,
"job", s,
"state", s,
)
payload = append(payload, newTimeSeriesPB([]float64{1}, []int64{ts.Unix()}, labels))
}
e := &executor{
previouslySentSeriesToRW: make(map[uint64]map[string][]prompbmarshal.Label),
}
ar := &AlertingRule{RuleID: 1}
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
e.getStaleSeries(ar, payload, ts)
}
}

View File

@@ -3,17 +3,16 @@ package rule
import (
"context"
"fmt"
"sort"
"strings"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/utils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logstorage"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
)
// RecordingRule is a Rule that supposed
@@ -35,8 +34,6 @@ type RecordingRule struct {
// during evaluations
state *ruleState
lastEvaluation map[string]struct{}
metrics *recordingRuleMetrics
}
@@ -116,7 +113,7 @@ func (rr *RecordingRule) execRange(ctx context.Context, start, end time.Time) ([
var tss []prompbmarshal.TimeSeries
for _, s := range res.Data {
ts := rr.toTimeSeries(s)
key := stringifyLabels(ts.Labels)
key := stringifyLabels(ts)
if _, ok := duplicates[key]; ok {
return nil, fmt.Errorf("original metric %v; resulting labels %q: %w", s.Labels, key, errDuplicate)
}
@@ -158,47 +155,28 @@ func (rr *RecordingRule) exec(ctx context.Context, ts time.Time, limit int) ([]p
return nil, curState.Err
}
curEvaluation := make(map[string]struct{}, len(qMetrics))
lastEvaluation := rr.lastEvaluation
duplicates := make(map[string]struct{}, len(qMetrics))
var tss []prompbmarshal.TimeSeries
for _, r := range qMetrics {
ts := rr.toTimeSeries(r)
key := stringifyLabels(ts.Labels)
if _, ok := curEvaluation[key]; ok {
key := stringifyLabels(ts)
if _, ok := duplicates[key]; ok {
curState.Err = fmt.Errorf("original metric %v; resulting labels %q: %w", r, key, errDuplicate)
return nil, curState.Err
}
curEvaluation[key] = struct{}{}
delete(lastEvaluation, key)
duplicates[key] = struct{}{}
tss = append(tss, ts)
}
// check for stale time series
for k := range lastEvaluation {
tss = append(tss, prompbmarshal.TimeSeries{
Labels: stringToLabels(k),
Samples: []prompbmarshal.Sample{
{Value: decimal.StaleNaN, Timestamp: ts.UnixNano() / 1e6},
}})
}
rr.lastEvaluation = curEvaluation
return tss, nil
}
func stringToLabels(s string) []prompbmarshal.Label {
labels := strings.Split(s, ",")
rLabels := make([]prompbmarshal.Label, 0, len(labels))
for i := range labels {
if label := strings.Split(labels[i], "="); len(label) == 2 {
rLabels = append(rLabels, prompbmarshal.Label{
Name: label[0],
Value: label[1],
})
}
func stringifyLabels(ts prompbmarshal.TimeSeries) string {
labels := ts.Labels
if len(labels) > 1 {
sort.Slice(labels, func(i, j int) bool {
return labels[i].Name < labels[j].Name
})
}
return rLabels
}
func stringifyLabels(labels []prompbmarshal.Label) string {
b := strings.Builder{}
for i, l := range labels {
b.WriteString(l.Name)
@@ -212,27 +190,19 @@ func stringifyLabels(labels []prompbmarshal.Label) string {
}
func (rr *RecordingRule) toTimeSeries(m datasource.Metric) prompbmarshal.TimeSeries {
if preN := promrelabel.GetLabelByName(m.Labels, "__name__"); preN != nil {
preN.Value = rr.Name
} else {
m.Labels = append(m.Labels, prompbmarshal.Label{
Name: "__name__",
Value: rr.Name,
})
labels := make(map[string]string)
for _, l := range m.Labels {
labels[l.Name] = l.Value
}
for k := range rr.Labels {
prevLabel := promrelabel.GetLabelByName(m.Labels, k)
if prevLabel != nil && prevLabel.Value != rr.Labels[k] {
// Rename the prevLabel to "exported_" + label.Name
prevLabel.Name = fmt.Sprintf("exported_%s", prevLabel.Name)
labels["__name__"] = rr.Name
// override existing labels with configured ones
for k, v := range rr.Labels {
if _, ok := labels[k]; ok && labels[k] != v {
labels[fmt.Sprintf("exported_%s", k)] = labels[k]
}
m.Labels = append(m.Labels, prompbmarshal.Label{
Name: k,
Value: rr.Labels[k],
})
labels[k] = v
}
ts := newTimeSeries(m.Values, m.Timestamps, m.Labels)
return ts
return newTimeSeries(m.Values, m.Timestamps, labels)
}
// updateWith copies all significant fields.

View File

@@ -9,151 +9,59 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/utils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
func TestRecordingRule_Exec(t *testing.T) {
ts, _ := time.Parse(time.RFC3339, "2024-10-29T00:00:00Z")
const defaultStep = 5 * time.Millisecond
f := func(rule *RecordingRule, steps [][]datasource.Metric, tssExpected [][]prompbmarshal.TimeSeries) {
f := func(rule *RecordingRule, metrics []datasource.Metric, tssExpected []prompbmarshal.TimeSeries) {
t.Helper()
fq := &datasource.FakeQuerier{}
for i, step := range steps {
fq.Reset()
fq.Add(step...)
rule.q = fq
rule.state = &ruleState{
entries: make([]StateEntry, 10),
}
tss, err := rule.exec(context.TODO(), ts, 0)
if err != nil {
t.Fatalf("unexpected RecordingRule.exec error: %s", err)
}
if err := compareTimeSeries(t, tssExpected[i], tss); err != nil {
t.Fatalf("time series mismatch: %s", err)
}
ts = ts.Add(defaultStep)
fq.Add(metrics...)
rule.q = fq
rule.state = &ruleState{
entries: make([]StateEntry, 10),
}
tss, err := rule.exec(context.TODO(), time.Now(), 0)
if err != nil {
t.Fatalf("unexpected RecordingRule.exec error: %s", err)
}
if err := compareTimeSeries(t, tssExpected, tss); err != nil {
t.Fatalf("timeseries missmatch: %s", err)
}
}
timestamp := time.Now()
f(&RecordingRule{
Name: "foo",
}, [][]datasource.Metric{{
}, []datasource.Metric{
metricWithValueAndLabels(t, 10, "__name__", "bar"),
}}, [][]prompbmarshal.TimeSeries{{
newTimeSeries([]float64{10}, []int64{ts.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foo",
},
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{10}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "foo",
}),
}})
})
f(&RecordingRule{
Name: "foobarbaz",
}, [][]datasource.Metric{
{
metricWithValueAndLabels(t, 1, "__name__", "foo", "job", "foo"),
metricWithValueAndLabels(t, 2, "__name__", "bar", "job", "bar"),
metricWithValueAndLabels(t, 3, "__name__", "baz", "job", "baz"),
},
{
metricWithValueAndLabels(t, 10, "__name__", "foo", "job", "foo"),
},
{
metricWithValueAndLabels(t, 10, "__name__", "foo", "job", "bar"),
},
}, [][]prompbmarshal.TimeSeries{{
newTimeSeries([]float64{1}, []int64{ts.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "foo",
},
}, []datasource.Metric{
metricWithValueAndLabels(t, 1, "__name__", "foo", "job", "foo"),
metricWithValueAndLabels(t, 2, "__name__", "bar", "job", "bar"),
metricWithValueAndLabels(t, 3, "__name__", "baz", "job", "baz"),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "foobarbaz",
"job": "foo",
}),
newTimeSeries([]float64{2}, []int64{ts.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "bar",
},
newTimeSeries([]float64{2}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "foobarbaz",
"job": "bar",
}),
newTimeSeries([]float64{3}, []int64{ts.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "baz",
},
newTimeSeries([]float64{3}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "foobarbaz",
"job": "baz",
}),
},
{
newTimeSeries([]float64{10}, []int64{ts.Add(defaultStep).UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "foo",
},
}),
// other series are with NaN values
newTimeSeries([]float64{decimal.StaleNaN}, []int64{ts.Add(defaultStep).UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "bar",
},
}),
newTimeSeries([]float64{decimal.StaleNaN}, []int64{ts.Add(defaultStep).UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "baz",
},
}),
},
{
newTimeSeries([]float64{10}, []int64{ts.Add(2 * defaultStep).UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "bar",
},
}),
newTimeSeries([]float64{decimal.StaleNaN}, []int64{ts.Add(2 * defaultStep).UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "foo",
},
}),
},
})
f(&RecordingRule{
@@ -161,44 +69,22 @@ func TestRecordingRule_Exec(t *testing.T) {
Labels: map[string]string{
"source": "test",
},
}, [][]datasource.Metric{{
}, []datasource.Metric{
metricWithValueAndLabels(t, 2, "__name__", "foo", "job", "foo"),
metricWithValueAndLabels(t, 1, "__name__", "bar", "job", "bar", "source", "origin"),
}}, [][]prompbmarshal.TimeSeries{{
newTimeSeries([]float64{2}, []int64{ts.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "job:foo",
},
{
Name: "job",
Value: "foo",
},
{
Name: "source",
Value: "test",
},
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{2}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "job:foo",
"job": "foo",
"source": "test",
}),
newTimeSeries([]float64{1}, []int64{ts.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: "job:foo",
},
{
Name: "job",
Value: "bar",
},
{
Name: "source",
Value: "test",
},
{
Name: "exported_source",
Value: "origin",
},
}),
}})
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "job:foo",
"job": "bar",
"source": "test",
"exported_source": "origin",
}),
})
}
func TestRecordingRule_ExecRange(t *testing.T) {
@@ -224,13 +110,9 @@ func TestRecordingRule_ExecRange(t *testing.T) {
}, []datasource.Metric{
metricWithValuesAndLabels(t, []float64{10, 20, 30}, "__name__", "bar"),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{10, 20, 30}, []int64{timestamp.UnixNano(), timestamp.UnixNano(), timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: "foo",
},
}),
newTimeSeries([]float64{10, 20, 30}, []int64{timestamp.UnixNano(), timestamp.UnixNano(), timestamp.UnixNano()}, map[string]string{
"__name__": "foo",
}),
})
f(&RecordingRule{
@@ -240,36 +122,18 @@ func TestRecordingRule_ExecRange(t *testing.T) {
metricWithValuesAndLabels(t, []float64{2, 3}, "__name__", "bar", "job", "bar"),
metricWithValuesAndLabels(t, []float64{4, 5, 6}, "__name__", "baz", "job", "baz"),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "foo",
},
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "foobarbaz",
"job": "foo",
}),
newTimeSeries([]float64{2, 3}, []int64{timestamp.UnixNano(), timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "bar",
},
newTimeSeries([]float64{2, 3}, []int64{timestamp.UnixNano(), timestamp.UnixNano()}, map[string]string{
"__name__": "foobarbaz",
"job": "bar",
}),
newTimeSeries([]float64{4, 5, 6},
[]int64{timestamp.UnixNano(), timestamp.UnixNano(), timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "foobarbaz",
},
{
Name: "job",
Value: "baz",
},
[]int64{timestamp.UnixNano(), timestamp.UnixNano(), timestamp.UnixNano()}, map[string]string{
"__name__": "foobarbaz",
"job": "baz",
}),
})
@@ -282,35 +146,16 @@ func TestRecordingRule_ExecRange(t *testing.T) {
metricWithValueAndLabels(t, 2, "__name__", "foo", "job", "foo"),
metricWithValueAndLabels(t, 1, "__name__", "bar", "job", "bar"),
}, []prompbmarshal.TimeSeries{
newTimeSeries([]float64{2}, []int64{timestamp.UnixNano()}, []prompbmarshal.Label{
{
Name: "__name__",
Value: "job:foo",
},
{
Name: "job",
Value: "foo",
},
{
Name: "source",
Value: "test",
},
newTimeSeries([]float64{2}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "job:foo",
"job": "foo",
"source": "test",
}),
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()}, map[string]string{
"__name__": "job:foo",
"job": "bar",
"source": "test",
}),
newTimeSeries([]float64{1}, []int64{timestamp.UnixNano()},
[]prompbmarshal.Label{
{
Name: "__name__",
Value: "job:foo",
},
{
Name: "job",
Value: "bar",
},
{
Name: "source",
Value: "test",
},
}),
})
}

