metrics(ticdc): refine exp boundary, record large row from 2k size (#…

…7861) ref #5928
pingcap · Dec 9, 2022 · 7aef2a0 · 7aef2a0
1 parent d651082
commit 7aef2a0
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 9 deletions.
diff --git a/cdc/sinkv2/metrics/metrics.go b/cdc/sinkv2/metrics/metrics.go
@@ -20,8 +20,8 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-// rowSizeLowBound is set to 128K, only track data event with size not smaller than it.
-const largeRowSizeLowBound = 128 * 1024
+// rowSizeLowBound is set to 2K, only track data event with size not smaller than it.
+const largeRowSizeLowBound = 2 * 1024
 
 // ---------- Metrics used in Statistics. ---------- //
 var (
@@ -42,7 +42,7 @@ var (
 			Subsystem: "sinkv2",
 			Name:      "large_row_size",
 			Help:      "The size of all received row changed events (in bytes).",
-			Buckets:   prometheus.ExponentialBuckets(largeRowSizeLowBound, 2, 10), // 128K~128M
+			Buckets:   prometheus.ExponentialBuckets(largeRowSizeLowBound, 2, 15), // 2K~32M
 		}, []string{"namespace", "changefeed", "type"}) // type is for `sinkType`
 
 	// ExecDDLHistogram records the exexution time of a DDL.

diff --git a/cdc/sinkv2/metrics/mq/metrics.go b/cdc/sinkv2/metrics/mq/metrics.go
@@ -26,7 +26,7 @@ var (
 			Subsystem: "sinkv2",
 			Name:      "mq_worker_send_message_duration",
 			Help:      "Send Message duration(s) for MQ worker.",
-			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 20), // 1ms~1000s
+			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 20), // 1ms~524s
 		}, []string{"namespace", "changefeed"})
 	// WorkerBatchSize record the size of each batched messages.
 	WorkerBatchSize = prometheus.NewHistogramVec(

diff --git a/cdc/sinkv2/metrics/txn/metrics.go b/cdc/sinkv2/metrics/txn/metrics.go
@@ -24,7 +24,7 @@ var (
 			Subsystem: "sinkv2",
 			Name:      "txn_conflict_detect_duration",
 			Help:      "Bucketed histogram of conflict detect time (s) for single DML statement.",
-			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 20), // 1ms~1000s
+			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 20), // 1ms~524s
 		}, []string{"namespace", "changefeed"})
 
 	WorkerFlushDuration = prometheus.NewHistogramVec(
@@ -33,7 +33,7 @@ var (
 			Subsystem: "sinkv2",
 			Name:      "txn_worker_flush_duration",
 			Help:      "Flush duration (s) for txn worker.",
-			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 20), // 1ms~1000s
+			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 20), // 1ms~524s
 		}, []string{"namespace", "changefeed"})
 
 	WorkerBusyRatio = prometheus.NewCounterVec(
@@ -58,7 +58,7 @@ var (
 			Subsystem: "sinkv2",
 			Name:      "txn_sink_dml_batch_commit",
 			Help:      "Duration of committing a DML batch",
-			Buckets:   prometheus.ExponentialBuckets(0.01, 2, 18), // 10ms~1000s
+			Buckets:   prometheus.ExponentialBuckets(0.01, 2, 18), // 10ms~1310s
 		}, []string{"namespace", "changefeed"})
 
 	SinkDMLBatchCallback = prometheus.NewHistogramVec(
@@ -67,7 +67,7 @@ var (
 			Subsystem: "sinkv2",
 			Name:      "txn_sink_dml_batch_callback",
 			Help:      "Duration of execuing a batch of callbacks",
-			Buckets:   prometheus.ExponentialBuckets(0.01, 2, 18), // 10ms~1000s
+			Buckets:   prometheus.ExponentialBuckets(0.01, 2, 18), // 10ms~1300s
 		}, []string{"namespace", "changefeed"})
 )
 

diff --git a/metrics/grafana/ticdc.json b/metrics/grafana/ticdc.json
@@ -7038,7 +7038,7 @@
           "dashLength": 10,
           "dashes": false,
           "datasource": "${DS_TEST-CLUSTER}",
-          "description": "Size of large rows (size >= 128K).",
+          "description": "Size of large rows (size >= 2K).",
           "fieldConfig": {
             "defaults": {},
             "overrides": []