feat(metrics): count api requests auth failures and 5xxs

test/api-server_test.ts

@@ -86,6 +86,89 @@ Deno.test('ApiServer serves health metrics and authenticated model listings', as
     assertEquals(authenticatedModels.status, 200);
     assertEquals(authenticatedBody.object, 'list');
     assertEquals(authenticatedBody.data[0].id, 'meta-llama/Llama-3.1-8B-Instruct');
+
+    const metricsAfterRequests = await fetch(`http://127.0.0.1:${port}/metrics`);
+    const metricsAfterRequestsBody = await metricsAfterRequests.text();
+    assertEquals(
+      metricsAfterRequestsBody.includes('modelgrid_api_requests_total{path="/v1/models"} 2'),
+      true,
+    );
+    assertEquals(
+      metricsAfterRequestsBody.includes('modelgrid_api_auth_failures_total{path="/v1/models"} 1'),
+      true,
+    );
+  } finally {
+    await server.stop();
+  }
+});
+
+Deno.test('ApiServer metrics expose 5xx counts for failing endpoints', async () => {
+  const port = 19100 + Math.floor(Math.random() * 1000);
+  let failModelListing = true;
+  const server = new ApiServer(
+    {
+      host: '127.0.0.1',
+      port,
+      apiKeys: ['valid-key'],
+      cors: false,
+      corsOrigins: [],
+    },
+    {
+      async getAllStatus() {
+        return new Map();
+      },
+      async getAllAvailableModels() {
+        if (failModelListing) {
+          failModelListing = false;
+          throw new Error('models unavailable');
+        }
+
+        return new Map();
+      },
+    } as never,
+    {
+      async getAllModels() {
+        return [];
+      },
+    } as never,
+    {} as never,
+    {
+      getStatus() {
+        return {
+          localNode: null,
+          nodes: [],
+          models: {},
+          desiredDeployments: [],
+        };
+      },
+    } as never,
+  );
+
+  (server as unknown as {
+    gpuDetector: { detectGpus: () => Promise<unknown[]> };
+  }).gpuDetector = {
+    async detectGpus() {
+      return [];
+    },
+  };
+
+  await server.start();
+
+  try {
+    const failedModels = await fetch(`http://127.0.0.1:${port}/v1/models`, {
+      headers: {
+        Authorization: 'Bearer valid-key',
+      },
+    });
+    assertEquals(failedModels.status, 500);
+    await failedModels.text();
+
+    const metricsResponse = await fetch(`http://127.0.0.1:${port}/metrics`);
+    const metricsBody = await metricsResponse.text();
+    assertEquals(
+      metricsBody.includes('modelgrid_api_server_errors_total{path="/v1/models"} 1'),
+      true,
+    );
   } finally {
     await server.stop();
   }