View File

@@ -8,7 +8,6 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/decimal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
@@ -88,7 +87,7 @@ func metricWithLabels(t *testing.T, labels ...string) datasource.Metric {
}
m := datasource.Metric{Values: []float64{1}, Timestamps: []int64{1}}
for i := 0; i < len(labels); i += 2 {
m.Labels = append(m.Labels, prompbmarshal.Label{
m.Labels = append(m.Labels, datasource.Label{
Name: labels[i],
Value: labels[i+1],
})
@@ -96,6 +95,21 @@ func metricWithLabels(t *testing.T, labels ...string) datasource.Metric {
return m
}
func toPromLabels(t testing.TB, labels ...string) []prompbmarshal.Label {
t.Helper()
if len(labels) == 0 || len(labels)%2 != 0 {
t.Fatalf("expected to get even number of labels")
}
var ls []prompbmarshal.Label
for i := 0; i < len(labels); i += 2 {
ls = append(ls, prompbmarshal.Label{
Name: labels[i],
Value: labels[i+1],
})
}
return ls
}
func compareTimeSeries(t *testing.T, a, b []prompbmarshal.TimeSeries) error {
t.Helper()
if len(a) != len(b) {
@@ -108,7 +122,7 @@ func compareTimeSeries(t *testing.T, a, b []prompbmarshal.TimeSeries) error {
}
for i, exp := range expTS.Samples {
got := gotTS.Samples[i]
if got.Value != exp.Value && (!decimal.IsStaleNaN(got.Value) || !decimal.IsStaleNaN(exp.Value)) {
if got.Value != exp.Value {
return fmt.Errorf("expected value %.2f; got %.2f", exp.Value, got.Value)
}
// timestamp validation isn't always correct for now.

View File

@@ -9,14 +9,10 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
)
// newTimeSeries first sorts given labels, then returns new time series.
func newTimeSeries(values []float64, timestamps []int64, labels []prompbmarshal.Label) prompbmarshal.TimeSeries {
promrelabel.SortLabels(labels)
func newTimeSeries(values []float64, timestamps []int64, labels map[string]string) prompbmarshal.TimeSeries {
ts := prompbmarshal.TimeSeries{
Labels: labels,
Samples: make([]prompbmarshal.Sample, len(values)),
}
for i := range values {
@@ -25,6 +21,34 @@ func newTimeSeries(values []float64, timestamps []int64, labels []prompbmarshal.
Timestamp: time.Unix(timestamps[i], 0).UnixNano() / 1e6,
}
}
keys := make([]string, 0, len(labels))
for k := range labels {
keys = append(keys, k)
}
sort.Strings(keys) // make order deterministic
for _, key := range keys {
ts.Labels = append(ts.Labels, prompbmarshal.Label{
Name: key,
Value: labels[key],
})
}
return ts
}
// newTimeSeriesPB creates prompbmarshal.TimeSeries with given
// values, timestamps and labels.
// It expects that labels are already sorted.
func newTimeSeriesPB(values []float64, timestamps []int64, labels []prompbmarshal.Label) prompbmarshal.TimeSeries {
ts := prompbmarshal.TimeSeries{
Samples: make([]prompbmarshal.Sample, len(values)),
}
for i := range values {
ts.Samples[i] = prompbmarshal.Sample{
Value: values[i],
Timestamp: time.Unix(timestamps[i], 0).UnixNano() / 1e6,
}
}
ts.Labels = labels
return ts
}

View File

@@ -169,8 +169,6 @@ func GetWithFuncs(funcs textTpl.FuncMap) (*textTpl.Template, error) {
if err != nil {
return nil, err
}
// Clone() doesn't copy tpl Options, so we set them manually
tmpl = tmpl.Option("missingkey=zero")
return tmpl.Funcs(funcs), nil
}

View File

@@ -72,7 +72,7 @@ services:
restart: always
vmanomaly:
container_name: vmanomaly
image: victoriametrics/vmanomaly:v1.18.1
image: victoriametrics/vmanomaly:v1.18.0
depends_on:
- "victoriametrics"
ports:

View File

@@ -11,25 +11,6 @@ aliases:
---
Please find the changelog for VictoriaMetrics Anomaly Detection below.
## v1.18.1
Released: 2024-11-12
- IMPROVEMENT: Added a [reader-level](https://docs.victoriametrics.com/anomaly-detection/components/reader/#vm-reader) `data_range` argument, allowing users to define a default *valid* data range for all input queries in `queries`. Individual queries can still override this default with their own `data_range` if needed.
- IMPROVEMENT: Added the `url` label to enhance labelset consistency across [self-monitoring metrics](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#metrics-generated-by-vmanomaly) in both [reader](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#reader-behaviour-metrics) and [writer](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#writer-behaviour-metrics) components. Metrics affected:
- `vmanomaly_reader_received_bytes`
- `vmanomaly_reader_response_parsing_seconds`
- `vmanomaly_reader_timeseries_received`
- `vmanomaly_reader_datapoints_received`
- `vmanomaly_writer_request_serialize_seconds`
- `vmanomaly_writer_datapoints_sent`
- `vmanomaly_writer_timeseries_sent`
- FIX: Resolved an issue where [rolling models](https://docs.victoriametrics.com/anomaly-detection/components/models/#rolling-models) incorrectly set their last seen `infer` timestamp during *first* `fit_infer` call, resulting in output being produced for *every datapoint* within the `fit_window` on its *first invocation*.
- FIX: Resolved an issue in multi-[scheduler](https://docs.victoriametrics.com/anomaly-detection/components/scheduler/) configurations where [self-monitoring metric](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#metrics-generated-by-vmanomaly) values were overwriting each other.
- FIX: Resolved an issue causing incorrect `query_key` label values in the `vmanomaly_model_datapoints_produced` [self-monitoring metric](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#models-behaviour-metrics) for [univariate models](https://docs.victoriametrics.com/anomaly-detection/components/models/#univariate-models).
- FIX: Resolved an issue that caused the `vmanomaly_model_runs` [self-monitoring metric](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#models-behaviour-metrics) to miss increments for [rolling models](https://docs.victoriametrics.com/anomaly-detection/components/models/#rolling-models).
- FIX: Aligned the calculations of `vmanomaly_model_datapoints_accepted` and `vmanomaly_model_datapoints_produced` [self-monitoring model metrics](https://docs.victoriametrics.com/anomaly-detection/components/monitoring/#models-behaviour-metrics) across all stages (`fit`, `infer`, and `fit_infer`) for consistency.
## v1.18.0
Released: 2024-10-28

View File

@@ -159,7 +159,7 @@ services:
# ...
vmanomaly:
container_name: vmanomaly
image: victoriametrics/vmanomaly:v1.18.1
image: victoriametrics/vmanomaly:v1.18.0
# ...
ports:
- "8490:8490"

View File

@@ -229,7 +229,7 @@ This will expose metrics at `http://0.0.0.0:8080/metrics` page.
To use *vmanomaly* you need to pull docker image:
```sh
docker pull victoriametrics/vmanomaly:v1.18.1
docker pull victoriametrics/vmanomaly:v1.18.0
```
> Note: please check what is latest release in [CHANGELOG](https://docs.victoriametrics.com/anomaly-detection/changelog/)
@@ -239,7 +239,7 @@ docker pull victoriametrics/vmanomaly:v1.18.1
You can put a tag on it for your convenience:
```sh
docker image tag victoriametrics/vmanomaly:v1.18.1 vmanomaly
docker image tag victoriametrics/vmanomaly:v1.18.0 vmanomaly
```
Here is an example of how to run *vmanomaly* docker container with [license file](#licensing):

View File

@@ -58,13 +58,13 @@ Below are the steps to get `vmanomaly` up and running inside a Docker container:
1. Pull Docker image:
```sh
docker pull victoriametrics/vmanomaly:v1.18.1
docker pull victoriametrics/vmanomaly:v1.18.0
```
2. (Optional step) tag the `vmanomaly` Docker image:
```sh
docker image tag victoriametrics/vmanomaly:v1.18.1 vmanomaly
docker image tag victoriametrics/vmanomaly:v1.18.0 vmanomaly
```
3. Start the `vmanomaly` Docker container with a *license file*, use the command below.
@@ -98,7 +98,7 @@ docker run -it --user 1000:1000 \
services:
# ...
vmanomaly:
image: victoriametrics/vmanomaly:v1.18.1
image: victoriametrics/vmanomaly:v1.18.0
volumes:
$YOUR_LICENSE_FILE_PATH:/license
$YOUR_CONFIG_FILE_PATH:/config.yml

