placement: add rule/group count metrics

Signed-off-by: nolouch <nolouch@gmail.com>
nolouch · Oct 19, 2023 · 4bc0ab7 · 4bc0ab7
1 parent cb9c70c
commit 4bc0ab7
Show file tree

Hide file tree

Showing 5 changed files with 149 additions and 2 deletions.
diff --git a/metrics/grafana/pd.json b/metrics/grafana/pd.json
@@ -1139,6 +1139,111 @@
       "timeFrom": null,
       "timeShift": null
     },
+    {
+      "bars": false,
+      "cacheTimeout": null,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": "${DS_TEST-CLUSTER}",
+      "description": "The current peer count of the cluster",
+      "fieldConfig": {
+        "defaults": {},
+        "overrides": []
+      },
+      "fill": 1,
+      "fillGradient": 0,
+      "gridPos": {
+        "h": 6,
+        "w": 4,
+        "x": 16,
+        "y": 13
+      },
+      "hiddenSeries": false,
+      "id": 22,
+      "interval": null,
+      "legend": {
+        "alignAsTable": true,
+        "avg": false,
+        "current": true,
+        "max": false,
+        "min": false,
+        "rightSide": true,
+        "show": true,
+        "total": false,
+        "values": true
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "maxDataPoints": 100,
+      "nullPointMode": "null",
+      "options": {
+        "alertThreshold": true
+      },
+      "percentage": false,
+      "pluginVersion": "7.5.10",
+      "pointradius": 2,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "exemplar": true,
+          "expr": "sum(pd_rule_manager_status{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\"}) by (type)",
+          "format": "time_series",
+          "interval": "",
+          "intervalFactor": 2,
+          "legendFormat": "{{type}}",
+          "refId": "A",
+          "step": 4
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeRegions": [],
+      "timeShift": null,
+      "title": "Placement Rules Status",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "$$hashKey": "object:192",
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        },
+        {
+          "$$hashKey": "object:193",
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        }
+      ],
+      "yaxis": {
+        "align": false,
+        "alignLevel": null
+      }
+    },
     {
       "collapsed": true,
       "gridPos": {

diff --git a/pkg/schedule/placement/rule_manager.go b/pkg/schedule/placement/rule_manager.go
@@ -321,6 +321,20 @@ func (m *RuleManager) GetAllRules() []*Rule {
 	return rules
 }
 
+// GetRulesCount returns the number of rules.
+func (m *RuleManager) GetRulesCount() int {
+	m.RLock()
+	defer m.RUnlock()
+	return len(m.ruleConfig.rules)
+}
+
+// GetGroupsCount returns the number of rule groups.
+func (m *RuleManager) GetGroupsCount() int {
+	m.RLock()
+	defer m.RUnlock()
+	return len(m.ruleConfig.groups)
+}
+
 // GetRulesByGroup returns sorted rules of a group.
 func (m *RuleManager) GetRulesByGroup(group string) []*Rule {
 	m.RLock()

diff --git a/pkg/schedule/placement/rule_manager_test.go b/pkg/schedule/placement/rule_manager_test.go
@@ -125,6 +125,7 @@ func TestAdjustRule(t *testing.T) {
 		IsWitness:        true,
 		LabelConstraints: []LabelConstraint{{Key: "engine", Op: "in", Values: []string{"tiflash"}}},
 	}, "tiflash"))
+
 }
 
 func TestLeaderCheck(t *testing.T) {
@@ -163,6 +164,8 @@ func TestSaveLoad(t *testing.T) {
 	re.Equal(rules[0].String(), m2.GetRule("pd", "default").String())
 	re.Equal(rules[1].String(), m2.GetRule("foo", "baz").String())
 	re.Equal(rules[2].String(), m2.GetRule("foo", "bar").String())
+	re.Equal(manager.GetRulesCount(), 3)
+	re.Equal(manager.GetGroupsCount(), 2)
 }
 
 func TestSetAfterGet(t *testing.T) {

diff --git a/pkg/schedule/schedulers/metrics.go b/pkg/schedule/schedulers/metrics.go
@@ -134,10 +134,19 @@ var (
 			Name:      "hot_pending_sum",
 			Help:      "Pending influence sum of store in hot region scheduler.",
 		}, []string{"store", "rw", "dim"})
+
+	ruleStatusGauge = prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Namespace: "pd",
+			Subsystem: "rule_manager",
+			Name:      "status",
+			Help:      "Status of the rule.",
+		}, []string{"type"})
 )
 
 func init() {
 	prometheus.MustRegister(schedulerStatusGauge)
+	prometheus.MustRegister(ruleStatusGauge)
 	prometheus.MustRegister(schedulerCounter)
 	prometheus.MustRegister(balanceWitnessCounter)
 	prometheus.MustRegister(hotSchedulerResultCounter)

diff --git a/pkg/schedule/schedulers/scheduler_controller.go b/pkg/schedule/schedulers/scheduler_controller.go
@@ -36,7 +36,11 @@ import (
 
 const maxScheduleRetries = 10
 
-var denySchedulersByLabelerCounter = labeler.LabelerEventCounter.WithLabelValues("schedulers", "deny")
+var (
+	denySchedulersByLabelerCounter = labeler.LabelerEventCounter.WithLabelValues("schedulers", "deny")
+	rulesCntStatusGauge            = ruleStatusGauge.WithLabelValues("rule_count")
+	groupsCntStatusGauge           = ruleStatusGauge.WithLabelValues("group_count")
+)
 
 // Controller is used to manage all schedulers.
 type Controller struct {
@@ -108,7 +112,6 @@ func (c *Controller) GetSchedulerHandlers() map[string]http.Handler {
 // CollectSchedulerMetrics collects metrics of all schedulers.
 func (c *Controller) CollectSchedulerMetrics() {
 	c.RLock()
-	defer c.RUnlock()
 	for _, s := range c.schedulers {
 		var allowScheduler float64
 		// If the scheduler is not allowed to schedule, it will disappear in Grafana panel.
@@ -118,6 +121,15 @@ func (c *Controller) CollectSchedulerMetrics() {
 		}
 		schedulerStatusGauge.WithLabelValues(s.Scheduler.GetName(), "allow").Set(allowScheduler)
 	}
+	c.RUnlock()
+	ruleMgr := c.cluster.GetRuleManager()
+	if ruleMgr == nil {
+		return
+	}
+	ruleCnt := ruleMgr.GetRulesCount()
+	groupCnt := ruleMgr.GetGroupsCount()
+	rulesCntStatusGauge.Set(float64(ruleCnt))
+	groupsCntStatusGauge.Set(float64(groupCnt))
 }
 
 func (c *Controller) isSchedulingHalted() bool {
@@ -127,6 +139,10 @@ func (c *Controller) isSchedulingHalted() bool {
 // ResetSchedulerMetrics resets metrics of all schedulers.
 func (c *Controller) ResetSchedulerMetrics() {
 	schedulerStatusGauge.Reset()
+	ruleStatusGauge.Reset()
+	// create in map again
+	rulesCntStatusGauge = ruleStatusGauge.WithLabelValues("rule_count")
+	groupsCntStatusGauge = ruleStatusGauge.WithLabelValues("group_count")
 }
 
 // AddSchedulerHandler adds the HTTP handler for a scheduler.