save

Author: AskAlexSharov

db/kv/kv_interface.go

@@ -603,9 +603,6 @@ type DBGauges struct { // these gauges are shared by all MDBX instances, but nee
 	GcLeafMetric     *metrics.GaugeVec
 	GcOverflowMetric *metrics.GaugeVec
 	GcPagesMetric    *metrics.GaugeVec
-
-	GCMaxRetainedPages *metrics.GaugeVec
-	GCMaxReaderLag     *metrics.GaugeVec
 }
 
 type DBSummaries struct { // the summaries are particular to a DB instance
@@ -616,6 +613,11 @@ type DBSummaries struct { // the summaries are particular to a DB instance
 	DbCommitTotal       metrics.Summary
 }
 
+type GCSummaries struct { // GC metric summaries for each DB instance
+	GCMaxRetainedPages metrics.Summary
+	GCMaxReaderLag     metrics.Summary
+}
+
 // InitMDBXMGauges this only needs to be called once during startup
 func InitMDBXMGauges() *DBGauges {
 	return &DBGauges{
@@ -638,9 +640,6 @@ func InitMDBXMGauges() *DBGauges {
 		GcLeafMetric:     metrics.GetOrCreateGaugeVec(`db_gc_leaf`, []string{dbLabelName}),
 		GcOverflowMetric: metrics.GetOrCreateGaugeVec(`db_gc_overflow`, []string{dbLabelName}),
 		GcPagesMetric:    metrics.GetOrCreateGaugeVec(`db_gc_pages`, []string{dbLabelName}),
-
-		GCMaxRetainedPages: metrics.GetOrCreateGaugeVec(`db_gc_max_retained_pages`, []string{dbLabelName}),
-		GCMaxReaderLag:     metrics.GetOrCreateGaugeVec(`db_gc_max_reader_lag`, []string{dbLabelName}),
 	}
 }
 
@@ -658,8 +657,20 @@ func InitSummaries(dbLabel Label) {
 	}
 }
 
+func InitGCSummaries(dbLabel Label) {
+	_, ok := MDBXGCSummaries.Load(dbLabel)
+	if !ok {
+		dbName := string(dbLabel)
+		MDBXGCSummaries.Store(dbName, &GCSummaries{
+			GCMaxRetainedPages: metrics.GetOrCreateSummaryWithLabels(`db_gc_max_retained_pages`, []string{dbLabelName}, []string{dbName}),
+			GCMaxReaderLag:     metrics.GetOrCreateSummaryWithLabels(`db_gc_max_reader_lag`, []string{dbLabelName}, []string{dbName}),
+		})
+	}
+}
+
 var MDBXGauges = InitMDBXMGauges() // global mdbx gauges. each gauge can be filtered by db name
 var MDBXSummaries sync.Map         // dbName => Summaries mapping
+var MDBXGCSummaries sync.Map       // dbName => GCSummaries mapping
 
 var (
 	ErrAttemptToDeleteNonDeprecatedBucket = errors.New("only buckets from dbutils.ChaindataDeprecatedTables can be deleted")

db/kv/mdbx/kv_mdbx.go

@@ -192,6 +192,7 @@ func (opts MdbxOpts) Open(ctx context.Context) (kv.RwDB, error) {
 
 	if opts.metrics {
 		kv.InitSummaries(opts.label)
+		kv.InitGCSummaries(opts.label)
 	}
 	if opts.HasFlag(mdbx.Accede) || opts.HasFlag(mdbx.Readonly) {
 		for retry := 0; ; retry++ {
@@ -1025,10 +1026,21 @@ func (tx *MdbxTx) Commit() error {
 			tx.db.opts.log.Error("failed to record mdbx summaries", "err", err)
 		}
 
-		kv.MDBXGauges.GCMaxRetainedPages.WithLabelValues(string(dbLabel)).SetUint64(uint64(latency.GCDetails.MaxRetainedPages))
-		kv.MDBXGauges.GCMaxReaderLag.WithLabelValues(string(dbLabel)).SetUint64(uint64(latency.GCDetails.MaxReaderLag))
+		// Record GC stats as Summary metrics to capture spike distribution
+		// Summary naturally aggregates all observations over time windows, showing when spikes happened
+		pages := float64(latency.GCDetails.MaxRetainedPages)
+		lag := float64(latency.GCDetails.MaxReaderLag)
+
+		// Get GC summaries for this DB instance
+		gcSummariesIface, ok := kv.MDBXGCSummaries.Load(string(dbLabel))
+		if ok {
+			gcSummaries := gcSummariesIface.(*kv.GCSummaries)
+			gcSummaries.GCMaxRetainedPages.Observe(pages)
+			gcSummaries.GCMaxReaderLag.Observe(lag)
+		}
+
 		if latency.GCDetails.MaxRetainedPages > 0 || latency.GCDetails.MaxReaderLag > 0 {
-			log.Warn("[dbg] latency.GCDetails", "MaxRetainedPages", latency.GCDetails.MaxRetainedPages, "MaxRetainedPages", latency.GCDetails.MaxRetainedPages)
+			log.Warn("[dbg] latency.GCDetails", "MaxRetainedPages", latency.GCDetails.MaxRetainedPages, "MaxReaderLag", latency.GCDetails.MaxReaderLag)
 		}
 
 		//kv.DbGcWorkPnlMergeTime.Update(latency.GCDetails.WorkPnlMergeTime.Seconds())