chore: sync grafana dashboard and EMQX dashboard

fix: #88 Signed-off-by: Rory Z <[email protected]>
emqx · May 6, 2024 · 9f8ab82 · 9f8ab82
1 parent e364f7d
commit 9f8ab82
Show file tree

Hide file tree

Showing 9 changed files with 55 additions and 73 deletions.
diff --git a/collector/collector_rule_engine.go b/collector/collector_rule_engine.go
@@ -33,9 +33,9 @@ const (
 
 	ruleTopicHitCount      = "topic_hit_count"
 	ruleExecPassCount      = "exec_pass_count"
-	ruleExecFailureCount   = "exec_failure_count"
-	ruleNoResultCount      = "exec_no_result_count"
-	ruleExecExceptionCount = "exec_exception_count"
+	ruleExecFailureCount   = "exec_failure_count"   // failure count = no result count + exec exception count, it's didn't show in EMQX dashboard
+	ruleNoResultCount      = "exec_no_result_count" // show in EMQX dashboard
+	ruleExecExceptionCount = "exec_exception_count" // show in EMQX dashboard
 	ruleExecRate           = "exec_rate"
 	ruleExecLast5mRate     = "exec_last5m_rate"
 	ruleExecMaxRate        = "exec_max_rate"

diff --git a/grafana-dashboard/grafanalib/Makefile b/grafana-dashboard/grafanalib/Makefile
@@ -1,20 +1,33 @@
 # Makefile to run Python scripts with customizable arguments
-
-# Default values for arguments
-# EDITION_ARG can be ee or ce
-EDITION_ARG ?= ee
-# VERSION_ARG can be 4 or 5
-VERSION_ARG ?= 5
 PROJECT_DIR := $(shell dirname $(abspath $(lastword $(MAKEFILE_LIST))))
-JSON_DIR = $(PROJECT_DIR)/json
 
 # Phony target to handle all tasks
-.PHONY: all clean
-ifeq ($(VERSION_ARG),4)
-all: $(JSON_DIR) overview.json client-events.json messages.json rule-engine-rate.json rule-engine-count.json
-else
-all: $(JSON_DIR) overview.json authentication.json authorization.json client-events.json messages.json rule-engine-rate.json rule-engine-count.json
-endif
+.PHONY: all
+all: emqx_4 emqx_ee_4 emqx_5 emqx_ee_5
+
+.PHONY: emqx_4
+emqx_4: EDITION_ARG=ce
+emqx_4: VERSION_ARG=4
+emqx_4: JSON_DIR = $(PROJECT_DIR)/../template/emqx-4
+emqx_4: $(JSON_DIR) overview.json client-events.json messages.json rule-engine-rate.json rule-engine-count.json
+
+.PHONY: emqx_ee_4
+emqx_ee_4: EDITION_ARG=ee
+emqx_ee_4: VERSION_ARG=4
+emqx_ee_4: JSON_DIR = $(PROJECT_DIR)/../template/emqx-enterprise-4
+emqx_ee_4: $(JSON_DIR) overview.json client-events.json messages.json rule-engine-rate.json rule-engine-count.json
+
+.PHONY: emqx_5
+emqx_5: EDITION_ARG=ce
+emqx_5: VERSION_ARG=5
+emqx_5: JSON_DIR = $(PROJECT_DIR)/../template/emqx-5
+emqx_5: $(JSON_DIR) overview.json authentication.json authorization.json client-events.json messages.json rule-engine-rate.json rule-engine-count.json
+
+.PHONY: emqx_ee_5
+emqx_ee_5: EDITION_ARG=ee
+emqx_ee_5: VERSION_ARG=5
+emqx_ee_5: JSON_DIR = $(PROJECT_DIR)/../template/emqx-enterprise-5
+emqx_ee_5: $(JSON_DIR) overview.json authentication.json authorization.json client-events.json messages.json rule-engine-rate.json rule-engine-count.json
 
 # Create json directory if it doesn't exist
 $(JSON_DIR):
@@ -41,8 +54,3 @@ rule-engine-rate.json:
 
 rule-engine-count.json:
 	./rule-engine-count.py > $(JSON_DIR)/rule-engine-count.json
-
-# Clean up by removing the json directory
-clean:
-	rm -rf $(JSON_DIR)
-
diff --git a/grafana-dashboard/grafanalib/README.md b/grafana-dashboard/grafanalib/README.md
@@ -24,32 +24,6 @@ pipenv shell
 