View File

@@ -984,7 +984,7 @@ monitoring:
Let's pull the docker image for `vmanomaly`:
```sh
docker pull victoriametrics/vmanomaly:v1.18.1
docker pull victoriametrics/vmanomaly:v1.18.0
```
Now we can run the docker container putting as volumes both config and model file:
@@ -998,7 +998,7 @@ docker run -it \
-v $(PWD)/license:/license \
-v $(PWD)/custom_model.py:/vmanomaly/model/custom.py \
-v $(PWD)/custom.yaml:/config.yaml \
victoriametrics/vmanomaly:v1.18.1 /config.yaml \
victoriametrics/vmanomaly:v1.18.0 /config.yaml \
--licenseFile=/license
```

View File

@@ -319,7 +319,7 @@ Label names [description](#labelnames)
<td>The total number of bytes received in responses for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`url`, `query_key`, `scheduler_alias`, `preset`
`query_key`, `scheduler_alias`, `preset`
</td>
</tr>
<tr>
@@ -331,7 +331,7 @@ Label names [description](#labelnames)
<td>The total time (in seconds) taken for data parsing at each `step` (json, dataframe) for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`step`, `url`, `query_key`, `scheduler_alias`, `preset`
`step`, `query_key`, `scheduler_alias`, `preset`
</td>
</tr>
<tr>
@@ -343,7 +343,7 @@ Label names [description](#labelnames)
<td>The total number of timeseries received from VictoriaMetrics for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`url`, `query_key`, `scheduler_alias`, `preset`
`query_key`, `scheduler_alias`, `preset`
</td>
</tr>
<tr>
@@ -355,7 +355,7 @@ Label names [description](#labelnames)
<td>The total number of datapoints received from VictoriaMetrics for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`url`, `query_key`, `scheduler_alias`, `preset`
`query_key`, `scheduler_alias`, `preset`
</td>
</tr>
</tbody>
@@ -530,7 +530,7 @@ Label names [description](#labelnames)
<td>The total time (in seconds) taken for serializing data for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`url`, `query_key`, `scheduler_alias`, `preset`
`query_key`, `scheduler_alias`, `preset`
</td>
</tr>
<tr>
@@ -542,7 +542,7 @@ Label names [description](#labelnames)
<td>The total number of datapoints sent to VictoriaMetrics for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`url`, `query_key`, `scheduler_alias`, `preset`
`query_key`, `scheduler_alias`, `preset`
</td>
</tr>
<tr>
@@ -552,8 +552,7 @@ Label names [description](#labelnames)
<td>`Counter`</td>
<td>The total number of timeseries sent to VictoriaMetrics for the `query_key` query within the specified scheduler `scheduler_alias`, in the `vmanomaly` service running in `preset` mode.</td>
<td>
`url`, `query_key`, `scheduler_alias`, `preset`
`query_key`, `scheduler_alias`, `preset`
</td>
</tr>
</tbody>

View File

@@ -68,8 +68,6 @@ Starting from [v1.13.0](https://docs.victoriametrics.com/anomaly-detection/chang
- **High anomaly scores** (>1) when the *data falls outside the expected range*, indicating a data constraint violation.
- **Lowest anomaly scores** (=0) when the *model's predictions (`yhat`) fall outside the expected range*, meaning uncertain predictions.
> **Note**: if not set explicitly (or if older config style prior to [v1.13.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1130)) is used, then it is set to reader-level `data_range` arg (since [v1.18.1](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1181))
- `max_points_per_query` (int): Introduced in [v1.17.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1170), optional arg overrides how `search.maxPointsPerTimeseries` flag (available since [v1.14.1](#v1141)) impacts `vmanomaly` on splitting long `fit_window` [queries](https://docs.victoriametrics.com/anomaly-detection/components/reader/?highlight=queries#vm-reader) into smaller sub-intervals. This helps users avoid hitting the `search.maxQueryDuration` limit for individual queries by distributing initial query across multiple subquery requests with minimal overhead. Set less than `search.maxPointsPerTimeseries` if hitting `maxQueryDuration` limits. If set on a query-level, it overrides the global `max_points_per_query` (reader-level).
- `tz` (string): Introduced in [v1.18.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1180), this optional argument enables timezone specification per query, overriding the readers default `tz`. This setting helps to account for local timezone shifts, such as [DST](https://en.wikipedia.org/wiki/Daylight_saving_time), in models that are sensitive to seasonal variations (e.g., [`ProphetModel`](https://docs.victoriametrics.com/anomaly-detection/components/models/#prophet) or [`OnlineQuantileModel`](https://docs.victoriametrics.com/anomaly-detection/components/models/#online-seasonal-quantile)).
@@ -325,17 +323,6 @@ Introduced in [v1.17.0](https://docs.victoriametrics.com/anomaly-detection/chang
Introduced in [v1.18.0](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1180), this optional argument specifies the [IANA](https://nodatime.org/TimeZones) timezone to account for local shifts, like [DST](https://en.wikipedia.org/wiki/Daylight_saving_time), in models sensitive to seasonal patterns (e.g., [`ProphetModel`](https://docs.victoriametrics.com/anomaly-detection/components/models/#prophet) or [`OnlineQuantileModel`](https://docs.victoriametrics.com/anomaly-detection/components/models/#online-seasonal-quantile)). Defaults to `UTC` if not set and can be overridden on a [per-query basis](#per-query-parameters).
</td>
</tr>
<tr>
<td>
`data_range`
</td>
<td>
`["-inf", "inf"]`
</td>
<td>
Added in [v1.18.1](https://docs.victoriametrics.com/anomaly-detection/changelog/#v1180), this optional argument allows defining **valid** data ranges for input of all the queries in `queries`. Defaults to `["-inf", "inf"]` if not set and can be overridden on a [per-query basis](#per-query-parameters).
</td>
</tr>
</tbody>
</table>
@@ -347,13 +334,12 @@ reader:
datasource_url: "https://play.victoriametrics.com/"
tenant_id: "0:0"
tz: 'America/New_York'
data_range: [1, 'inf'] # reader-level
queries:
ingestion_rate:
expr: 'sum(rate(vm_rows_inserted_total[5m])) by (type) > 0'
step: '1m' # can override reader-level `sampling_period` on per-query level
data_range: [0, 'inf'] # if set, overrides reader-level data_range
tz: 'Australia/Sydney' # if set, overrides reader-level tz
step: '1m' # can override global `sampling_period` on per-query level
data_range: [0, 'inf']
tz: 'Australia/Sydney' # if set, overrides reader-wise tz
sampling_period: '1m'
query_from_last_seen_timestamp: True # false by default
latency_offset: '1ms'

View File

@@ -385,7 +385,7 @@ services:
restart: always
vmanomaly:
container_name: vmanomaly
image: victoriametrics/vmanomaly:v1.18.1
image: victoriametrics/vmanomaly:v1.18.0
depends_on:
- "victoriametrics"
ports:

View File

@@ -27,7 +27,7 @@ func (wr *WriteRequest) Reset() {
// ResetTimeSeries clears all the GC references from tss and returns an empty tss ready for further use.
func ResetTimeSeries(tss []TimeSeries) []TimeSeries {
clear(tss)
return make([]TimeSeries, 0)
return tss[:0]
}
// MustParsePromMetrics parses metrics in Prometheus text exposition format from s and returns them.

View File

@@ -11,16 +11,77 @@ import (
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
)
type queue struct {
in chan uint64
out chan uint64
storage []uint64
}
func newQueue() *queue {
q := &queue{
out: make(chan uint64),
in: make(chan uint64),
}
go func() {
defer close(q.out)
for {
if len(q.storage) == 0 {
item, ok := <-q.in
if !ok {
return
}
q.storage = append(q.storage, item)
continue
}
select {
case item, ok := <-q.in:
if ok {
q.storage = append(q.storage, item)
} else {
// unwind storage
for _, item := range q.storage {
q.out <- item
}
return
}
case q.out <- q.storage[0]:
if len(q.storage) == 1 {
q.storage = nil
} else {
q.storage = q.storage[1:]
}
}
}
}()
return q
}
type compressedLabel struct {
mu sync.Mutex
code uint64
deleted bool
deleteDeadline int64
}
// LabelsCompressor compresses []prompbmarshal.Label into short binary strings
type LabelsCompressor struct {
labelToIdx sync.Map
idxToLabel labelsMap
freeIdxs *queue
nextIdx atomic.Uint64
totalSizeBytes atomic.Uint64
}
func NewLabelsCompressor() *LabelsCompressor {
return &LabelsCompressor{
freeIdxs: newQueue(),
}
}
// SizeBytes returns the size of lc data in bytes
func (lc *LabelsCompressor) SizeBytes() uint64 {
return uint64(unsafe.Sizeof(*lc)) + lc.totalSizeBytes.Load()
@@ -28,13 +89,60 @@ func (lc *LabelsCompressor) SizeBytes() uint64 {
// ItemsCount returns the number of items in lc
func (lc *LabelsCompressor) ItemsCount() uint64 {
return lc.nextIdx.Load()
return lc.nextIdx.Load() - uint64(len(lc.freeIdxs.storage))
}
// Delete adds stale labels idx to lc.freeIdxs list
func (lc *LabelsCompressor) Delete(src []byte, ts int64) {
labelsLen, nSize := encoding.UnmarshalVarUint64(src)
if nSize <= 0 {
logger.Panicf("BUG: cannot unmarshal labels length from uvarint")
}
tail := src[nSize:]
if labelsLen == 0 {
// fast path - nothing to decode
if len(tail) > 0 {
logger.Panicf("BUG: unexpected non-empty tail left; len(tail)=%d; tail=%X", len(tail), tail)
}
return
}
a := encoding.GetUint64s(int(labelsLen))
var err error
tail, err = encoding.UnmarshalVarUint64s(a.A, tail)
if err != nil {
logger.Panicf("BUG: cannot unmarshal label indexes: %s", err)
}
if len(tail) > 0 {
logger.Panicf("BUG: unexpected non-empty tail left: len(tail)=%d; tail=%X", len(tail), tail)
}
for _, idx := range a.A {
label, ok := lc.idxToLabel.Load(idx)
if !ok {
logger.Panicf("BUG: missing label for idx=%d", idx)
}
v, ok := lc.labelToIdx.Load(label)
if !ok {
continue
}
cl := v.(compressedLabel)
cl.mu.Lock()
if cl.deleteDeadline < ts {
cl.deleted = true
cl.mu.Unlock()
lc.freeIdxs.in <- idx
} else {
cl.mu.Unlock()
}
}
encoding.PutUint64s(a)
}
// Compress compresses labels, appends the compressed labels to dst and returns the result.
//
// It is safe calling Compress from concurrent goroutines.
func (lc *LabelsCompressor) Compress(dst []byte, labels []prompbmarshal.Label) []byte {
func (lc *LabelsCompressor) Compress(dst []byte, labels []prompbmarshal.Label, deleteDeadline int64) []byte {
if len(labels) == 0 {
// Fast path
return append(dst, 0)
