Merge pull request #2899 from Infisical/misc/add-custom-metrics-for-errors

misc: added metric for api errors

backend/src/server/plugins/error-handler.ts

@@ -1,8 +1,10 @@
 import { ForbiddenError, PureAbility } from "@casl/ability";
+import opentelemetry from "@opentelemetry/api";
 import fastifyPlugin from "fastify-plugin";
 import jwt from "jsonwebtoken";
 import { ZodError } from "zod";
 
+import { getConfig } from "@app/lib/config/env";
 import {
   BadRequestError,
   DatabaseError,
@@ -35,8 +37,30 @@ enum HttpStatusCodes {
 }
 
 export const fastifyErrHandler = fastifyPlugin(async (server: FastifyZodProvider) => {
+  const appCfg = getConfig();
+
+  const apiMeter = opentelemetry.metrics.getMeter("API");
+  const errorHistogram = apiMeter.createHistogram("API_errors", {
+    description: "API errors by type, status code, and name",
+    unit: "1"
+  });
+
   server.setErrorHandler((error, req, res) => {
     req.log.error(error);
+    if (appCfg.OTEL_TELEMETRY_COLLECTION_ENABLED) {
+      const { method } = req;
+      const route = req.routerPath;
+      const errorType =
+        error instanceof jwt.JsonWebTokenError ? "TokenError" : error.constructor.name || "UnknownError";
+
+      errorHistogram.record(1, {
+        route,
+        method,
+        type: errorType,
+        name: error.name
+      });
+    }
+
     if (error instanceof BadRequestError) {
       void res
         .status(HttpStatusCodes.BadRequest)

backend/src/services/secret/secret-queue.ts

@@ -1,4 +1,5 @@
 /* eslint-disable no-await-in-loop */
+import opentelemetry from "@opentelemetry/api";
 import { AxiosError } from "axios";
 
 import {
@@ -158,6 +159,12 @@ export const secretQueueFactory = ({
   projectUserMembershipRoleDAL,
   projectKeyDAL
 }: TSecretQueueFactoryDep) => {
+  const integrationMeter = opentelemetry.metrics.getMeter("Integrations");
+  const errorHistogram = integrationMeter.createHistogram("integration_secret_sync_errors", {
+    description: "Integration secret sync errors",
+    unit: "1"
+  });
+
   const removeSecretReminder = async (dto: TRemoveSecretReminderDTO) => {
     const appCfg = getConfig();
     await queueService.stopRepeatableJob(
@@ -933,6 +940,19 @@ export const secretQueueFactory = ({
               `Secret integration sync error [projectId=${job.data.projectId}] [environment=${environment}]  [secretPath=${job.data.secretPath}]`
             );
 
+            const appCfg = getConfig();
+            if (appCfg.OTEL_TELEMETRY_COLLECTION_ENABLED) {
+              errorHistogram.record(1, {
+                version: 1,
+                integration: integration.integration,
+                integrationId: integration.id,
+                type: err instanceof AxiosError ? "AxiosError" : err?.constructor?.name || "UnknownError",
+                status: err instanceof AxiosError ? err.response?.status : undefined,
+                name: err instanceof Error ? err.name : undefined,
+                projectId: integration.projectId
+              });
+            }
+
             const message =
               // eslint-disable-next-line no-nested-ternary
               (err instanceof AxiosError