 ## Generate Dashboard JSON
 
-- Generate EMQX 5.x Enterprise Dashboard
-
 ```shell
 make
 ```
-
-- Generate EMQX 5.x Community Edition Dashboard
-
-```shell
-make EDITION_ARG=ce VERSION_ARG=5
-```
-
-- Generate EMQX 4.x Enterprise Dashboard
-
-```shell
-make EDITION_ARG=ee VERSION_ARG=4
-```
-
-- Generate EMQX 4.x Community Edition Dashboard
-
-```shell
-make EDITION_ARG=ce VERSION_ARG=4
-```
-
-## Remove Dashboard JSON
-
-```shell
-make clean
-```
diff --git a/grafana-dashboard/grafanalib/metrics.py b/grafana-dashboard/grafanalib/metrics.py
@@ -264,12 +264,12 @@
         "format": "timeseries"
     },
     "rule_engine_exec_failure": {
-        "title": "Rule Engine Exec Failure",
+        "title": "Rule Engine Exec Exception",
         "subchart_links": (subcharts["rule-engine-count"], ["subchart_rule_engine_exec_failure"]),
         "targets": [
             {
                 "legendFormat": "{{rule}}",
-                "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))"
+                "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))"
             }
         ],
         "format": "timeseries"
@@ -1270,7 +1270,7 @@
             },
             {
                 "legendFormat": "Exec Exception last 15m",
-                "expr": "sum by(node, rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
+                "expr": "sum by(node, rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
                 "thresholds": {
                     "mode": "absolute",
                     "steps": [
@@ -1341,11 +1341,11 @@
         "format": "timeseries"
     },
     "subchart_rule_engine_exec_failure": {
-        "title": "Rule Engine Exec Failure",
+        "title": "Rule Engine Exec Exception",
         "targets": [
             {
                 "legendFormat": "{{ node }}:{{ rule }}",
-                "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
+                "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
 
             }
         ],

diff --git a/grafana-dashboard/grafanalib/submetrics.py b/grafana-dashboard/grafanalib/submetrics.py
@@ -317,7 +317,7 @@
                     },
                     {
                         "legendFormat": "Exec Exception last 15m",
-                        "expr": "sum by(node, rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
+                        "expr": "sum by(node, rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
                         "thresholds": {
                             "mode": "absolute",
                             "steps": [
@@ -388,11 +388,11 @@
                 "format": "timeseries"
             },
             "subchart_rule_engine_exec_failure": {
-                "title": "Rule Engine Exec Failure",
+                "title": "Rule Engine Exec Exception",
                 "targets": [
                     {
                         "legendFormat": "{{ node }}:{{ rule }}",
-                        "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
+                        "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
 
                     }
                 ],

diff --git a/grafana-dashboard/template/emqx-4/overview.json b/grafana-dashboard/template/emqx-4/overview.json
@@ -1486,7 +1486,7 @@
             {
               "targetBlank": true,
               "title": "Show Rule Engine Exec Failure Node Detail",
-              "url": "/d/rule-engine-count/Rule Engine Exec Failure?orgId=1&refresh=10s&var-node=All&viewPanel=3"
+              "url": "/d/rule-engine-count/Rule Engine Exec Exception?orgId=1&refresh=10s&var-node=All&viewPanel=3"
             }
           ],
           "maxDataPoints": 100,
@@ -1508,23 +1508,23 @@
           "targets": [
             {
               "datasource": null,
-              "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
+              "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
               "format": "timeseries",
               "hide": false,
               "instant": false,
               "interval": "",
               "intervalFactor": 1,
               "legendFormat": "{{rule}}",
               "metric": "",
-              "query": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
+              "query": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
               "refId": "{{rule}}",
               "step": 10,
               "target": ""
             }
           ],
           "timeFrom": null,
           "timeShift": null,
-          "title": "Rule Engine Exec Failure",
+          "title": "Rule Engine Exec Exception",
           "transformations": [],
           "transparent": false,
           "type": "timeseries"