@@ -42,22 +150,27 @@ func (lc *LabelsCompressor) Compress(dst []byte, labels []prompbmarshal.Label) [
a := encoding.GetUint64s(len(labels) + 1)
a.A[0] = uint64(len(labels))
lc.compress(a.A[1:], labels)
lc.compress(a.A[1:], labels, deleteDeadline)
dst = encoding.MarshalVarUint64s(dst, a.A)
encoding.PutUint64s(a)
return dst
}
func (lc *LabelsCompressor) compress(dst []uint64, labels []prompbmarshal.Label) {
func (lc *LabelsCompressor) compress(dst []uint64, labels []prompbmarshal.Label, deleteDeadline int64) {
if len(labels) == 0 {
return
}
_ = dst[len(labels)-1]
for i, label := range labels {
again:
v, ok := lc.labelToIdx.Load(label)
if !ok {
idx := lc.nextIdx.Add(1)
v = idx
var idx uint64
select {
case idx = <-lc.freeIdxs.out:
default:
idx = lc.nextIdx.Add(1)
}
labelCopy := cloneLabel(label)
// Must store idxToLabel entry before labelToIdx,
@@ -66,6 +179,10 @@ func (lc *LabelsCompressor) compress(dst []uint64, labels []prompbmarshal.Label)
// We might store duplicated entries for single label with different indexes,
// and it's fine, see https://github.com/VictoriaMetrics/VictoriaMetrics/pull/7118.
lc.idxToLabel.Store(idx, labelCopy)
v = &compressedLabel{
deleteDeadline: deleteDeadline,
code: idx,
}
vNew, loaded := lc.labelToIdx.LoadOrStore(labelCopy, v)
if loaded {
// This label has been stored by a concurrent goroutine with different index,
@@ -78,7 +195,17 @@ func (lc *LabelsCompressor) compress(dst []uint64, labels []prompbmarshal.Label)
entrySizeBytes := labelSizeBytes + uint64(2*(unsafe.Sizeof(label)+unsafe.Sizeof(&label))+unsafe.Sizeof(v))
lc.totalSizeBytes.Add(entrySizeBytes)
}
dst[i] = v.(uint64)
cl := v.(*compressedLabel)
dst[i] = cl.code
cl.mu.Lock()
deleted := cl.deleted
if !deleted {
cl.deleteDeadline = deleteDeadline
}
cl.mu.Unlock()
if deleted {
goto again
}
}
}

View File

@@ -9,14 +9,14 @@ import (
)
func TestLabelsCompressorSerial(t *testing.T) {
var lc LabelsCompressor
lc := NewLabelsCompressor()
f := func(labels []prompbmarshal.Label) {
t.Helper()
sExpected := labelsToString(labels)
data := lc.Compress(nil, labels)
data := lc.Compress(nil, labels, 0)
labelsResult := lc.Decompress(nil, data)
sResult := labelsToString(labelsResult)
@@ -67,7 +67,7 @@ func TestLabelsCompressorSerial(t *testing.T) {
func TestLabelsCompressorConcurrent(t *testing.T) {
const concurrency = 5
var lc LabelsCompressor
lc := NewLabelsCompressor()
var expectCompressedKeys sync.Map
var wg sync.WaitGroup
@@ -78,7 +78,7 @@ func TestLabelsCompressorConcurrent(t *testing.T) {
series := newTestSeries(100, 20)
for n, labels := range series {
sExpected := labelsToString(labels)
data := lc.Compress(nil, labels)
data := lc.Compress(nil, labels, 0)
if expectData, ok := expectCompressedKeys.LoadOrStore(n, data); ok {
if string(data) != string(expectData.([]byte)) {
panic(fmt.Errorf("unexpected compress result at series/%d in iteration %d ", n, i))

View File

@@ -8,7 +8,7 @@ import (
)
func BenchmarkLabelsCompressorCompress(b *testing.B) {
var lc LabelsCompressor
lc := NewLabelsCompressor()
series := newTestSeries(100, 10)
b.ReportAllocs()
@@ -19,7 +19,7 @@ func BenchmarkLabelsCompressorCompress(b *testing.B) {
for pb.Next() {
dst = dst[:0]
for _, labels := range series {
dst = lc.Compress(dst, labels)
dst = lc.Compress(dst, labels, 0)
}
Sink.Add(uint64(len(dst)))
}
@@ -27,13 +27,13 @@ func BenchmarkLabelsCompressorCompress(b *testing.B) {
}
func BenchmarkLabelsCompressorDecompress(b *testing.B) {
var lc LabelsCompressor
lc := NewLabelsCompressor()
series := newTestSeries(100, 10)
datas := make([][]byte, len(series))
var dst []byte
for i, labels := range series {
dstLen := len(dst)
dst = lc.Compress(dst, labels)
dst = lc.Compress(dst, labels, 0)
datas[i] = dst[dstLen:]
}

View File

@@ -12,37 +12,33 @@ type avgAggrState struct {
}
type avgStateValue struct {
mu sync.Mutex
sum float64
count int64
deleted bool
mu sync.Mutex
sum float64
count int64
deleted bool
deleteDeadline int64
}
func newAvgAggrState() *avgAggrState {
return &avgAggrState{}
}
func (as *avgAggrState) pushSamples(samples []pushSample) {
func (as *avgAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &avgStateValue{
sum: s.value,
count: 1,
}
v = &avgStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The entry has been successfully stored
continue
if loaded {
// Update the entry created by a concurrent goroutine.
v = vNew
}
// Update the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*avgStateValue)
sv.mu.Lock()
@@ -50,6 +46,7 @@ func (as *avgAggrState) pushSamples(samples []pushSample) {
if !deleted {
sv.sum += s.value
sv.count++
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -68,9 +65,22 @@ func (as *avgAggrState) flushState(ctx *flushCtx) {
sv := v.(*avgStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
if sv.count == 0 {
sv.mu.Unlock()
return true
}
avg := sv.sum / float64(sv.count)
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.sum = 0
sv.count = 0
sv.mu.Unlock()
key := k.(string)

View File

@@ -12,41 +12,39 @@ type countSamplesAggrState struct {
}
type countSamplesStateValue struct {
mu sync.Mutex
n uint64
deleted bool
mu sync.Mutex
n uint64
deleted bool
deleteDeadline int64
}
func newCountSamplesAggrState() *countSamplesAggrState {
return &countSamplesAggrState{}
}
func (as *countSamplesAggrState) pushSamples(samples []pushSample) {
func (as *countSamplesAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &countSamplesStateValue{
n: 1,
}
v = &countSamplesStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The new entry has been successfully created.
continue
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
// Use the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*countSamplesStateValue)
sv.mu.Lock()
deleted := sv.deleted
if !deleted {
sv.n++
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -65,9 +63,21 @@ func (as *countSamplesAggrState) flushState(ctx *flushCtx) {
sv := v.(*countSamplesStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
n := sv.n
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
if n == 0 {
sv.mu.Unlock()
return true
}
sv.n = 0
sv.mu.Unlock()
key := k.(string)

View File

@@ -13,16 +13,17 @@ type countSeriesAggrState struct {
}
type countSeriesStateValue struct {
mu sync.Mutex
m map[uint64]struct{}
deleted bool
mu sync.Mutex
m map[uint64]struct{}
deleted bool
deleteDeadline int64
}
func newCountSeriesAggrState() *countSeriesAggrState {
return &countSeriesAggrState{}
}
func (as *countSeriesAggrState) pushSamples(samples []pushSample) {
func (as *countSeriesAggrState) pushSamples(samples []pushSample, deleteDeadline int64, _ bool) {
for i := range samples {
s := &samples[i]
inputKey, outputKey := getInputOutputKey(s.key)
@@ -36,18 +37,14 @@ func (as *countSeriesAggrState) pushSamples(samples []pushSample) {
if !ok {
// The entry is missing in the map. Try creating it.
v = &countSeriesStateValue{
m: map[uint64]struct{}{
h: {},
},
m: make(map[uint64]struct{}),
}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The entry has been added to the map.
continue
if loaded {
// Update the entry created by a concurrent goroutine.
v = vNew
}
// Update the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*countSeriesStateValue)
sv.mu.Lock()
@@ -56,6 +53,7 @@ func (as *countSeriesAggrState) pushSamples(samples []pushSample) {
if _, ok := sv.m[h]; !ok {
sv.m[h] = struct{}{}
}
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -74,9 +72,21 @@ func (as *countSeriesAggrState) flushState(ctx *flushCtx) {
sv := v.(*countSeriesStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
n := len(sv.m)
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
if n == 0 {
sv.mu.Unlock()
return true
}
sv.m = make(map[uint64]struct{})
sv.mu.Unlock()
key := k.(string)

View File

@@ -8,12 +8,14 @@ import (
"github.com/cespare/xxhash/v2"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/slicesutil"
)
const dedupAggrShardsCount = 128
type dedupAggr struct {
lc *promutils.LabelsCompressor
shards []dedupAggrShard
}
@@ -26,7 +28,7 @@ type dedupAggrShard struct {
}
type dedupAggrShardNopad struct {
mu sync.Mutex
mu sync.RWMutex
m map[string]*dedupAggrSample
samplesBuf []dedupAggrSample
@@ -36,14 +38,16 @@ type dedupAggrShardNopad struct {
}
type dedupAggrSample struct {
value float64
timestamp int64
value float64
timestamp int64
deleteDeadline int64
}
func newDedupAggr() *dedupAggr {
func newDedupAggr(lc *promutils.LabelsCompressor) *dedupAggr {
shards := make([]dedupAggrShard, dedupAggrShardsCount)
return &dedupAggr{
shards: shards,
lc: lc,
}
}
@@ -63,7 +67,7 @@ func (da *dedupAggr) itemsCount() uint64 {
return n
}
func (da *dedupAggr) pushSamples(samples []pushSample) {
func (da *dedupAggr) pushSamples(samples []pushSample, deleteDeadline int64) {
pss := getPerShardSamples()
shards := pss.shards
for _, sample := range samples {
@@ -75,17 +79,21 @@ func (da *dedupAggr) pushSamples(samples []pushSample) {
if len(shardSamples) == 0 {
continue
}
da.shards[i].pushSamples(shardSamples)
da.shards[i].pushSamples(shardSamples, deleteDeadline)
}
putPerShardSamples(pss)
}
func getDedupFlushCtx() *dedupFlushCtx {
func getDedupFlushCtx(flushTimestamp, deleteDeadline int64, lc *promutils.LabelsCompressor) *dedupFlushCtx {
v := dedupFlushCtxPool.Get()
if v == nil {
return &dedupFlushCtx{}
v = &dedupFlushCtx{}
}
return v.(*dedupFlushCtx)
ctx := v.(*dedupFlushCtx)
ctx.lc = lc
ctx.deleteDeadline = deleteDeadline
ctx.flushTimestamp = flushTimestamp
return ctx
}
func putDedupFlushCtx(ctx *dedupFlushCtx) {
@@ -96,15 +104,24 @@ func putDedupFlushCtx(ctx *dedupFlushCtx) {
var dedupFlushCtxPool sync.Pool
type dedupFlushCtx struct {
samples []pushSample
keysToDelete []string
samples []pushSample
deleteDeadline int64
flushTimestamp int64
lc *promutils.LabelsCompressor
}
func (ctx *dedupFlushCtx) reset() {
ctx.deleteDeadline = 0
ctx.flushTimestamp = 0
ctx.lc = nil
clear(ctx.keysToDelete)
ctx.keysToDelete = ctx.keysToDelete[:0]
clear(ctx.samples)
ctx.samples = ctx.samples[:0]
}
func (da *dedupAggr) flush(f func(samples []pushSample)) {
func (da *dedupAggr) flush(f func(samples []pushSample, deleteDeadline int64), flushTimestamp, deleteDeadline int64) {
var wg sync.WaitGroup
