metrics: add StaleRegionFromPDCounter (#1530)

ref tikv/pd#7431 Signed-off-by: okJiang <[email protected]>
tikv · Dec 19, 2024 · 2572ee4 · 2572ee4
1 parent 8e0275c
commit 2572ee4
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 0 deletions.
diff --git a/internal/locate/region_cache.go b/internal/locate/region_cache.go
@@ -1902,6 +1902,7 @@ func (mu *regionIndexMu) insertRegionToCache(cachedRegion *Region, invalidateOld
 	// and there is the synchronization time between the pd follower and the leader.
 	// So we should check the epoch.
 	if ok && (oldVer.GetVer() > newVer.GetVer() || oldVer.GetConfVer() > newVer.GetConfVer()) {
+		metrics.TiKVStaleRegionFromPDCounter.Inc()
 		logutil.BgLogger().Debug("get stale region",
 			zap.Uint64("region", newVer.GetID()), zap.Uint64("new-ver", newVer.GetVer()), zap.Uint64("new-conf", newVer.GetConfVer()),
 			zap.Uint64("old-ver", oldVer.GetVer()), zap.Uint64("old-conf", oldVer.GetConfVer()))

diff --git a/internal/locate/sorted_btree.go b/internal/locate/sorted_btree.go
@@ -40,6 +40,7 @@ import (
 
 	"github.com/google/btree"
 	"github.com/tikv/client-go/v2/internal/logutil"
+	"github.com/tikv/client-go/v2/metrics"
 	"go.uber.org/zap"
 )
 
@@ -112,6 +113,7 @@ func (s *SortedRegions) removeIntersecting(r *Region, verID RegionVerID) ([]*btr
 			return false
 		}
 		if item.cachedRegion.meta.GetRegionEpoch().GetVersion() > verID.ver {
+			metrics.TiKVStaleRegionFromPDCounter.Inc()
 			logutil.BgLogger().Debug("get stale region",
 				zap.Uint64("region", verID.GetID()), zap.Uint64("ver", verID.GetVer()), zap.Uint64("conf", verID.GetConfVer()),
 				zap.Uint64("intersecting-ver", item.cachedRegion.meta.GetRegionEpoch().GetVersion()))

diff --git a/metrics/metrics.go b/metrics/metrics.go
@@ -117,6 +117,7 @@ var (
 	TiKVPipelinedFlushDuration                     prometheus.Histogram
 	TiKVValidateReadTSFromPDCount                  prometheus.Counter
 	TiKVLowResolutionTSOUpdateIntervalSecondsGauge prometheus.Gauge
+	TiKVStaleRegionFromPDCounter                   prometheus.Counter
 )
 
 // Label constants.
@@ -852,6 +853,14 @@ func initMetrics(namespace, subsystem string, constLabels prometheus.Labels) {
 			Help:      "The actual working update interval for the low resolution TSO. As there are adaptive mechanism internally, this value may differ from the config.",
 		})
 
+	TiKVStaleRegionFromPDCounter = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "stale_region_from_pd",
+			Help:      "Counter of stale region from PD",
+		})
+
 	initShortcuts()
 }
 
@@ -948,6 +957,7 @@ func RegisterMetrics() {
 	prometheus.MustRegister(TiKVPipelinedFlushDuration)
 	prometheus.MustRegister(TiKVValidateReadTSFromPDCount)
 	prometheus.MustRegister(TiKVLowResolutionTSOUpdateIntervalSecondsGauge)
+	prometheus.MustRegister(TiKVStaleRegionFromPDCounter)
 }
 
 // readCounter reads the value of a prometheus.Counter.