fix: metrics, add dynamic service name (#1112)

* fix: metrics, add dynamic service name

* fix: apply serviceName to tracing, prometheus prefix, and document in env samples

Agent-Logs-Url: https://github.com/supabase/storage/sessions/cacef517-4708-4b90-ba6c-6e85f0356a21

Co-authored-by: ferhatelmas <648018+ferhatelmas@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: ferhatelmas <648018+ferhatelmas@users.noreply.github.com>

Author: 3 people

.env.sample

@@ -156,6 +156,7 @@ WEBHOOK_API_KEY=
 # Monitoring
 #######################################
 LOG_LEVEL=info
+SERVICE_NAME=storage_api
 OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://localhost:4317
 OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=http://localhost:4317
 OTEL_METRICS_EXPORT_INTERVAL_MS=5000

.env.test.sample

@@ -37,3 +37,4 @@ ICEBERG_BUCKET_DETECTION_MODE="BUCKET"
 
 OTEL_METRICS_ENABLED=false
 PROMETHEUS_METRICS_ENABLED=false
+SERVICE_NAME=storage_api

src/config.ts

@@ -51,6 +51,7 @@ export interface JwksConfig {
 }
 
 type StorageConfigType = {
+  serviceName: string
   isProduction: boolean
   version: string
   numWorkers: number
@@ -264,6 +265,7 @@ export function getConfig(options?: { reload?: boolean }): StorageConfigType {
   const isMultitenant = getOptionalConfigFromEnv('MULTI_TENANT', 'IS_MULTITENANT') === 'true'
 
   config = {
+    serviceName: getOptionalConfigFromEnv('SERVICE_NAME') || 'storage_api',
     numWorkers: envNumber(getOptionalConfigFromEnv('WORKERS_NUM'), 1),
     isProduction: process.env.NODE_ENV === 'production',
     exposeDocs: getOptionalConfigFromEnv('EXPOSE_DOCS') !== 'false',

src/internal/monitoring/otel-metrics.ts

@@ -27,6 +27,7 @@ const {
   otelMetricsTemporality,
   prometheusMetricsEnabled,
   region,
+  serviceName,
 } = getConfig()
 
 let prometheusExporter: PrometheusExporter | undefined
@@ -79,7 +80,7 @@ Object.keys(exporterHeaders).forEach((key) => {
 })
 
 const resource = resourceFromAttributes({
-  [ATTR_SERVICE_NAME]: 'storage_api',
+  [ATTR_SERVICE_NAME]: serviceName,
   [ATTR_SERVICE_VERSION]: version,
   'metric.version': '1',
   region,
@@ -237,9 +238,9 @@ if (otelMetricsEnabled) {
 
   if (prometheusMetricsEnabled) {
     prometheusExporter = new PrometheusExporter({
-      prefix: 'storage_api',
+      prefix: serviceName,
       preventServerStart: true,
-      withResourceConstantLabels: /^(region|instance|metric\.version)$/,
+      withResourceConstantLabels: /^(region|instance|metric\.version|service\.name)$/,
     })
     readers.push(prometheusExporter)
   }

src/internal/monitoring/otel-tracing.ts

@@ -7,6 +7,7 @@ const {
   requestXForwardedHostRegExp,
   tenantId: defaultTenantId,
   region,
+  serviceName,
   storageS3InternalTracesEnabled,
 } = getConfig()
 
@@ -114,7 +115,7 @@ if (tracingEnabled && traceExporter && spanProcessors.length > 0) {
   // Configure the OpenTelemetry Node SDK
   tracingSdk = new NodeSDK({
     resource: resourceFromAttributes({
-      [ATTR_SERVICE_NAME]: 'storage',
+      [ATTR_SERVICE_NAME]: serviceName,
       [ATTR_SERVICE_VERSION]: version,
     }),
     spanProcessors,