diff --git a/grafana-dashboard/template/emqx-4/rule-engine-count.json b/grafana-dashboard/template/emqx-4/rule-engine-count.json
@@ -287,15 +287,15 @@
             },
             {
               "datasource": null,
-              "expr": "sum by(node, rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
+              "expr": "sum by(node, rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
               "format": "table",
               "hide": false,
               "instant": true,
               "interval": "",
               "intervalFactor": 1,
               "legendFormat": "Exec Exception last 15m",
               "metric": "",
-              "query": "sum by(node, rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
+              "query": "sum by(node, rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
               "refId": "exec_exception_last_15m",
               "step": 10,
               "target": ""
@@ -583,23 +583,23 @@
           "targets": [
             {
               "datasource": null,
-              "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
+              "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
               "format": "timeseries",
               "hide": false,
               "instant": false,
               "interval": "",
               "intervalFactor": 1,
               "legendFormat": "{{ node }}:{{ rule }}",
               "metric": "",
-              "query": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
+              "query": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
               "refId": "{{_node_}}:{{_rule_}}",
               "step": 10,
               "target": ""
             }
           ],
           "timeFrom": null,
           "timeShift": null,
-          "title": "Rule Engine Exec Failure",
+          "title": "Rule Engine Exec Exception",
           "transformations": [],
           "transparent": false,
           "type": "timeseries"

diff --git a/grafana-dashboard/template/emqx-enterprise-5/overview.json b/grafana-dashboard/template/emqx-enterprise-5/overview.json
@@ -1857,7 +1857,7 @@
             {
               "targetBlank": true,
               "title": "Show Rule Engine Exec Failure Node Detail",
-              "url": "/d/rule-engine-count/Rule Engine Exec Failure?orgId=1&refresh=10s&var-node=All&viewPanel=3"
+              "url": "/d/rule-engine-count/Rule Engine Exec Exception?orgId=1&refresh=10s&var-node=All&viewPanel=3"
             }
           ],
           "maxDataPoints": 100,
@@ -1879,23 +1879,23 @@
           "targets": [
             {
               "datasource": null,
-              "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
+              "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
               "format": "timeseries",
               "hide": false,
               "instant": false,
               "interval": "",
               "intervalFactor": 1,
               "legendFormat": "{{rule}}",
               "metric": "",
-              "query": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
+              "query": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\".*\"}[$__rate_interval]))",
               "refId": "{{rule}}",
               "step": 10,
               "target": ""
             }
           ],
           "timeFrom": null,
           "timeShift": null,
-          "title": "Rule Engine Exec Failure",
+          "title": "Rule Engine Exec Exception",
           "transformations": [],
           "transparent": false,
           "type": "timeseries"

diff --git a/grafana-dashboard/template/emqx-enterprise-5/rule-engine-count.json b/grafana-dashboard/template/emqx-enterprise-5/rule-engine-count.json
@@ -287,15 +287,15 @@
             },
             {
               "datasource": null,
-              "expr": "sum by(node, rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
+              "expr": "sum by(node, rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
               "format": "table",
               "hide": false,
               "instant": true,
               "interval": "",
               "intervalFactor": 1,
               "legendFormat": "Exec Exception last 15m",
               "metric": "",
-              "query": "sum by(node, rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
+              "query": "sum by(node, rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[15m]))",
               "refId": "exec_exception_last_15m",
               "step": 10,
               "target": ""
@@ -583,23 +583,23 @@
           "targets": [
             {
               "datasource": null,
-              "expr": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
+              "expr": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
               "format": "timeseries",
               "hide": false,
               "instant": false,
               "interval": "",
               "intervalFactor": 1,
               "legendFormat": "{{ node }}:{{ rule }}",
               "metric": "",
-              "query": "sum by(rule) (irate(emqx_rule_exec_failure_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
+              "query": "sum by(rule) (irate(emqx_rule_exec_exception_count{cluster=\"$cluster\", node=~\"$node\", rule=~\"$rule\"}[$__rate_interval]))",
               "refId": "{{_node_}}:{{_rule_}}",
               "step": 10,
               "target": ""
             }
           ],
           "timeFrom": null,
           "timeShift": null,
-          "title": "Rule Engine Exec Failure",
+          "title": "Rule Engine Exec Exception",
           "transformations": [],
           "transparent": false,
           "type": "timeseries"