WC-2695 Add better visibility for router-worker (#6941)

This commit instruments the router worker with analytics such
as request time, colo metadata, error, etc, in order for us to
have better visibility into the router worker.

These changes were tested using gradual rollouts with a 0% version
and a Cloudflare-Workers-Version-Overrides header.

.changeset/poor-shoes-tickle.md

@@ -0,0 +1,5 @@
+---
+"@cloudflare/workers-shared": minor
+---
+
+feat: Add observability to router-worker

packages/workers-shared/router-worker/src/analytics.ts

@@ -0,0 +1,71 @@
+import type { Environment, ReadyAnalytics } from "./types";
+
+// This will allow us to make breaking changes to the analytic schema
+const VERSION = 1;
+
+export enum DISPATCH_TYPE {
+	ASSETS = "asset",
+	WORKER = "worker",
+}
+
+// When adding new columns please update the schema
+type Data = {
+	// -- Doubles --
+	// double1 - The time it takes for the whole request to complete in milliseconds
+	requestTime?: number;
+	// double2 - Colo ID
+	coloId?: number;
+	// double3 - Metal ID
+	metalId?: number;
+	// double4 - Colo tier (e.g. tier 1, tier 2, tier 3)
+	coloTier?: number;
+
+	// -- Blobs --
+	// blob1 - Hostname of the request
+	hostname?: string;
+	// blob2 - Dispatch type - what kind of thing did we dispatch
+	dispatchtype?: DISPATCH_TYPE;
+	// blob3 - Error message
+	error?: string;
+	// blob4 - The current version UUID of router-server
+	version?: string;
+	// blob5 - Region of the colo (e.g. WEUR)
+	coloRegion?: string;
+};
+
+export class Analytics {
+	private data: Data = {};
+
+	setData(newData: Partial<Data>) {
+		this.data = { ...this.data, ...newData };
+	}
+
+	getData(key: keyof Data) {
+		return this.data[key];
+	}
+
+	write(env: Environment, readyAnalytics?: ReadyAnalytics, hostname?: string) {
+		if (!readyAnalytics) {
+			return;
+		}
+
+		readyAnalytics.logEvent({
+			version: VERSION,
+			accountId: 0, // TODO: need to plumb through
+			indexId: hostname,
+			doubles: [
+				this.data.requestTime ?? -1, // double1
+				this.data.coloId ?? -1, // double2
+				this.data.metalId ?? -1, // double3
+				this.data.coloTier ?? -1, // double4
+			],
+			blobs: [
+				this.data.hostname?.substring(0, 256), // blob1 - trim to 256 bytes
+				this.data.dispatchtype, // blob2
+				this.data.error?.substring(0, 256), // blob3 - trim to 256 bytes
+				this.data.version, // blob4
+				this.data.coloRegion, // blob5
+			],
+		});
+	}
+}

packages/workers-shared/router-worker/src/index.ts

@@ -1,13 +1,26 @@
 import { setupSentry } from "../../utils/sentry";
+import { Analytics, DISPATCH_TYPE } from "./analytics";
+import { PerformanceTimer } from "./performance";
 import type AssetWorker from "../../asset-worker/src/index";
 import type { RoutingConfig } from "../../utils/types";
+import type {
+	ColoMetadata,
+	Environment,
+	ReadyAnalytics,
+	UnsafePerformanceTimer,
+} from "./types";
 
 interface Env {
 	ASSET_WORKER: Service<AssetWorker>;
 	USER_WORKER: Fetcher;
 	CONFIG: RoutingConfig;
 
 	SENTRY_DSN: string;
+	ENVIRONMENT: Environment;
+	ANALYTICS: ReadyAnalytics;
+	COLO_METADATA: ColoMetadata;
+	UNSAFE_PERFORMANCE: UnsafePerformanceTimer;
+	VERSION_METADATA: WorkerVersionMetadata;
 
