cluster: support ng-monitoring config via server_configs and per-instance overrides

embed/examples/cluster/topology.example.yaml

@@ -63,6 +63,11 @@ monitored:
   # tiflash:
   # tiflash-learner:
   # kvcdc:
+  # ng_monitoring:
+  #   storage.type: "sqlite"
+  #   continuous_profiling.enable: true
+  #   continuous_profiling.profile_seconds: 5
+  #   continuous_profiling.interval_seconds: 15
 
 # # Server configs are used to specify the configuration of PD Servers.
 pd_servers:
@@ -363,6 +368,10 @@ monitoring_servers:
     # rule_dir: /home/tidb/prometheus_rule
     # scrape_interval: 15s
     # scrape_timeout: 10s
+    # # The following configs are used to overwrite the `server_configs.ng_monitoring` values.
+    # ng_monitoring_config:
+    #   storage.path: "/tidb-data/prometheus-8249/docdb"
+    #   continuous_profiling.data_retention_seconds: 259200
 # # Server configs are used to specify the configuration of Grafana Servers.  
 grafana_servers:
   # # The ip address of the Grafana Server.

pkg/cluster/spec/monitoring.go

@@ -63,7 +63,8 @@ type PrometheusSpec struct {
 	ScrapeInterval        string                 `yaml:"scrape_interval,omitempty" validate:"scrape_interval:editable"`
 	ScrapeTimeout         string                 `yaml:"scrape_timeout,omitempty" validate:"scrape_timeout:editable"`
 
-	AdditionalArgs []string `yaml:"additional_args,omitempty" validate:"additional_args:ignore"`
+	AdditionalArgs     []string       `yaml:"additional_args,omitempty" validate:"additional_args:ignore"`
+	NgMonitoringConfig map[string]any `yaml:"ng_monitoring_config,omitempty" validate:"ng_monitoring_config:ignore"`
 }
 
 // Remote prometheus remote config
@@ -486,24 +487,37 @@ func (i *MonitorInstance) InitConfig(
 	}
 
 	if spec.NgPort > 0 {
-		pds := []string{}
+		pdAddrs := []string{}
 		if servers, found := topoHasField("PDServers"); found {
 			for i := 0; i < servers.Len(); i++ {
 				pd := servers.Index(i).Interface().(*PDSpec)
-				pds = append(pds, fmt.Sprintf("\"%s\"", utils.JoinHostPort(pd.Host, pd.ClientPort)))
+				pdAddrs = append(pdAddrs, utils.JoinHostPort(pd.Host, pd.ClientPort))
 			}
 		}
-		ngcfg := &config.NgMonitoringConfig{
-			ClusterName:      clusterName,
-			Address:          utils.JoinHostPort(i.GetListenHost(), spec.NgPort),
-			AdvertiseAddress: utils.JoinHostPort(i.GetHost(), spec.NgPort),
-			PDAddrs:          strings.Join(pds, ","),
-			TLSEnabled:       enableTLS,
 
-			DeployDir: paths.Deploy,
-			DataDir:   paths.Data[0],
-			LogDir:    paths.Log,
+		// Build base ng-monitoring config as a map so user overrides via
+		// server_configs.ng_monitoring and per-instance ng_monitoring_config
+		// are merged on top (same pattern as PD/TiKV/TiDB).
+		baseConfig := map[string]any{
+			"address":           utils.JoinHostPort(i.GetListenHost(), spec.NgPort),
+			"advertise-address": utils.JoinHostPort(i.GetHost(), spec.NgPort),
+			"log.path":          paths.Log,
+			"log.level":         "INFO",
+			"pd.endpoints":      pdAddrs,
+			"storage.path":      paths.Data[0],
 		}
+		if enableTLS {
+			baseConfig["security.ca-path"] = fmt.Sprintf("%s/tls/ca.crt", paths.Deploy)
+			baseConfig["security.cert-path"] = fmt.Sprintf("%s/tls/prometheus.crt", paths.Deploy)
+			baseConfig["security.key-path"] = fmt.Sprintf("%s/tls/prometheus.pem", paths.Deploy)
+		}
+
+		// Gather global and per-instance ng-monitoring user config.
+		var globalNgConfig map[string]any
+		if s, ok := i.topo.(*Specification); ok {
+			globalNgConfig = s.ServerConfigs.NGMonitoring
+		}
+		userConfig := MergeConfig(globalNgConfig, spec.NgMonitoringConfig)
 