for i := range da.shards {
flushConcurrencyCh <- struct{}{}
@@ -115,7 +132,7 @@ func (da *dedupAggr) flush(f func(samples []pushSample)) {
wg.Done()
}()
ctx := getDedupFlushCtx()
ctx := getDedupFlushCtx(flushTimestamp, deleteDeadline, da.lc)
shard.flush(ctx, f)
putDedupFlushCtx(ctx)
}(&da.shards[i])
@@ -154,7 +171,7 @@ func putPerShardSamples(pss *perShardSamples) {
var perShardSamplesPool sync.Pool
func (das *dedupAggrShard) pushSamples(samples []pushSample) {
func (das *dedupAggrShard) pushSamples(samples []pushSample, deleteDeadline int64) {
das.mu.Lock()
defer das.mu.Unlock()
@@ -171,6 +188,7 @@ func (das *dedupAggrShard) pushSamples(samples []pushSample) {
s = &samplesBuf[len(samplesBuf)-1]
s.value = sample.value
s.timestamp = sample.timestamp
s.deleteDeadline = deleteDeadline
key := bytesutil.InternString(sample.key)
m[key] = s
@@ -183,30 +201,28 @@ func (das *dedupAggrShard) pushSamples(samples []pushSample) {
if sample.timestamp > s.timestamp || (sample.timestamp == s.timestamp && sample.value > s.value) {
s.value = sample.value
s.timestamp = sample.timestamp
s.deleteDeadline = deleteDeadline
}
}
das.samplesBuf = samplesBuf
}
func (das *dedupAggrShard) flush(ctx *dedupFlushCtx, f func(samples []pushSample)) {
das.mu.Lock()
m := das.m
if len(m) > 0 {
das.m = make(map[string]*dedupAggrSample, len(m))
das.sizeBytes.Store(0)
das.itemsCount.Store(0)
das.samplesBuf = make([]dedupAggrSample, 0, len(das.samplesBuf))
}
das.mu.Unlock()
if len(m) == 0 {
func (das *dedupAggrShard) flush(ctx *dedupFlushCtx, f func(samples []pushSample, deleteDeadline int64)) {
if len(das.m) == 0 {
return
}
keysToDelete := ctx.keysToDelete
dstSamples := ctx.samples
for key, s := range m {
das.mu.RLock()
for key, s := range das.m {
if ctx.flushTimestamp > s.deleteDeadline {
das.itemsCount.Add(^uint64(0))
//ctx.lc.Delete(key)
das.sizeBytes.Add(^(uint64(len(key)) + uint64(unsafe.Sizeof(key)+unsafe.Sizeof(s)+unsafe.Sizeof(*s)) - 1))
keysToDelete = append(keysToDelete, key)
continue
}
dstSamples = append(dstSamples, pushSample{
key: key,
value: s.value,
@@ -215,11 +231,17 @@ func (das *dedupAggrShard) flush(ctx *dedupFlushCtx, f func(samples []pushSample
// Limit the number of samples per each flush in order to limit memory usage.
if len(dstSamples) >= 10_000 {
f(dstSamples)
f(dstSamples, ctx.deleteDeadline)
clear(dstSamples)
dstSamples = dstSamples[:0]
}
}
f(dstSamples)
das.mu.RUnlock()
das.mu.Lock()
for _, key := range keysToDelete {
delete(das.m, key)
}
das.mu.Unlock()
f(dstSamples, ctx.deleteDeadline)
ctx.samples = dstSamples
}

View File

@@ -5,10 +5,13 @@ import (
"reflect"
"sync"
"testing"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
)
func TestDedupAggrSerial(t *testing.T) {
da := newDedupAggr()
var lc promutils.LabelsCompressor
da := newDedupAggr(&lc)
const seriesCount = 100_000
expectedSamplesMap := make(map[string]pushSample)
@@ -20,11 +23,11 @@ func TestDedupAggrSerial(t *testing.T) {
sample.value = float64(i + j)
expectedSamplesMap[sample.key] = *sample
}
da.pushSamples(samples)
da.pushSamples(samples, 0)
}
if n := da.sizeBytes(); n > 5_000_000 {
t.Fatalf("too big dedupAggr state before flush: %d bytes; it shouldn't exceed 5_000_000 bytes", n)
if n := da.sizeBytes(); n > 6_000_000 {
t.Fatalf("too big dedupAggr state before flush: %d bytes; it shouldn't exceed 6_000_000 bytes", n)
}
if n := da.itemsCount(); n != seriesCount {
t.Fatalf("unexpected itemsCount; got %d; want %d", n, seriesCount)
@@ -32,19 +35,21 @@ func TestDedupAggrSerial(t *testing.T) {
flushedSamplesMap := make(map[string]pushSample)
var mu sync.Mutex
flushSamples := func(samples []pushSample) {
flushSamples := func(samples []pushSample, _ int64) {
mu.Lock()
for _, sample := range samples {
flushedSamplesMap[sample.key] = sample
}
mu.Unlock()
}
da.flush(flushSamples)
da.flush(flushSamples, 0, 0)
if !reflect.DeepEqual(expectedSamplesMap, flushedSamplesMap) {
t.Fatalf("unexpected samples;\ngot\n%v\nwant\n%v", flushedSamplesMap, expectedSamplesMap)
}
da.flush(flushSamples, 1, 0)
if n := da.sizeBytes(); n > 17_000 {
t.Fatalf("too big dedupAggr state after flush; %d bytes; it shouldn't exceed 17_000 bytes", n)
}
@@ -54,9 +59,10 @@ func TestDedupAggrSerial(t *testing.T) {
}
func TestDedupAggrConcurrent(_ *testing.T) {
var lc promutils.LabelsCompressor
const concurrency = 5
const seriesCount = 10_000
da := newDedupAggr()
da := newDedupAggr(&lc)
var wg sync.WaitGroup
for i := 0; i < concurrency; i++ {
@@ -70,7 +76,7 @@ func TestDedupAggrConcurrent(_ *testing.T) {
sample.key = fmt.Sprintf("key_%d", j)
sample.value = float64(i + j)
}
da.pushSamples(samples)
da.pushSamples(samples, 0)
}
}()
}

View File

@@ -6,6 +6,7 @@ import (
"testing"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
)
func BenchmarkDedupAggr(b *testing.B) {
@@ -17,9 +18,11 @@ func BenchmarkDedupAggr(b *testing.B) {
}
func benchmarkDedupAggr(b *testing.B, samplesPerPush int) {
var lc promutils.LabelsCompressor
const loops = 2
benchSamples := newBenchSamples(samplesPerPush)
da := newDedupAggr()
da := newDedupAggr(&lc)
b.ResetTimer()
b.ReportAllocs()
@@ -27,13 +30,14 @@ func benchmarkDedupAggr(b *testing.B, samplesPerPush int) {
b.RunParallel(func(pb *testing.PB) {
for pb.Next() {
for i := 0; i < loops; i++ {
da.pushSamples(benchSamples)
da.pushSamples(benchSamples, 0)
}
}
})
}
func newBenchSamples(count int) []pushSample {
var lc promutils.LabelsCompressor
labels := []prompbmarshal.Label{
{
Name: "app",
@@ -65,7 +69,7 @@ func newBenchSamples(count int) []pushSample {
Name: "app",
Value: fmt.Sprintf("instance-%d", i),
})
keyBuf = compressLabels(keyBuf[:0], labels[:labelsLen], labels[labelsLen:])
keyBuf = compressLabels(keyBuf[:0], &lc, labels[:labelsLen], labels[labelsLen:], false, 0)
sample.key = string(keyBuf)
sample.value = float64(i)
}

View File

@@ -16,8 +16,10 @@ import (
// Deduplicator deduplicates samples per each time series.
type Deduplicator struct {
da *dedupAggr
lc *promutils.LabelsCompressor
dropLabels []string
dropLabels []string
stalenessInterval int64
wg sync.WaitGroup
stopCh chan struct{}
@@ -40,13 +42,16 @@ type Deduplicator struct {
// MustStop must be called on the returned deduplicator in order to free up occupied resources.
func NewDeduplicator(pushFunc PushFunc, dedupInterval time.Duration, dropLabels []string, alias string) *Deduplicator {
d := &Deduplicator{
da: newDedupAggr(),
dropLabels: dropLabels,
dropLabels: dropLabels,
stalenessInterval: 2 * dedupInterval.Milliseconds(),
lc: promutils.NewLabelsCompressor(),
stopCh: make(chan struct{}),
ms: metrics.NewSet(),
}
d.da = newDedupAggr(d.lc)
ms := d.ms
metricLabels := fmt.Sprintf(`name="dedup",url=%q`, alias)
@@ -57,6 +62,12 @@ func NewDeduplicator(pushFunc PushFunc, dedupInterval time.Duration, dropLabels
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_dedup_state_items_count{%s}`, metricLabels), func() float64 {
return float64(d.da.itemsCount())
})
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_labels_compressor_size_bytes{%s}`, metricLabels), func() float64 {
return float64(d.lc.SizeBytes())
})
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_labels_compressor_items_count{%s}`, metricLabels), func() float64 {
return float64(d.lc.ItemsCount())
})
d.dedupFlushDuration = ms.NewHistogram(fmt.Sprintf(`vm_streamaggr_dedup_flush_duration_seconds{%s}`, metricLabels))
d.dedupFlushTimeouts = ms.NewCounter(fmt.Sprintf(`vm_streamaggr_dedup_flush_timeouts_total{%s}`, metricLabels))
@@ -87,6 +98,7 @@ func (d *Deduplicator) Push(tss []prompbmarshal.TimeSeries) {
pss := ctx.pss
labels := &ctx.labels
buf := ctx.buf
deleteDeadline := time.Now().UnixMilli() + d.stalenessInterval
dropLabels := d.dropLabels
for _, ts := range tss {
@@ -101,7 +113,7 @@ func (d *Deduplicator) Push(tss []prompbmarshal.TimeSeries) {
labels.Sort()
bufLen := len(buf)
buf = lc.Compress(buf, labels.Labels)
buf = d.lc.Compress(buf, labels.Labels, deleteDeadline)
key := bytesutil.ToUnsafeString(buf[bufLen:])
for _, s := range ts.Samples {
pss = append(pss, pushSample{
@@ -112,7 +124,7 @@ func (d *Deduplicator) Push(tss []prompbmarshal.TimeSeries) {
}
}
d.da.pushSamples(pss)
d.da.pushSamples(pss, deleteDeadline)
ctx.pss = pss
ctx.buf = buf
@@ -145,7 +157,8 @@ func (d *Deduplicator) flush(pushFunc PushFunc, dedupInterval time.Duration) {
startTime := time.Now()
timestamp := startTime.UnixMilli()
d.da.flush(func(pss []pushSample) {
deleteDeadline := timestamp + d.stalenessInterval
d.da.flush(func(pss []pushSample, deleteDeadline int64) {
ctx := getDeduplicatorFlushCtx()
tss := ctx.tss
@@ -153,7 +166,7 @@ func (d *Deduplicator) flush(pushFunc PushFunc, dedupInterval time.Duration) {
samples := ctx.samples
for _, ps := range pss {
labelsLen := len(labels)
labels = decompressLabels(labels, ps.key)
labels = decompressLabels(labels, d.lc, ps.key)
samplesLen := len(samples)
samples = append(samples, prompbmarshal.Sample{
@@ -172,7 +185,7 @@ func (d *Deduplicator) flush(pushFunc PushFunc, dedupInterval time.Duration) {
ctx.labels = labels
ctx.samples = samples
putDeduplicatorFlushCtx(ctx)
})
}, timestamp, deleteDeadline)
duration := time.Since(startTime)
d.dedupFlushDuration.Update(duration.Seconds())

View File

@@ -1,42 +1,32 @@
package streamaggr
import (
"math"
"sync"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
"github.com/VictoriaMetrics/metrics"
)
// histogramBucketAggrState calculates output=histogram_bucket, e.g. VictoriaMetrics histogram over input samples.
type histogramBucketAggrState struct {
m sync.Map
stalenessSecs uint64
}
type histogramBucketStateValue struct {
mu sync.Mutex
h metrics.Histogram
deleteDeadline uint64
deleteDeadline int64
deleted bool
}
func newHistogramBucketAggrState(stalenessInterval time.Duration) *histogramBucketAggrState {
stalenessSecs := roundDurationToSecs(stalenessInterval)
return &histogramBucketAggrState{
stalenessSecs: stalenessSecs,
}
func newHistogramBucketAggrState() *histogramBucketAggrState {
return &histogramBucketAggrState{}
}
func (as *histogramBucketAggrState) pushSamples(samples []pushSample) {
currentTime := fasttime.UnixTimestamp()
deleteDeadline := currentTime + as.stalenessSecs
