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

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

backend/src/server/app.ts

@@ -22,6 +22,7 @@ import { TSmtpService } from "@app/services/smtp/smtp-service";
 
 import { globalRateLimiterCfg } from "./config/rateLimiter";
 import { addErrorsToResponseSchemas } from "./plugins/add-errors-to-response-schemas";
+import { apiMetrics } from "./plugins/api-metrics";
 import { fastifyErrHandler } from "./plugins/error-handler";
 import { registerExternalNextjs } from "./plugins/external-nextjs";
 import { serializerCompiler, validatorCompiler, ZodTypeProvider } from "./plugins/fastify-zod";
@@ -86,6 +87,10 @@ export const main = async ({ db, hsmModule, auditLogDb, smtp, logger, queue, key
     // pull ip based on various proxy headers
     await server.register(fastifyIp);
 
+    if (appCfg.OTEL_TELEMETRY_COLLECTION_ENABLED) {
+      await server.register(apiMetrics);
+    }
+
     await server.register(fastifySwagger);
     await server.register(fastifyFormBody);
     await server.register(fastifyErrHandler);

backend/src/server/plugins/api-metrics.ts

@@ -0,0 +1,21 @@
+import opentelemetry from "@opentelemetry/api";
+import fp from "fastify-plugin";
+
+export const apiMetrics = fp(async (fastify) => {
+  const apiMeter = opentelemetry.metrics.getMeter("API");
+  const latencyHistogram = apiMeter.createHistogram("API_latency", {
+    unit: "ms"
+  });
+
+  fastify.addHook("onResponse", async (request, reply) => {
+    const { method } = request;
+    const route = request.routerPath;
+    const { statusCode } = reply;
+
+    latencyHistogram.record(reply.elapsedTime, {
+      route,
+      method,
+      statusCode
+    });
+  });
+});