chore: add metrics of exception and no result

emqx · Sep 20, 2023 · 443f5af · 443f5af
1 parent 952c0ee
commit 443f5af
Show file tree

Hide file tree

Showing 5 changed files with 32 additions and 19 deletions.
diff --git a/client/client_4.x.go b/client/client_4.x.go
@@ -153,22 +153,22 @@ func (n *cluster4x) getRuleEngineMetrics() (metrics []collector.RuleEngine, err
 	resp := struct {
 		Data []struct {
 			Metrics []struct {
-				Node        string
+				Node        string `json:"node"`
 				SpeedMax    float64 `json:"speed_max"`
 				SpeedLast5m float64 `json:"speed_last5m"`
 				Speed       float64 `json:"speed"`
-				Matched     int64
-				Passed      int64
-				//NoResult    int64
-				//Exception   int64
-				Failed int64
+				Matched     int64 `json:"matched"`
+				Passed      int64 `json:"passed"`
+				NoResult    int64 `json:"no_result"`
+				Exception   int64 `json:"exception"`
+				Failed int64 `json:"failed"`
 			}
 			Actions []struct {
 				Metrics []struct {
-					Node    string
-					Taken   int64
-					Success int64
-					Failed  int64
+					Node    string `json:"node"`
+					Taken   int64 `json:"taken"`
+					Success int64 `json:"success"`
+					Failed  int64 `json:"failed"`
 				}
 			}
 			ID      string `json:"id"`
@@ -208,10 +208,11 @@ func (n *cluster4x) getRuleEngineMetrics() (metrics []collector.RuleEngine, err
 				NodeName: cutNodeName(m.Node),
 				RuleID:   rule.ID,
 				//ResStatus:           unknown,
-				TopicHitCount:    m.Matched,
-				ExecPassCount:    m.Passed,
-				ExecFailureCount: m.Failed,
-				//NoResultCount:      m.NoResult,
+				TopicHitCount:      m.Matched,
+				ExecPassCount:      m.Passed,
+				ExecFailureCount:   m.Failed,
+				NoResultCount:      m.NoResult,
+				ExecExceptionCount: m.Exception,
 				ExecRate:           m.Speed,
 				ExecLast5mRate:     m.SpeedLast5m,
 				ExecMaxRate:        m.SpeedMax,

diff --git a/client/client_5.x.go b/client/client_5.x.go
@@ -146,7 +146,7 @@ func (n *cluster5x) getRuleEngineMetrics() (metrics []collector.RuleEngine, err
 					Matched    int64
 					Passed     int64
 					Failed     int64
-					//Exception     int64 `json:"failed.exception"`
+					Exception     int64 `json:"failed.exception"`
 					NoResult      int64 `json:"failed.no_result"`
 					ActionTotal   int64 `json:"actions.total"`
 					ActionSuccess int64 `json:"actions.success"`
@@ -166,6 +166,7 @@ func (n *cluster5x) getRuleEngineMetrics() (metrics []collector.RuleEngine, err
 				TopicHitCount:      node.Metrics.Matched,
 				ExecPassCount:      node.Metrics.Passed,
 				ExecFailureCount:   node.Metrics.Failed,
+				ExecExceptionCount: node.Metrics.Exception,
 				NoResultCount:      node.Metrics.NoResult,
 				ExecRate:           node.Metrics.Rate,
 				ExecLast5mRate:     node.Metrics.RateLast5m,

diff --git a/collector/client.go b/collector/client.go
@@ -33,6 +33,7 @@ type RuleEngine struct {
 	TopicHitCount      int64
 	ExecPassCount      int64
 	ExecFailureCount   int64
+	ExecExceptionCount int64
 	NoResultCount      int64
 	ExecRate           float64
 	ExecLast5mRate     float64

diff --git a/collector/rule_engine.go b/collector/rule_engine.go
@@ -34,6 +34,7 @@ const (
 	ruleExecPassCount    = "exec_pass_count"
 	ruleExecFailureCount = "exec_failure_count"
 	ruleNoResultCount    = "exec_no_result_count"
+	ruleExecExceptionCount   = "exec_exception_count"
 	ruleExecRate         = "exec_rate"
 	ruleExecLast5mRate   = "exec_last5m_rate"
 	ruleExecMaxRate      = "exec_max_rate"
@@ -111,6 +112,11 @@ func NewRuleEngineCollector(client Cluster, logger log.Logger) (Collector, error
 			help:   "The failure count of rule exec",
 			labels: []string{"node", "rule"},
 		},
+		{
+			name:   ruleExecExceptionCount,
+			help:   "The exception count of rule exec",
+			labels: []string{"node", "rule"},
+		},
 		{
 			name:   ruleNoResultCount,
 			help:   "The no result count of rule exec",
@@ -199,6 +205,10 @@ func (c *ruleEngineCollector) Update(ch chan<- prometheus.Metric) error {
 			c.desc[ruleExecFailureCount],
 			prometheus.CounterValue, float64(metric.ExecFailureCount), metric.NodeName, metric.RuleID,
 		)
+		ch <- prometheus.MustNewConstMetric(
+			c.desc[ruleExecExceptionCount],
+			prometheus.CounterValue, float64(metric.ExecExceptionCount), metric.NodeName, metric.RuleID,
+		)
 		ch <- prometheus.MustNewConstMetric(
 			c.desc[ruleNoResultCount],
 			prometheus.CounterValue, float64(metric.NoResultCount), metric.NodeName, metric.RuleID,

diff --git a/config/grafana-template/grafanalib-emqx/metrics.py b/config/grafana-template/grafanalib-emqx/metrics.py
@@ -601,8 +601,8 @@
                 "expr": "sum by(rule) (emqx_rule_exec_pass_count{cluster=\"$cluster\", node=~\".*\"})",
             },
             {
-                "legendFormat": "Exec Failed last 15m",
-                "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\".*\"}[15m]))",
+                "legendFormat": "Exec Exception last 15m",
+                "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\".*\"}[15m]))",
                 "thresholds": {
                     "mode": "absolute",
                     "steps": [
@@ -618,8 +618,8 @@
                 }
             },
             {
-                "legendFormat": "Exec No Result",
-                "expr": "sum by(rule) (emqx_rule_exec_no_result_count{cluster=\"$cluster\", node=~\".*\"})",
+                "legendFormat": "Exec No Result 15m",
+                "expr": "sum by(rule) (irate(emqx_rule_exec_no_result_count{cluster=\"$cluster\", node=~\".*\"}[15m]))",
                 "thresholds": {
                     "mode": "absolute",
                     "steps": [