func (as *histogramBucketAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
@@ -66,13 +56,13 @@ func (as *histogramBucketAggrState) pushSamples(samples []pushSample) {
}
}
func (as *histogramBucketAggrState) removeOldEntries(currentTime uint64) {
func (as *histogramBucketAggrState) removeOldEntries(ctx *flushCtx) {
m := &as.m
m.Range(func(k, v any) bool {
sv := v.(*histogramBucketStateValue)
sv.mu.Lock()
deleted := currentTime > sv.deleteDeadline
deleted := ctx.flushTimestamp > sv.deleteDeadline
if deleted {
// Mark the current entry as deleted
sv.deleted = deleted
@@ -80,6 +70,8 @@ func (as *histogramBucketAggrState) removeOldEntries(currentTime uint64) {
sv.mu.Unlock()
if deleted {
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
}
return true
@@ -87,9 +79,7 @@ func (as *histogramBucketAggrState) removeOldEntries(currentTime uint64) {
}
func (as *histogramBucketAggrState) flushState(ctx *flushCtx) {
currentTime := fasttime.UnixTimestamp()
as.removeOldEntries(currentTime)
as.removeOldEntries(ctx)
m := &as.m
m.Range(func(k, v any) bool {
@@ -105,11 +95,3 @@ func (as *histogramBucketAggrState) flushState(ctx *flushCtx) {
return true
})
}
func roundDurationToSecs(d time.Duration) uint64 {
if d < 0 {
return 0
}
secs := d.Seconds()
return uint64(math.Ceil(secs))
}

View File

@@ -12,46 +12,45 @@ type lastAggrState struct {
}
type lastStateValue struct {
mu sync.Mutex
last float64
timestamp int64
deleted bool
mu sync.Mutex
last float64
timestamp int64
deleted bool
defined bool
deleteDeadline int64
}
func newLastAggrState() *lastAggrState {
return &lastAggrState{}
}
func (as *lastAggrState) pushSamples(samples []pushSample) {
func (as *lastAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &lastStateValue{
last: s.value,
timestamp: s.timestamp,
}
v = &lastStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The new entry has been successfully created.
continue
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
// Use the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*lastStateValue)
sv.mu.Lock()
deleted := sv.deleted
if !deleted {
if s.timestamp >= sv.timestamp {
if !sv.defined || s.timestamp >= sv.timestamp {
sv.last = s.value
sv.timestamp = s.timestamp
sv.deleteDeadline = deleteDeadline
}
sv.defined = true
}
sv.mu.Unlock()
if deleted {
@@ -70,6 +69,14 @@ func (as *lastAggrState) flushState(ctx *flushCtx) {
sv := v.(*lastStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
last := sv.last
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true

View File

@@ -12,35 +12,33 @@ type maxAggrState struct {
}
type maxStateValue struct {
mu sync.Mutex
max float64
deleted bool
mu sync.Mutex
max float64
deleted bool
defined bool
deleteDeadline int64
}
func newMaxAggrState() *maxAggrState {
return &maxAggrState{}
}
func (as *maxAggrState) pushSamples(samples []pushSample) {
func (as *maxAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &maxStateValue{
max: s.value,
}
v = &maxStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The new entry has been successfully created.
continue
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
// Use the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*maxStateValue)
sv.mu.Lock()
@@ -49,6 +47,10 @@ func (as *maxAggrState) pushSamples(samples []pushSample) {
if s.value > sv.max {
sv.max = s.value
}
if !sv.defined {
sv.defined = true
}
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -67,9 +69,21 @@ func (as *maxAggrState) flushState(ctx *flushCtx) {
sv := v.(*maxStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
if !sv.defined {
sv.mu.Unlock()
return true
}
max := sv.max
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.max = 0
sv.defined = false
sv.mu.Unlock()
key := k.(string)

View File

@@ -12,43 +12,45 @@ type minAggrState struct {
}
type minStateValue struct {
mu sync.Mutex
min float64
deleted bool
mu sync.Mutex
min float64
deleted bool
defined bool
deleteDeadline int64
}
func newMinAggrState() *minAggrState {
return &minAggrState{}
}
func (as *minAggrState) pushSamples(samples []pushSample) {
func (as *minAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &minStateValue{
min: s.value,
}
v = &minStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The new entry has been successfully created.
continue
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
// Use the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*minStateValue)
sv.mu.Lock()
deleted := sv.deleted
if !deleted {
if s.value < sv.min {
if !sv.defined || s.value < sv.min {
sv.min = s.value
}
sv.deleteDeadline = deleteDeadline
if !sv.defined {
sv.defined = true
}
}
sv.mu.Unlock()
if deleted {
@@ -67,10 +69,23 @@ func (as *minAggrState) flushState(ctx *flushCtx) {
sv := v.(*minStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
if !sv.defined {
sv.mu.Unlock()
return true
}
min := sv.min
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.min = 0
sv.defined = false
sv.mu.Unlock()
key := k.(string)
ctx.appendSeries(key, "min", min)
return true

View File

@@ -16,9 +16,10 @@ type quantilesAggrState struct {
}
type quantilesStateValue struct {
mu sync.Mutex
h *histogram.Fast
deleted bool
mu sync.Mutex
h *histogram.Fast
deleted bool
deleteDeadline int64
}
func newQuantilesAggrState(phis []float64) *quantilesAggrState {
@@ -27,10 +28,10 @@ func newQuantilesAggrState(phis []float64) *quantilesAggrState {
}
}
func (as *quantilesAggrState) pushSamples(samples []pushSample) {
func (as *quantilesAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
@@ -53,6 +54,7 @@ func (as *quantilesAggrState) pushSamples(samples []pushSample) {
deleted := sv.deleted
if !deleted {
sv.h.Update(s.value)
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -74,10 +76,16 @@ func (as *quantilesAggrState) flushState(ctx *flushCtx) {
sv := v.(*quantilesStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
quantiles = sv.h.Quantiles(quantiles[:0], phis)
histogram.PutFast(sv.h)
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.h.Reset()
sv.mu.Unlock()
key := k.(string)

View File

@@ -2,10 +2,8 @@ package streamaggr
import (
"sync"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
)
// rateAggrState calculates output=rate_avg and rate_sum, e.g. the average per-second increase rate for counter metrics.
@@ -14,22 +12,19 @@ type rateAggrState struct {
// isAvg is set to true if rate_avg() must be calculated instead of rate_sum().
isAvg bool
// Time series state is dropped if no new samples are received during stalenessSecs.
stalenessSecs uint64
}
type rateStateValue struct {
mu sync.Mutex
lastValues map[string]rateLastValueState
deleteDeadline uint64
deleteDeadline int64
deleted bool
}
type rateLastValueState struct {
value float64
timestamp int64
deleteDeadline uint64
deleteDeadline int64
// increase stores cumulative increase for the current time series on the current aggregation interval
increase float64
@@ -38,17 +33,13 @@ type rateLastValueState struct {
prevTimestamp int64
}
func newRateAggrState(stalenessInterval time.Duration, isAvg bool) *rateAggrState {
stalenessSecs := roundDurationToSecs(stalenessInterval)
func newRateAggrState(isAvg bool) *rateAggrState {
return &rateAggrState{
isAvg: isAvg,
stalenessSecs: stalenessSecs,
isAvg: isAvg,
}
}
func (as *rateAggrState) pushSamples(samples []pushSample) {
currentTime := fasttime.UnixTimestamp()
deleteDeadline := currentTime + as.stalenessSecs
func (as *rateAggrState) pushSamples(samples []pushSample, deleteDeadline int64, _ bool) {
for i := range samples {
s := &samples[i]
inputKey, outputKey := getInputOutputKey(s.key)
@@ -106,11 +97,9 @@ func (as *rateAggrState) pushSamples(samples []pushSample) {
}
func (as *rateAggrState) flushState(ctx *flushCtx) {
currentTime := fasttime.UnixTimestamp()
suffix := as.getSuffix()
as.removeOldEntries(currentTime)
as.removeOldEntries(ctx)
m := &as.m
m.Range(func(k, v any) bool {
@@ -156,16 +145,18 @@ func (as *rateAggrState) getSuffix() string {
return "rate_sum"
}
func (as *rateAggrState) removeOldEntries(currentTime uint64) {
func (as *rateAggrState) removeOldEntries(ctx *flushCtx) {
m := &as.m
m.Range(func(k, v any) bool {
sv := v.(*rateStateValue)
sv.mu.Lock()
if currentTime > sv.deleteDeadline {
if ctx.flushTimestamp > sv.deleteDeadline {
// Mark the current entry as deleted
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
@@ -173,7 +164,7 @@ func (as *rateAggrState) removeOldEntries(currentTime uint64) {
// Delete outdated entries in sv.lastValues
lvs := sv.lastValues
for k1, lv := range lvs {
if currentTime > lv.deleteDeadline {
if ctx.flushTimestamp > lv.deleteDeadline {
delete(lvs, k1)
}
}

View File

@@ -13,21 +13,22 @@ type stddevAggrState struct {
}
type stddevStateValue struct {
mu sync.Mutex
count float64
avg float64
q float64
deleted bool
mu sync.Mutex
count float64
avg float64
q float64
deleted bool
deleteDeadline int64
}
func newStddevAggrState() *stddevAggrState {
return &stddevAggrState{}
}
func (as *stddevAggrState) pushSamples(samples []pushSample) {
func (as *stddevAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
@@ -50,6 +51,7 @@ func (as *stddevAggrState) pushSamples(samples []pushSample) {
avg := sv.avg + (s.value-sv.avg)/sv.count
sv.q += (s.value - sv.avg) * (s.value - avg)
sv.avg = avg
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -68,9 +70,23 @@ func (as *stddevAggrState) flushState(ctx *flushCtx) {
sv := v.(*stddevStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
if sv.count == 0 {
sv.mu.Unlock()
return true
}
stddev := math.Sqrt(sv.q / sv.count)
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.q = 0
sv.count = 0
sv.avg = 0
sv.mu.Unlock()
key := k.(string)

View File

@@ -12,21 +12,22 @@ type stdvarAggrState struct {
}
type stdvarStateValue struct {
mu sync.Mutex
count float64
avg float64
q float64
deleted bool
mu sync.Mutex
count float64
avg float64
q float64
deleted bool
deleteDeadline int64
}
func newStdvarAggrState() *stdvarAggrState {
return &stdvarAggrState{}
}
func (as *stdvarAggrState) pushSamples(samples []pushSample) {
func (as *stdvarAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
@@ -49,6 +50,7 @@ func (as *stdvarAggrState) pushSamples(samples []pushSample) {
