add monitor

Signed-off-by: Ryan Leung <rleungx@gmail.com>
tikv · Dec 12, 2022 · d3f1da6 · d3f1da6
1 parent 1f13a24
commit d3f1da6
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 3 deletions.
diff --git a/metrics/grafana/pd.json b/metrics/grafana/pd.json
@@ -796,7 +796,7 @@
           "intervalFactor": 2,
           "legendFormat": "{{type}}",
           "refId": "B"
-        },
+        }
       ],
       "thresholds": [
         {

diff --git a/server/cluster/coordinator.go b/server/cluster/coordinator.go
@@ -52,8 +52,8 @@ const (
 	maxScheduleRetries         = 10
 	maxLoadConfigRetries       = 10
 
-	patrolScanRegionLimit = 128 // It takes about 14 minutes to iterate 1 million regions.
-	statsScanRegionLimit  = 1000
+	patrolScanRegionLimit = 128  // It takes about 14 minutes to iterate 1 million regions.
+	statsScanRegionLimit  = 1000 // It takes about 3.5 minutes to iterate 2 million regions.
 	defaultScrapInterval  = 15 * time.Second
 	// PluginLoad means action for load plugin
 	PluginLoad = "PluginLoad"
@@ -231,6 +231,7 @@ func (c *coordinator) collectRegionStats() {
 				case <-timer.C:
 				}
 			}
+			collectRegionStatsGauge.Set(time.Since(start).Seconds())
 			start = time.Now()
 		}
 	}

diff --git a/server/cluster/metrics.go b/server/cluster/metrics.go
@@ -73,6 +73,14 @@ var (
 			Help:      "Time spent of patrol checks region.",
 		})
 
+	collectRegionStatsGauge = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Namespace: "pd",
+			Subsystem: "statistics",
+			Name:      "collect_region_stats_time",
+			Help:      "Time spent of collecting region stats.",
+		})
+
 	updateStoreStatsGauge = prometheus.NewGauge(
 		prometheus.GaugeOpts{
 			Namespace: "pd",
@@ -152,4 +160,5 @@ func init() {
 	prometheus.MustRegister(storesETAGauge)
 	prometheus.MustRegister(storeSyncConfigEvent)
 	prometheus.MustRegister(updateStoreStatsGauge)
+	prometheus.MustRegister(collectRegionStatsGauge)
 }