 	SENTRY_ACCESS_CLIENT_ID: string;
 	SENTRY_ACCESS_CLIENT_SECRET: string;
@@ -16,7 +29,9 @@ interface Env {
 export default {
 	async fetch(request: Request, env: Env, ctx: ExecutionContext) {
 		let sentry: ReturnType<typeof setupSentry> | undefined;
-		const maybeSecondRequest = request.clone();
+		const analytics = new Analytics();
+		const performance = new PerformanceTimer(env.UNSAFE_PERFORMANCE);
+		const startTimeMs = performance.now();
 
 		try {
 			sentry = setupSentry(
@@ -27,21 +42,50 @@ export default {
 				env.SENTRY_ACCESS_CLIENT_SECRET
 			);
 
+			const url = new URL(request.url);
+			if (sentry) {
+				sentry.setUser({ username: url.hostname });
+				sentry.setTag("colo", env.COLO_METADATA.coloId);
+				sentry.setTag("metal", env.COLO_METADATA.metalId);
+			}
+
+			if (env.COLO_METADATA && env.VERSION_METADATA) {
+				analytics.setData({
+					coloId: env.COLO_METADATA.coloId,
+					metalId: env.COLO_METADATA.metalId,
+					coloTier: env.COLO_METADATA.coloTier,
+					coloRegion: env.COLO_METADATA.coloRegion,
+					hostname: url.hostname,
+					version: env.VERSION_METADATA.id,
+				});
+			}
+
+			const maybeSecondRequest = request.clone();
 			if (env.CONFIG.has_user_worker) {
 				if (await env.ASSET_WORKER.unstable_canFetch(request)) {
+					analytics.setData({ dispatchtype: DISPATCH_TYPE.ASSETS });
 					return await env.ASSET_WORKER.fetch(maybeSecondRequest);
 				} else {
+					analytics.setData({ dispatchtype: DISPATCH_TYPE.WORKER });
 					return env.USER_WORKER.fetch(maybeSecondRequest);
 				}
 			}
 
+			analytics.setData({ dispatchtype: DISPATCH_TYPE.ASSETS });
 			return await env.ASSET_WORKER.fetch(request);
 		} catch (err) {
+			if (err instanceof Error) {
+				analytics.setData({ error: err.message });
+			}
+
 			// Log to Sentry if we can
 			if (sentry) {
 				sentry.captureException(err);
 			}
 			throw err;
+		} finally {
+			analytics.setData({ requestTime: performance.now() - startTimeMs });
+			analytics.write(env.ENVIRONMENT, env.ANALYTICS);
 		}
 	},
 };

packages/workers-shared/router-worker/src/performance.ts

@@ -0,0 +1,16 @@
+import type { UnsafePerformanceTimer } from "./types";
+
+export class PerformanceTimer {
+	private performanceTimer;
+
+	constructor(performanceTimer?: UnsafePerformanceTimer) {
+		this.performanceTimer = performanceTimer;
+	}
+
+	now() {
+		if (this.performanceTimer) {
+			return this.performanceTimer.timeOrigin + this.performanceTimer.now();
+		}
+		return Date.now();
+	}
+}

packages/workers-shared/router-worker/src/types.ts

@@ -0,0 +1,25 @@
+export type Environment = "production" | "staging";
+
+export interface ReadyAnalytics {
+	logEvent: (e: ReadyAnalyticsEvent) => void;
+}
+
+export interface ColoMetadata {
+	metalId: number;
+	coloId: number;
+	coloRegion: string;
+	coloTier: number;
+}
+
+export interface UnsafePerformanceTimer {
+	readonly timeOrigin: number;
+	now: () => number;
+}
+
+export interface ReadyAnalyticsEvent {
+	accountId?: number;
+	indexId?: string;
+	version?: number;
+	doubles?: (number | undefined)[];
+	blobs?: (string | undefined)[];
+}

packages/workers-shared/router-worker/wrangler.toml

@@ -13,6 +13,9 @@ workers_dev = false
 main = "src/index.ts"
 compatibility_date = "2024-07-31"
 
+[version_metadata]
+binding = "VERSION_METADATA"
+
 [[unsafe.bindings]]
 name = "CONFIG"
 type = "param"
@@ -29,4 +32,8 @@ type = "origin"
 
 [unsafe.metadata.build_options]
 stable_id = "cloudflare/cf_router_worker"
-networks = ["cf","jdc"]
+networks = ["cf","jdc"]
+
+[[unsafe.bindings]]
+name = "workers-router-worker"
+type = "internal_capability_grants"