avg := sv.avg + (s.value-sv.avg)/sv.count
sv.q += (s.value - sv.avg) * (s.value - avg)
sv.avg = avg
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -67,9 +69,22 @@ func (as *stdvarAggrState) flushState(ctx *flushCtx) {
sv := v.(*stdvarStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
if sv.count == 0 {
sv.mu.Unlock()
return true
}
stdvar := sv.q / sv.count
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.q = 0
sv.count = 0
sv.avg = 0
sv.mu.Unlock()
key := k.(string)

View File

@@ -47,19 +47,6 @@ var supportedOutputs = []string{
"unique_samples",
}
var (
// lc contains information about all compressed labels for streaming aggregation
lc promutils.LabelsCompressor
_ = metrics.NewGauge(`vm_streamaggr_labels_compressor_size_bytes`, func() float64 {
return float64(lc.SizeBytes())
})
_ = metrics.NewGauge(`vm_streamaggr_labels_compressor_items_count`, func() float64 {
return float64(lc.ItemsCount())
})
)
// LoadFromFile loads Aggregators from the given path and uses the given pushFunc for pushing the aggregated data.
//
// opts can contain additional options. If opts is nil, then default options are used.
@@ -373,13 +360,15 @@ type aggregator struct {
keepMetricNames bool
ignoreOldSamples bool
includeInputKey bool
by []string
without []string
aggregateOnlyByTime bool
// interval is the interval between flushes
interval time.Duration
interval time.Duration
stalenessInterval int64
// dedupInterval is optional deduplication interval for incoming samples
dedupInterval time.Duration
@@ -390,6 +379,9 @@ type aggregator struct {
// aggrOutputs contains aggregate states for the given outputs
aggrOutputs []aggrOutput
// lc is used for compressing series keys before passing them to dedupAggr and aggrState
lc *promutils.LabelsCompressor
// minTimestamp is used for ignoring old samples when ignoreOldSamples is set
minTimestamp atomic.Int64
@@ -424,7 +416,7 @@ type aggrState interface {
// pushSamples must push samples to the aggrState.
//
// samples[].key must be cloned by aggrState, since it may change after returning from pushSamples.
pushSamples(samples []pushSample)
pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool)
// flushState must flush aggrState data to ctx.
flushState(ctx *flushCtx)
@@ -556,11 +548,15 @@ func newAggregator(cfg *Config, path string, pushFunc PushFunc, ms *metrics.Set,
}
aggrOutputs := make([]aggrOutput, len(cfg.Outputs))
outputsSeen := make(map[string]struct{}, len(cfg.Outputs))
includeInputKey := false
for i, output := range cfg.Outputs {
as, err := newAggrState(output, outputsSeen, stalenessInterval)
as, ik, err := newAggrState(output, outputsSeen)
if err != nil {
return nil, err
}
if ik {
includeInputKey = true
}
aggrOutputs[i] = aggrOutput{
as: as,
@@ -592,9 +588,11 @@ func newAggregator(cfg *Config, path string, pushFunc PushFunc, ms *metrics.Set,
by: by,
without: without,
aggregateOnlyByTime: aggregateOnlyByTime,
lc: promutils.NewLabelsCompressor(),
interval: interval,
dedupInterval: dedupInterval,
interval: interval,
dedupInterval: dedupInterval,
stalenessInterval: stalenessInterval.Milliseconds(),
aggrOutputs: aggrOutputs,
@@ -614,18 +612,25 @@ func newAggregator(cfg *Config, path string, pushFunc PushFunc, ms *metrics.Set,
}
if dedupInterval > 0 {
a.da = newDedupAggr()
includeInputKey = true
a.da = newDedupAggr(a.lc)
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_dedup_state_size_bytes{%s}`, metricLabels), func() float64 {
n := a.da.sizeBytes()
return float64(n)
return float64(a.da.sizeBytes())
})
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_dedup_state_items_count{%s}`, metricLabels), func() float64 {
n := a.da.itemsCount()
return float64(n)
return float64(a.da.itemsCount())
})
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_labels_compressor_size_bytes{%s}`, metricLabels), func() float64 {
return float64(a.lc.SizeBytes())
})
_ = ms.NewGauge(fmt.Sprintf(`vm_streamaggr_labels_compressor_items_count{%s}`, metricLabels), func() float64 {
return float64(a.lc.ItemsCount())
})
}
a.includeInputKey = includeInputKey
alignFlushToInterval := !opts.NoAlignFlushToInterval
if v := cfg.NoAlignFlushToInterval; v != nil {
alignFlushToInterval = !*v
@@ -645,20 +650,20 @@ func newAggregator(cfg *Config, path string, pushFunc PushFunc, ms *metrics.Set,
return a, nil
}
func newAggrState(output string, outputsSeen map[string]struct{}, stalenessInterval time.Duration) (aggrState, error) {
func newAggrState(output string, outputsSeen map[string]struct{}) (aggrState, bool, error) {
// check for duplicated output
if _, ok := outputsSeen[output]; ok {
return nil, fmt.Errorf("`outputs` list contains duplicate aggregation function: %s", output)
return nil, true, fmt.Errorf("`outputs` list contains duplicate aggregation function: %s", output)
}
outputsSeen[output] = struct{}{}
if strings.HasPrefix(output, "quantiles(") {
if !strings.HasSuffix(output, ")") {
return nil, fmt.Errorf("missing closing brace for `quantiles()` output")
return nil, false, fmt.Errorf("missing closing brace for `quantiles()` output")
}
argsStr := output[len("quantiles(") : len(output)-1]
if len(argsStr) == 0 {
return nil, fmt.Errorf("`quantiles()` must contain at least one phi")
return nil, false, fmt.Errorf("`quantiles()` must contain at least one phi")
}
args := strings.Split(argsStr, ",")
phis := make([]float64, len(args))
@@ -666,57 +671,57 @@ func newAggrState(output string, outputsSeen map[string]struct{}, stalenessInter
arg = strings.TrimSpace(arg)
phi, err := strconv.ParseFloat(arg, 64)
if err != nil {
return nil, fmt.Errorf("cannot parse phi=%q for quantiles(%s): %w", arg, argsStr, err)
return nil, false, fmt.Errorf("cannot parse phi=%q for quantiles(%s): %w", arg, argsStr, err)
}
if phi < 0 || phi > 1 {
return nil, fmt.Errorf("phi inside quantiles(%s) must be in the range [0..1]; got %v", argsStr, phi)
return nil, false, fmt.Errorf("phi inside quantiles(%s) must be in the range [0..1]; got %v", argsStr, phi)
}
phis[i] = phi
}
if _, ok := outputsSeen["quantiles"]; ok {
return nil, fmt.Errorf("`outputs` list contains duplicated `quantiles()` function, please combine multiple phi* like `quantiles(0.5, 0.9)`")
return nil, false, fmt.Errorf("`outputs` list contains duplicated `quantiles()` function, please combine multiple phi* like `quantiles(0.5, 0.9)`")
}
outputsSeen["quantiles"] = struct{}{}
return newQuantilesAggrState(phis), nil
return newQuantilesAggrState(phis), false, nil
}
switch output {
case "avg":
return newAvgAggrState(), nil
return newAvgAggrState(), false, nil
case "count_samples":
return newCountSamplesAggrState(), nil
return newCountSamplesAggrState(), false, nil
case "count_series":
return newCountSeriesAggrState(), nil
return newCountSeriesAggrState(), true, nil
case "histogram_bucket":
return newHistogramBucketAggrState(stalenessInterval), nil
return newHistogramBucketAggrState(), false, nil
case "increase":
return newTotalAggrState(stalenessInterval, true, true), nil
return newTotalAggrState(true, true), true, nil
case "increase_prometheus":
return newTotalAggrState(stalenessInterval, true, false), nil
return newTotalAggrState(true, false), true, nil
case "last":
return newLastAggrState(), nil
return newLastAggrState(), false, nil
case "max":
return newMaxAggrState(), nil
return newMaxAggrState(), false, nil
case "min":
return newMinAggrState(), nil
return newMinAggrState(), false, nil
case "rate_avg":
return newRateAggrState(stalenessInterval, true), nil
return newRateAggrState(true), true, nil
case "rate_sum":
return newRateAggrState(stalenessInterval, false), nil
return newRateAggrState(false), true, nil
case "stddev":
return newStddevAggrState(), nil
return newStddevAggrState(), false, nil
case "stdvar":
return newStdvarAggrState(), nil
return newStdvarAggrState(), false, nil
case "sum_samples":
return newSumSamplesAggrState(), nil
return newSumSamplesAggrState(), false, nil
case "total":
return newTotalAggrState(stalenessInterval, false, true), nil
return newTotalAggrState(false, true), true, nil
case "total_prometheus":
return newTotalAggrState(stalenessInterval, false, false), nil
return newTotalAggrState(false, false), true, nil
case "unique_samples":
return newUniqueSamplesAggrState(), nil
return newUniqueSamplesAggrState(), false, nil
default:
return nil, fmt.Errorf("unsupported output=%q; supported values: %s; see https://docs.victoriametrics.com/stream-aggregation/", output, supportedOutputs)
return nil, false, fmt.Errorf("unsupported output=%q; supported values: %s; see https://docs.victoriametrics.com/stream-aggregation/", output, supportedOutputs)
}
}
@@ -823,8 +828,10 @@ func (a *aggregator) dedupFlush() {
}
startTime := time.Now()
timestamp := startTime.UnixMilli()
deleteDeadline := timestamp + a.stalenessInterval
a.da.flush(a.pushSamples)
a.da.flush(a.pushSamples, timestamp, deleteDeadline)
d := time.Since(startTime)
a.dedupFlushDuration.Update(d.Seconds())
@@ -902,6 +909,7 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
minTimestamp := a.minTimestamp.Load()
nowMsec := time.Now().UnixMilli()
deleteDeadline := nowMsec + a.stalenessInterval
var maxLagMsec int64
for idx, ts := range tss {
if !a.match.Match(ts.Labels) {
@@ -930,7 +938,7 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
}
bufLen := len(buf)
buf = compressLabels(buf, inputLabels.Labels, outputLabels.Labels)
buf = compressLabels(buf, a.lc, inputLabels.Labels, outputLabels.Labels, a.includeInputKey, deleteDeadline)
// key remains valid only by the end of this function and can't be reused after
// do not intern key because number of unique keys could be too high
key := bytesutil.ToUnsafeString(buf[bufLen:])
@@ -964,52 +972,60 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
ctx.buf = buf
if a.da != nil {
a.da.pushSamples(samples)
a.da.pushSamples(samples, deleteDeadline)
} else {
a.pushSamples(samples)
a.pushSamples(samples, deleteDeadline)
}
}
func compressLabels(dst []byte, inputLabels, outputLabels []prompbmarshal.Label) []byte {
func compressLabels(dst []byte, lc *promutils.LabelsCompressor, inputLabels, outputLabels []prompbmarshal.Label, includeInputKey bool, deleteDeadline int64) []byte {
bb := bbPool.Get()
