Merge pull request #1924 from Infisical/misc/metrics-observability

feat: added setup for production observability (metrics via OTEL)

backend/src/lib/config/env.ts

@@ -157,6 +157,15 @@ const envSchema = z
     INFISICAL_CLOUD: zodStrBool.default("false"),
     MAINTENANCE_MODE: zodStrBool.default("false"),
     CAPTCHA_SECRET: zpStr(z.string().optional()),
+
+    // TELEMETRY
+    OTEL_TELEMETRY_COLLECTION_ENABLED: zodStrBool.default("false"),
+    OTEL_EXPORT_OTLP_ENDPOINT: zpStr(z.string().optional()),
+    OTEL_OTLP_PUSH_INTERVAL: z.coerce.number().default(30000),
+    OTEL_COLLECTOR_BASIC_AUTH_USERNAME: zpStr(z.string().optional()),
+    OTEL_COLLECTOR_BASIC_AUTH_PASSWORD: zpStr(z.string().optional()),
+    OTEL_EXPORT_TYPE: z.enum(["prometheus", "otlp"]).optional(),
+
     PLAIN_API_KEY: zpStr(z.string().optional()),
     PLAIN_WISH_LABEL_IDS: zpStr(z.string().optional()),
     DISABLE_AUDIT_LOG_GENERATION: zodStrBool.default("false"),
@@ -203,11 +212,11 @@ let envCfg: Readonly<z.infer<typeof envSchema>>;
 
 export const getConfig = () => envCfg;
 // cannot import singleton logger directly as it needs config to load various transport
-export const initEnvConfig = (logger: Logger) => {
+export const initEnvConfig = (logger?: Logger) => {
   const parsedEnv = envSchema.safeParse(process.env);
   if (!parsedEnv.success) {
-    logger.error("Invalid environment variables. Check the error below");
-    logger.error(parsedEnv.error.issues);
+    (logger ?? console).error("Invalid environment variables. Check the error below");
+    (logger ?? console).error(parsedEnv.error.issues);
     process.exit(-1);
   }
 

backend/src/lib/telemetry/instrumentation.ts

@@ -0,0 +1,91 @@
+import opentelemetry, { diag, DiagConsoleLogger, DiagLogLevel } from "@opentelemetry/api";
+import { getNodeAutoInstrumentations } from "@opentelemetry/auto-instrumentations-node";
+import { OTLPMetricExporter } from "@opentelemetry/exporter-metrics-otlp-proto";
+import { PrometheusExporter } from "@opentelemetry/exporter-prometheus";
+import { registerInstrumentations } from "@opentelemetry/instrumentation";
+import { Resource } from "@opentelemetry/resources";
+import { AggregationTemporality, MeterProvider, PeriodicExportingMetricReader } from "@opentelemetry/sdk-metrics";
+import { ATTR_SERVICE_NAME, ATTR_SERVICE_VERSION } from "@opentelemetry/semantic-conventions";
+import dotenv from "dotenv";
+
+import { initEnvConfig } from "../config/env";
+
+dotenv.config();
+
+const initTelemetryInstrumentation = ({
+  exportType,
+  otlpURL,
+  otlpUser,
+  otlpPassword,
+  otlpPushInterval
+}: {
+  exportType?: string;
+  otlpURL?: string;
+  otlpUser?: string;
+  otlpPassword?: string;
+  otlpPushInterval?: number;
+}) => {
+  diag.setLogger(new DiagConsoleLogger(), DiagLogLevel.DEBUG);
+
+  const resource = Resource.default().merge(
+    new Resource({
+      [ATTR_SERVICE_NAME]: "infisical-core",
+      [ATTR_SERVICE_VERSION]: "0.1.0"
+    })
+  );
+
+  const metricReaders = [];
+  switch (exportType) {
+    case "prometheus": {
+      const promExporter = new PrometheusExporter();
+      metricReaders.push(promExporter);
+      break;
+    }
+    case "otlp": {
+      const otlpExporter = new OTLPMetricExporter({
+        url: `${otlpURL}/v1/metrics`,
+        headers: {
+          Authorization: `Basic ${btoa(`${otlpUser}:${otlpPassword}`)}`
+        },
+        temporalityPreference: AggregationTemporality.DELTA
+      });
+      metricReaders.push(
+        new PeriodicExportingMetricReader({
+          exporter: otlpExporter,
+          exportIntervalMillis: otlpPushInterval
+        })
+      );
+      break;
+    }
+    default:
+      throw new Error("Invalid OTEL export type");
+  }
+
+  const meterProvider = new MeterProvider({
+    resource,
+    readers: metricReaders
+  });
+
+  opentelemetry.metrics.setGlobalMeterProvider(meterProvider);
+
+  registerInstrumentations({
+    instrumentations: [getNodeAutoInstrumentations()]
+  });
+};
+
+const setupTelemetry = () => {
+  const appCfg = initEnvConfig();
+
+  if (appCfg.OTEL_TELEMETRY_COLLECTION_ENABLED) {
+    console.log("Initializing telemetry instrumentation");
+    initTelemetryInstrumentation({
+      otlpURL: appCfg.OTEL_EXPORT_OTLP_ENDPOINT,
+      otlpUser: appCfg.OTEL_COLLECTOR_BASIC_AUTH_USERNAME,
+      otlpPassword: appCfg.OTEL_COLLECTOR_BASIC_AUTH_PASSWORD,
+      otlpPushInterval: appCfg.OTEL_OTLP_PUSH_INTERVAL,
+      exportType: appCfg.OTEL_EXPORT_TYPE
+    });
+  }
+};
+
+void setupTelemetry();