Track metrics ports with prometheus (#111)

* Track metrics ports with prometheus

* Add an external scrape endpoint

Author: ferranbt

internal/components.go

@@ -137,6 +137,7 @@ func (o *OpGeth) Run(service *service, ctx *ExContext) {
 		WithImage("us-docker.pkg.dev/oplabs-tools-artifacts/images/op-geth").
 		WithTag("v1.101503.2-rc.5").
 		WithEntrypoint("/bin/sh").
+		WithLabel("metrics_path", "/debug/metrics/prometheus").
 		WithArgs(
 			"-c",
 			"geth init --datadir {{.Dir}}/data_opgeth --state.scheme hash {{.Dir}}/l2-genesis.json && "+
@@ -262,6 +263,7 @@ func (r *RethEL) Run(svc *service, ctx *ExContext) {
 			"--authrpc.port", `{{Port "authrpc" 8551}}`,
 			"--authrpc.addr", "0.0.0.0",
 			"--authrpc.jwtsecret", "{{.Dir}}/jwtsecret",
+			"--metrics", `0.0.0.0:{{Port "metrics" 9090}}`,
 			// For reth version 1.2.0 the "legacy" engine was removed, so we now require these arguments:
 			"--engine.persistence-threshold", "0", "--engine.memory-block-buffer-target", "0",
 			logLevelToRethVerbosity(ctx.LogLevel),
@@ -513,6 +515,7 @@ func (o *OpReth) Run(service *service, ctx *ExContext) {
 			"--datadir", "{{.Dir}}/data_op_reth",
 			"--disable-discovery",
 			"--color", "never",
+			"--metrics", `0.0.0.0:{{Port "metrics" 9090}}`,
 			"--port", `{{Port "rpc" 30303}}`)
 }
 

internal/local_runner.go

@@ -755,6 +755,68 @@ func (d *LocalRunner) trackContainerStatusAndLogs() {
 	}
 }
 
+func CreatePrometheusServices(manifest *Manifest, out *output) error {
+	// Read all the components to be deployed and find all the ports with name 'metrics'
+	// to create the prometheus scrapper config
+	var scrapeConfigs []map[string]interface{}
+
+	// global scrape config
+	scrapeConfigs = append(scrapeConfigs, map[string]interface{}{
+		"job_name":     "external",
+		"metrics_path": "/metrics",
+		"static_configs": []map[string]interface{}{
+			{
+				"targets": []string{"host.docker.internal:5555"},
+			},
+		},
+	})
+
+	for _, c := range manifest.services {
+		for _, port := range c.ports {
+			if port.Name == "metrics" {
+				metricsPath := "/metrics"
+				if overrideMetricsPath, ok := c.labels["metrics_path"]; ok {
+					metricsPath = overrideMetricsPath
+				}
+
+				scrapeConfig := map[string]interface{}{
+					"job_name":     c.Name,
+					"metrics_path": metricsPath,
+					"static_configs": []map[string]interface{}{
+						{
+							"targets": []string{fmt.Sprintf("%s:%d", c.Name, port.Port)},
+						},
+					},
+				}
+				scrapeConfigs = append(scrapeConfigs, scrapeConfig)
+			}
+		}
+	}
+
+	promConfig := map[string]interface{}{
+		"global": map[string]interface{}{
+			"scrape_interval":     "1s",
+			"evaluation_interval": "1s",
+		},
+		"scrape_configs": scrapeConfigs,
+	}
+
+	if err := out.WriteFile("prometheus.yaml", promConfig); err != nil {
+		return fmt.Errorf("failed to write prometheus.yml: %w", err)
+	}
+
+	// add to the manifest the prometheus service
+	// This is a bit of a hack.
+	srv := manifest.NewService("prometheus").
+		WithImage("prom/prometheus").
+		WithTag("latest").
+		WithArgs("--config.file", "{{.Dir}}/prometheus.yaml").
+		WithPort("metrics", 9090, "tcp")
+	manifest.services = append(manifest.services, srv)
+
+	return nil
+}
+
 func (d *LocalRunner) Run() error {
 	go d.trackContainerStatusAndLogs()
 

main.go

@@ -25,6 +25,7 @@ var interactive bool
 var timeout time.Duration
 var logLevelFlag string
 var bindExternal bool
+var withPrometheus bool
 
 var rootCmd = &cobra.Command{
 	Use:   "playground",
@@ -169,6 +170,7 @@ func main() {
 		recipeCmd.Flags().DurationVar(&timeout, "timeout", 0, "") // Used for CI
 		recipeCmd.Flags().StringVar(&logLevelFlag, "log-level", "info", "log level")
 		recipeCmd.Flags().BoolVar(&bindExternal, "bind-external", false, "bind host ports to external interface")
+		recipeCmd.Flags().BoolVar(&withPrometheus, "with-prometheus", false, "whether to gather the Prometheus metrics")
 
 		cookCmd.AddCommand(recipeCmd)
 	}
@@ -225,6 +227,12 @@ func runIt(recipe internal.Recipe) error {
 		return err
 	}
 
+	if withPrometheus {
+		if err := internal.CreatePrometheusServices(svcManager, artifacts.Out); err != nil {
+			return fmt.Errorf("failed to create prometheus services: %w", err)
+		}
+	}
+
 	if dryRun {
 		return nil
 	}