From bd05e2d37900472c66bc51b00f282e1d85fa038c Mon Sep 17 00:00:00 2001
From: nolouch <nolouch@gmail.com>
Date: Mon, 4 Mar 2024 19:56:49 +0800
Subject: [PATCH] seperate the lock metrics

Signed-off-by: nolouch <nolouch@gmail.com>
---
 pkg/core/metrics.go | 26 ++++++++++++++++++++++++--
 pkg/core/region.go  | 41 ++++++++++++++++++++++++++---------------
 2 files changed, 50 insertions(+), 17 deletions(-)

diff --git a/pkg/core/metrics.go b/pkg/core/metrics.go
index 4928df8d0979..6f75c70c100b 100644
--- a/pkg/core/metrics.go
+++ b/pkg/core/metrics.go
@@ -40,8 +40,28 @@ var (
 			Help:      "Bucketed histogram of processing count of handle the heartbeat stage.",
 		}, []string{"name"})
 
-	waitLockDurationSum       = HeartbeatBreakdownHandleDurationSum.WithLabelValues("WaitLock")
-	waitLockCount             = HeartbeatBreakdownHandleCount.WithLabelValues("WaitLock")
+	AcquireRegionsLockWaitDurationSum = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "pd",
+			Subsystem: "core",
+			Name:      "acquire_regions_lock_wait_duration_seconds_sum",
+			Help:      "Bucketed histogram of processing time (s) of waiting for acquiring regions lock.",
+		}, []string{"type"})
+	AcquireRegionsLockWaitCount = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "pd",
+			Subsystem: "core",
+			Name:      "acquire_regions_lock_wait_duration_seconds_count",
+			Help:      "Bucketed histogram of processing count of waiting for acquiring regions lock.",
+		}, []string{"type"})
+
+	// lock statistics
+	waitRegionsLockDurationSum    = AcquireRegionsLockWaitDurationSum.WithLabelValues("WaitRegionsLock")
+	waitRegionsLockCount          = AcquireRegionsLockWaitCount.WithLabelValues("WaitRegionsLock")
+	waitSubRegionsLockDurationSum = AcquireRegionsLockWaitDurationSum.WithLabelValues("WaitSubRegionsLock")
+	waitSubRegionsLockCount       = AcquireRegionsLockWaitCount.WithLabelValues("WaitSubRegionsLock")
+
+	// heartbeat breakdown statistics
 	preCheckDurationSum       = HeartbeatBreakdownHandleDurationSum.WithLabelValues("PreCheck")
 	preCheckCount             = HeartbeatBreakdownHandleCount.WithLabelValues("PreCheck")
 	asyncHotStatsDurationSum  = HeartbeatBreakdownHandleDurationSum.WithLabelValues("AsyncHotStatsDuration")
@@ -63,6 +83,8 @@ var (
 func init() {
 	prometheus.MustRegister(HeartbeatBreakdownHandleDurationSum)
 	prometheus.MustRegister(HeartbeatBreakdownHandleCount)
+	prometheus.MustRegister(AcquireRegionsLockWaitDurationSum)
+	prometheus.MustRegister(AcquireRegionsLockWaitCount)
 }
 
 type saveCacheStats struct {
diff --git a/pkg/core/region.go b/pkg/core/region.go
index c4c8d53abd9e..6c058fd424f4 100644
--- a/pkg/core/region.go
+++ b/pkg/core/region.go
@@ -1699,23 +1699,34 @@ const magicCount = 15 * time.Second
 
 // CollectWaitLockMetrics collects the metrics of waiting time for lock
 func (r *RegionsInfo) CollectWaitLockMetrics() {
-	sTotalTime := atomic.LoadInt64(&r.t.totalWaitTime)
-	stTotalTime := atomic.LoadInt64(&r.st.totalWaitTime)
-	sLockCount := atomic.LoadInt64(&r.t.lockCount)
-	stLockCount := atomic.LoadInt64(&r.st.lockCount)
-	lastTotalWaitTime := atomic.LoadInt64(&r.t.lastTotalWaitTime) + atomic.LoadInt64(&r.st.lastTotalWaitTime)
-	lastLockCount := atomic.LoadInt64(&r.t.lastLockCount) + atomic.LoadInt64(&r.st.lastLockCount)
-	totalLockCount := sLockCount + stLockCount
-	totalWaitTime := sTotalTime + stTotalTime
-	atomic.StoreInt64(&r.t.lastTotalWaitTime, sTotalTime)
-	atomic.StoreInt64(&r.t.lastLockCount, sLockCount)
-	atomic.StoreInt64(&r.st.lastTotalWaitTime, stTotalTime)
-	atomic.StoreInt64(&r.st.lastLockCount, stLockCount)
-	if lastTotalWaitTime == 0 || lastLockCount == 0 || totalLockCount-lastLockCount < 0 || totalLockCount-lastLockCount > int64(magicCount) {
+	regionsLockTotalWaitTime := atomic.LoadInt64(&r.t.totalWaitTime)
+	regionsLockCount := atomic.LoadInt64(&r.t.lockCount)
+
+	subRegionsLockTotalWaitTime := atomic.LoadInt64(&r.st.totalWaitTime)
+	subRegionsLockCount := atomic.LoadInt64(&r.st.lockCount)
+
+	lastRegionsLockTotalWaitTime := atomic.LoadInt64(&r.t.lastTotalWaitTime)
+	lastsRegionsLockCount := atomic.LoadInt64(&r.t.lastLockCount)
+
+	lastSubRegionsLockTotalWaitTime := atomic.LoadInt64(&r.st.lastTotalWaitTime)
+	lastSubRegionsLockCount := atomic.LoadInt64(&r.st.lastLockCount)
+
+	// skip invalid situation like initial status
+	if lastRegionsLockTotalWaitTime == 0 || lastsRegionsLockCount == 0 || lastSubRegionsLockTotalWaitTime == 0 || lastSubRegionsLockCount == 0 ||
+		regionsLockTotalWaitTime-lastRegionsLockTotalWaitTime < 0 || regionsLockCount-lastsRegionsLockCount > int64(magicCount) ||
+		subRegionsLockTotalWaitTime-lastSubRegionsLockTotalWaitTime < 0 || subRegionsLockCount-lastSubRegionsLockCount > int64(magicCount) {
 		return
 	}
-	waitLockDurationSum.Add(time.Duration(totalWaitTime - lastTotalWaitTime).Seconds())
-	waitLockCount.Add(float64(totalLockCount - lastLockCount))
+
+	waitRegionsLockDurationSum.Add(float64(regionsLockTotalWaitTime - lastRegionsLockTotalWaitTime))
+	waitRegionsLockCount.Add(float64(regionsLockCount - lastsRegionsLockCount))
+	waitSubRegionsLockDurationSum.Add(float64(subRegionsLockTotalWaitTime - lastSubRegionsLockTotalWaitTime))
+	waitSubRegionsLockCount.Add(float64(subRegionsLockCount - lastSubRegionsLockCount))
+
+	atomic.StoreInt64(&r.t.lastTotalWaitTime, regionsLockTotalWaitTime)
+	atomic.StoreInt64(&r.t.lastLockCount, regionsLockCount)
+	atomic.StoreInt64(&r.st.lastTotalWaitTime, subRegionsLockTotalWaitTime)
+	atomic.StoreInt64(&r.st.lastLockCount, subRegionsLockCount)
 }
 
 // GetAdjacentRegions returns region's info that is adjacent with specific region