 		if servers, found := topoHasField("Monitors"); found {
 			for idx := 0; idx < servers.Len(); idx++ {
@@ -514,7 +528,11 @@ func (i *MonitorInstance) InitConfig(
 		}
 
 		fp = filepath.Join(paths.Cache, fmt.Sprintf("ngmonitoring_%s_%d.toml", i.GetHost(), i.GetPort()))
-		if err := ngcfg.ConfigToFile(fp); err != nil {
+		ngConf, err := Merge2Toml("ng_monitoring", baseConfig, userConfig)
+		if err != nil {
+			return err
+		}
+		if err := utils.WriteFile(fp, ngConf, 0755); err != nil {
 			return err
 		}
 		dst = filepath.Join(paths.Deploy, "conf", "ngmonitoring.toml")

pkg/cluster/spec/server_config_test.go

@@ -32,6 +32,66 @@ server_configs:
 	require.True(t, decimal)
 }
 
+func TestNGMonitoringServerConfig(t *testing.T) {
+	yamlData := []byte(`
+server_configs:
+  ng_monitoring:
+    storage.type: "sqlite"
+    log.level: "WARN"
+    continuous_profiling.enable: true
+    continuous_profiling.profile_seconds: 5
+    continuous_profiling.interval_seconds: 15
+
+monitoring_servers:
+  - host: 10.0.1.21
+    ng_port: 12020
+    ng_monitoring_config:
+      storage.path: "/custom/data/path"
+      continuous_profiling.data_retention_seconds: 259200
+`)
+
+	topo := new(Specification)
+	err := yaml.Unmarshal(yamlData, topo)
+	require.NoError(t, err)
+
+	// Verify global config parsed
+	require.Equal(t, "sqlite", topo.ServerConfigs.NGMonitoring["storage.type"])
+	require.Equal(t, "WARN", topo.ServerConfigs.NGMonitoring["log.level"])
+	require.Equal(t, true, topo.ServerConfigs.NGMonitoring["continuous_profiling.enable"])
+
+	// Verify per-instance config parsed
+	require.Len(t, topo.Monitors, 1)
+	require.Equal(t, "/custom/data/path", topo.Monitors[0].NgMonitoringConfig["storage.path"])
+
+	// Build base config (simulating what InitConfig does)
+	baseConfig := map[string]any{
+		"address":           "0.0.0.0:12020",
+		"advertise-address": "10.0.1.21:12020",
+		"log.path":          "/tidb-deploy/prometheus-9090/log",
+		"log.level":         "INFO",
+		"pd.endpoints":      []string{"10.0.1.10:2379", "10.0.1.11:2379"},
+		"storage.path":      "/tidb-data/prometheus-9090",
+	}
+
+	// Merge: base + global + per-instance (same as InitConfig logic)
+	userConfig := MergeConfig(topo.ServerConfigs.NGMonitoring, topo.Monitors[0].NgMonitoringConfig)
+	got, err := Merge2Toml("ng_monitoring", baseConfig, userConfig)
+	require.NoError(t, err)
+
+	tomlStr := string(got)
+
+	// User overrides should take precedence
+	require.Contains(t, tomlStr, `type = "sqlite"`)                 // from global server_configs
+	require.Contains(t, tomlStr, `path = "/custom/data/path"`)      // per-instance overrides base
+	require.Contains(t, tomlStr, `level = "WARN"`)                  // global overrides default
+	require.Contains(t, tomlStr, `enable = true`)                   // from global
+	require.Contains(t, tomlStr, `data_retention_seconds = 259200`) // from per-instance
+	require.Contains(t, tomlStr, `address = "0.0.0.0:12020"`)       // from base
+
+	// pd.endpoints must be a TOML array of quoted strings (matches old template behavior)
+	require.Contains(t, tomlStr, `endpoints = ["10.0.1.10:2379", "10.0.1.11:2379"]`)
+}
+
 func TestGetValueFromPath(t *testing.T) {
 	yamlData := []byte(`
 server_configs:

pkg/cluster/spec/spec.go

@@ -134,6 +134,7 @@ type (
 		CDC             map[string]any    `yaml:"cdc"`
 		TiKVCDC         map[string]any    `yaml:"kvcdc"`
 		Grafana         map[string]string `yaml:"grafana"`
+		NGMonitoring    map[string]any    `yaml:"ng_monitoring"`
 	}
 
 	// ComponentVersions represents the versions of components