@rocicorp/zero 1.9.0-head-177d6c1f-20260713 → 1.9.0

package/out/zero-cache/src/services/view-syncer/view-syncer.js

@@ -25,6 +25,7 @@ import { ClientHandler, startPoke } from "./client-handler.js";
 import { tracer } from "./tracer.js";
 import { CVRStore, ClientNotFoundError } from "./cvr-store.js";
 import { CVRConfigDrivenUpdater, CVRQueryDrivenUpdater, nextEvictionTime } from "./cvr.js";
+import { E2EServingLagTracker } from "./e2e-serving-lag.js";
 import { handleInspect } from "./inspect-handler.js";
 import { QueryCoveringIndex } from "./query-covering.js";
 import { resolver } from "@rocicorp/resolver";
@@ -90,6 +91,7 @@ var ViewSyncerService = class {
 	#cvr;
 	#pipelinesSynced = false;
 	#servedVersion = null;
+	#e2eServingLagTracker = new E2EServingLagTracker();
 	#expiredQueriesTimer = 0;
 	#authMaintenanceTimer = 0;
 	#setTimeout;
@@ -102,6 +104,14 @@ var ViewSyncerService = class {
 		description: "Time from ViewSyncer query sync requiring hydration to output for a client group. Includes query transformation, query materialization, CVR flush, catchup, and pokeEnd.",
 		unit: "s"
 	});
+	#e2eServingLag = getOrCreateNativeHistogram("sync", "e2e_serving_lag", {
+		description: "End-to-end lag from upstream commit to ViewSyncer output. Spans the whole pipeline: the upstream transaction commit, replication to the replica, IVM advancement, CVR flush, and pokeEnd. Recorded once per served version, not sampled, so each observation is the completion latency of real replicated work.",
+		unit: "s"
+	});
+	#e2eServingLagClamps = getOrCreateCounter("sync", "e2e_serving_lag_clamps", {
+		description: "Observations of sync.e2e_serving_lag that came out negative and were clamped to zero. Non-zero means the upstream database clock is running ahead of this pod by more than the entire pipeline latency, so sync.e2e_serving_lag is biased low and reads healthier than reality. See replication.upstream_clock_skew for the magnitude.",
+		unit: "{observation}"
+	});
 	#transactionAdvanceTime = getOrCreateLatencyHistogram("sync", "advance-time", "Time to advance all queries for a given client group after applying a new transaction to the replica.");
 	#queryTransformations = getOrCreateCounter("sync", "query.transformations", "Number of query transformations performed");
 	#queryTransformationTime = getOrCreateLatencyHistogram("sync", "query.transformation-time", "Time to transform custom queries via API server.");
@@ -186,12 +196,14 @@ var ViewSyncerService = class {
 				this.#lc.debug?.(`draining view-syncer ${this.id} before running`);
 				this.stop();
 			}
-			for await (const { state } of this.#stateChanges) {
+			for await (const replicaState of this.#stateChanges) {
+				const { state } = replicaState;
 				if (this.#drainCoordinator.shouldDrain()) {
 					this.#lc.debug?.(`draining view-syncer ${this.id} (elective)`);
 					break;
 				}
 				assert(state === "version-ready", "state should be version-ready");
+				this.#e2eServingLagTracker.onVersionReady(replicaState);
 				await this.#runInLockWithCVR(async (lc, cvr) => {
 					const clientSchema = must(cvr.clientSchema, "cvr.clientSchema missing after initialization");
 					if (!this.#pipelines.initialized()) this.#pipelines.init(clientSchema);
@@ -258,6 +270,11 @@ var ViewSyncerService = class {
 	}
 	#markVersionServed(version) {
 		this.#servedVersion = version.stateVersion;
+		const observation = this.#e2eServingLagTracker.onVersionServed(version.stateVersion, Date.now());
+		if (observation !== null) {
+			this.#e2eServingLag.recordMs(observation.lagMs);
+			if (observation.clamped) this.#e2eServingLagClamps.add(1);
+		}
 	}
 	#keepAliveUntil = 0;
 	/**