statistics: increase batchInsertSize from 10 to 8192 for improved performance (#60405) (#60414)

close #60183

Author: ti-chi-bot

pkg/metrics/grafana/tidb.json

@@ -17911,16 +17911,24 @@
               "expr": "histogram_quantile(0.99, sum(rate(tidb_statistics_stats_delta_update_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\"}[1m])) by (le))",
               "interval": "",
               "intervalFactor": 2,
-              "legendFormat": "99",
+              "legendFormat": "stats-meta-99",
               "queryType": "randomWalk",
               "refId": "A"
+            },
+            {
+              "exemplar": true,
+              "expr": "histogram_quantile(0.99, sum(rate(tidb_statistics_stats_usage_update_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\"}[1m])) by (le))",
+              "hide": false,
+              "interval": "",
+              "legendFormat": "stats-usage-99",
+              "refId": "B"
             }
           ],
           "thresholds": [],
           "timeFrom": null,
           "timeRegions": [],
           "timeShift": null,
-          "title": "Stats Meta Updating Duration",
+          "title": "Stats Meta/Usage Updating Duration",
           "tooltip": {
             "shared": true,
             "sort": 0,

pkg/metrics/metrics.go

@@ -240,6 +240,7 @@ func RegisterMetrics() {
 	prometheus.MustRegister(StatsHealthyGauge)
 	prometheus.MustRegister(StatsDeltaLoadHistogram)
 	prometheus.MustRegister(StatsDeltaUpdateHistogram)
+	prometheus.MustRegister(StatsUsageUpdateHistogram)
 	prometheus.MustRegister(TxnStatusEnteringCounter)
 	prometheus.MustRegister(TxnDurationHistogram)
 	prometheus.MustRegister(LastCheckpoint)

pkg/metrics/stats.go

@@ -34,6 +34,7 @@ var (
 	StatsHealthyGauge         *prometheus.GaugeVec
 	StatsDeltaLoadHistogram   prometheus.Histogram
 	StatsDeltaUpdateHistogram prometheus.Histogram
+	StatsUsageUpdateHistogram prometheus.Histogram
 
 	HistoricalStatsCounter        *prometheus.CounterVec
 	PlanReplayerTaskCounter       *prometheus.CounterVec
@@ -162,4 +163,13 @@ func InitStatsMetrics() {
 			Buckets:   prometheus.ExponentialBuckets(0.01, 2, 24), // 10ms ~ 24h
 		},
 	)
+	StatsUsageUpdateHistogram = NewHistogram(
+		prometheus.HistogramOpts{
+			Namespace: "tidb",
+			Subsystem: "statistics",
+			Name:      "stats_usage_update_duration_seconds",
+			Help:      "Bucketed histogram of processing time for the background stats usage update job",
+			Buckets:   prometheus.ExponentialBuckets(0.01, 2, 24), // 10ms ~ 24h
+		},
+	)
 }

pkg/statistics/handle/usage/session_stats_collect.go

@@ -45,8 +45,8 @@ var (
 	// dumpStatsMaxDuration is the max duration since last update.
 	dumpStatsMaxDuration = 5 * time.Minute
 
-	// batchInsertSize is the batch size used by internal SQL to insert values to some system table.
-	batchInsertSize = 10
+	// batchInsertSize is the batch size used by internal SQL to insert values to stats usage table.
+	batchInsertSize = 8192
 )
 
 // needDumpStatsDelta checks whether to dump stats delta.
@@ -314,6 +314,11 @@ func (s *statsUsageImpl) dumpStatsDeltaToKV(
 // DumpColStatsUsageToKV sweeps the whole list, updates the column stats usage map and dumps it to KV.
 func (s *statsUsageImpl) DumpColStatsUsageToKV() error {
 	defer util.Recover(metrics.LabelStats, "DumpColStatsUsageToKV", nil, false)
+	start := time.Now()
+	defer func() {
+		dur := time.Since(start)
+		metrics.StatsUsageUpdateHistogram.Observe(dur.Seconds())
+	}()
 	s.SweepSessionStatsList()
 	colMap := s.SessionStatsUsage().GetUsageAndReset()
 	defer func() {