bb.B = lc.Compress(bb.B, inputLabels)
dst = encoding.MarshalVarUint64(dst, uint64(len(bb.B)))
dst = append(dst, bb.B...)
bbPool.Put(bb)
dst = lc.Compress(dst, outputLabels)
bb.B = lc.Compress(bb.B, outputLabels, deleteDeadline)
if includeInputKey {
dst = encoding.MarshalVarUint64(dst, uint64(len(bb.B)))
dst = append(dst, bb.B...)
bbPool.Put(bb)
dst = lc.Compress(dst, inputLabels, deleteDeadline)
} else {
dst = append(dst, bb.B...)
bbPool.Put(bb)
}
return dst
}
func decompressLabels(dst []prompbmarshal.Label, key string) []prompbmarshal.Label {
func decompressLabels(dst []prompbmarshal.Label, lc *promutils.LabelsCompressor, key string) []prompbmarshal.Label {
return lc.Decompress(dst, bytesutil.ToUnsafeBytes(key))
}
func getOutputKey(key string) string {
func getOutputKey(key string, includeInputKey bool) string {
src := bytesutil.ToUnsafeBytes(key)
inputKeyLen, nSize := encoding.UnmarshalVarUint64(src)
if nSize <= 0 {
logger.Panicf("BUG: cannot unmarshal inputKeyLen from uvarint")
outputKey := src
if includeInputKey {
outputKeyLen, nSize := encoding.UnmarshalVarUint64(src)
if nSize <= 0 {
logger.Panicf("BUG: cannot unmarshal outputKeyLen from uvarint")
}
src = src[nSize:]
outputKey = src[:outputKeyLen]
}
src = src[nSize:]
outputKey := src[inputKeyLen:]
return bytesutil.ToUnsafeString(outputKey)
}
func getInputOutputKey(key string) (string, string) {
src := bytesutil.ToUnsafeBytes(key)
inputKeyLen, nSize := encoding.UnmarshalVarUint64(src)
outputKeyLen, nSize := encoding.UnmarshalVarUint64(src)
if nSize <= 0 {
logger.Panicf("BUG: cannot unmarshal inputKeyLen from uvarint")
logger.Panicf("BUG: cannot unmarshal outputKeyLen from uvarint")
}
src = src[nSize:]
inputKey := src[:inputKeyLen]
outputKey := src[inputKeyLen:]
outputKey := src[:outputKeyLen]
inputKey := src[outputKeyLen:]
return bytesutil.ToUnsafeString(inputKey), bytesutil.ToUnsafeString(outputKey)
}
func (a *aggregator) pushSamples(samples []pushSample) {
func (a *aggregator) pushSamples(samples []pushSample, deleteDeadline int64) {
for _, ao := range a.aggrOutputs {
ao.as.pushSamples(samples)
ao.as.pushSamples(samples, deleteDeadline, a.includeInputKey)
}
}
@@ -1169,7 +1185,7 @@ func (ctx *flushCtx) flushSeries() {
func (ctx *flushCtx) appendSeries(key, suffix string, value float64) {
labelsLen := len(ctx.labels)
samplesLen := len(ctx.samples)
ctx.labels = decompressLabels(ctx.labels, key)
ctx.labels = decompressLabels(ctx.labels, ctx.a.lc, key)
if !ctx.a.keepMetricNames {
ctx.labels = addMetricSuffix(ctx.labels, labelsLen, ctx.a.suffix, suffix)
}
@@ -1191,7 +1207,7 @@ func (ctx *flushCtx) appendSeries(key, suffix string, value float64) {
func (ctx *flushCtx) appendSeriesWithExtraLabel(key, suffix string, value float64, extraName, extraValue string) {
labelsLen := len(ctx.labels)
samplesLen := len(ctx.samples)
ctx.labels = decompressLabels(ctx.labels, key)
ctx.labels = decompressLabels(ctx.labels, ctx.a.lc, key)
if !ctx.a.keepMetricNames {
ctx.labels = addMetricSuffix(ctx.labels, labelsLen, ctx.a.suffix, suffix)
}

View File

@@ -12,41 +12,46 @@ type sumSamplesAggrState struct {
}
type sumSamplesStateValue struct {
mu sync.Mutex
sum float64
deleted bool
mu sync.Mutex
sum float64
deleted bool
defined bool
deleteDeadline int64
}
func newSumSamplesAggrState() *sumSamplesAggrState {
return &sumSamplesAggrState{}
}
func (as *sumSamplesAggrState) pushSamples(samples []pushSample) {
func (as *sumSamplesAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &sumSamplesStateValue{
sum: s.value,
}
v = &sumSamplesStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The new entry has been successfully created.
continue
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
// Use the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*sumSamplesStateValue)
sv.mu.Lock()
if !sv.defined {
sv.defined = true
}
deleted := sv.deleted
if !deleted {
sv.sum += s.value
sv.deleteDeadline = deleteDeadline
if !sv.defined {
sv.defined = true
}
}
sv.mu.Unlock()
if deleted {
@@ -65,9 +70,22 @@ func (as *sumSamplesAggrState) flushState(ctx *flushCtx) {
sv := v.(*sumSamplesStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
if !sv.defined {
sv.mu.Unlock()
return true
}
sum := sv.sum
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.defined = false
sv.sum = 0
sv.mu.Unlock()
key := k.(string)

View File

@@ -3,10 +3,9 @@ package streamaggr
import (
"math"
"sync"
"time"
"sync/atomic"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
)
// totalAggrState calculates output=total, total_prometheus, increase and increase_prometheus.
@@ -19,48 +18,37 @@ type totalAggrState struct {
// Whether to take into account the first sample in new time series when calculating the output value.
keepFirstSample bool
// Time series state is dropped if no new samples are received during stalenessSecs.
//
// Aslo, the first sample per each new series is ignored during stalenessSecs even if keepFirstSample is set.
// see ignoreFirstSampleDeadline for more details.
stalenessSecs uint64
// The first sample per each new series is ignored until this unix timestamp deadline in seconds even if keepFirstSample is set.
// The first sample per each new series is ignored first two intervals
// This allows avoiding an initial spike of the output values at startup when new time series
// cannot be distinguished from already existing series. This is tracked with ignoreFirstSampleDeadline.
ignoreFirstSampleDeadline uint64
// cannot be distinguished from already existing series. This is tracked with ignoreFirstSamples.
ignoreFirstSamples atomic.Int32
}
type totalStateValue struct {
mu sync.Mutex
lastValues map[string]totalLastValueState
total float64
deleteDeadline uint64
deleteDeadline int64
deleted bool
}
type totalLastValueState struct {
value float64
timestamp int64
deleteDeadline uint64
deleteDeadline int64
}
func newTotalAggrState(stalenessInterval time.Duration, resetTotalOnFlush, keepFirstSample bool) *totalAggrState {
stalenessSecs := roundDurationToSecs(stalenessInterval)
ignoreFirstSampleDeadline := fasttime.UnixTimestamp() + stalenessSecs
return &totalAggrState{
resetTotalOnFlush: resetTotalOnFlush,
keepFirstSample: keepFirstSample,
stalenessSecs: stalenessSecs,
ignoreFirstSampleDeadline: ignoreFirstSampleDeadline,
func newTotalAggrState(resetTotalOnFlush, keepFirstSample bool) *totalAggrState {
as := &totalAggrState{
resetTotalOnFlush: resetTotalOnFlush,
keepFirstSample: keepFirstSample,
}
as.ignoreFirstSamples.Store(2)
return as
}
func (as *totalAggrState) pushSamples(samples []pushSample) {
currentTime := fasttime.UnixTimestamp()
deleteDeadline := currentTime + as.stalenessSecs
keepFirstSample := as.keepFirstSample && currentTime > as.ignoreFirstSampleDeadline
func (as *totalAggrState) pushSamples(samples []pushSample, deleteDeadline int64, _ bool) {
keepFirstSample := as.keepFirstSample && as.ignoreFirstSamples.Load() <= 0
for i := range samples {
s := &samples[i]
inputKey, outputKey := getInputOutputKey(s.key)
@@ -116,11 +104,9 @@ func (as *totalAggrState) pushSamples(samples []pushSample) {
}
func (as *totalAggrState) flushState(ctx *flushCtx) {
currentTime := fasttime.UnixTimestamp()
suffix := as.getSuffix()
as.removeOldEntries(currentTime)
as.removeOldEntries(ctx)
m := &as.m
m.Range(func(k, v any) bool {
@@ -143,6 +129,10 @@ func (as *totalAggrState) flushState(ctx *flushCtx) {
}
return true
})
ignoreFirstSamples := as.ignoreFirstSamples.Load()
if ignoreFirstSamples > 0 {
as.ignoreFirstSamples.Add(-1)
}
}
func (as *totalAggrState) getSuffix() string {
@@ -159,16 +149,18 @@ func (as *totalAggrState) getSuffix() string {
return "total_prometheus"
}
func (as *totalAggrState) removeOldEntries(currentTime uint64) {
func (as *totalAggrState) removeOldEntries(ctx *flushCtx) {
m := &as.m
m.Range(func(k, v any) bool {
sv := v.(*totalStateValue)
sv.mu.Lock()
if currentTime > sv.deleteDeadline {
if ctx.flushTimestamp > sv.deleteDeadline {
// Mark the current entry as deleted
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
@@ -176,7 +168,7 @@ func (as *totalAggrState) removeOldEntries(currentTime uint64) {
// Delete outdated entries in sv.lastValues
lvs := sv.lastValues
for k1, lv := range lvs {
if currentTime > lv.deleteDeadline {
if ctx.flushTimestamp > lv.deleteDeadline {
delete(lvs, k1)
}
}

View File

@@ -12,37 +12,34 @@ type uniqueSamplesAggrState struct {
}
type uniqueSamplesStateValue struct {
mu sync.Mutex
m map[float64]struct{}
deleted bool
mu sync.Mutex
m map[float64]struct{}
deleted bool
deleteDeadline int64
}
func newUniqueSamplesAggrState() *uniqueSamplesAggrState {
return &uniqueSamplesAggrState{}
}
func (as *uniqueSamplesAggrState) pushSamples(samples []pushSample) {
func (as *uniqueSamplesAggrState) pushSamples(samples []pushSample, deleteDeadline int64, includeInputKey bool) {
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
outputKey := getOutputKey(s.key, includeInputKey)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &uniqueSamplesStateValue{
m: map[float64]struct{}{
s.value: {},
},
m: make(map[float64]struct{}),
}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if !loaded {
// The new entry has been successfully created.
continue
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
// Use the entry created by a concurrent goroutine.
v = vNew
}
sv := v.(*uniqueSamplesStateValue)
sv.mu.Lock()
@@ -51,6 +48,7 @@ func (as *uniqueSamplesAggrState) pushSamples(samples []pushSample) {
if _, ok := sv.m[s.value]; !ok {
sv.m[s.value] = struct{}{}
}
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
@@ -69,13 +67,21 @@ func (as *uniqueSamplesAggrState) flushState(ctx *flushCtx) {
sv := v.(*uniqueSamplesStateValue)
sv.mu.Lock()
if ctx.flushTimestamp > sv.deleteDeadline {
sv.deleted = true
sv.mu.Unlock()
key := k.(string)
ctx.a.lc.Delete(bytesutil.ToUnsafeBytes(key), ctx.flushTimestamp)
m.Delete(k)
return true
}
n := len(sv.m)
// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
sv.deleted = true
sv.m = make(map[float64]struct{})
sv.mu.Unlock()
key := k.(string)
ctx.appendSeries(key, "unique_samples", float64(n))
if n > 0 {
key := k.(string)
ctx.appendSeries(key, "unique_samples", float64(n))
}
return true
})
}