feat(webapp): make the snapshot store completed-state TTL configurable

The snapshot store's completed-state retention is now configurable with
RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS (default 4 days).

Mono-RevId: 5d229fefab916dfc1e608023ffb36a512e72b6db

Author: d-cs

apps/webapp/app/env.server.ts

@@ -1459,6 +1459,13 @@ const EnvironmentSchema = z
     RUN_ENGINE_SNAPSHOT_STORE_REDIS_URL: z.string().optional(),
     // How long a fetched org snapshotStoreMode override is reused per process. 0 disables caching.
     RUN_ENGINE_SNAPSHOT_STORE_DIAL_CACHE_MS: z.coerce.number().int().min(0).default(5000),
+    // How long a finished run's snapshot state stays in the store. Redis-primary runs fail closed
+    // after it expires.
+    RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS: z.coerce
+      .number()
+      .int()
+      .positive()
+      .default(4 * 24 * 60 * 60 * 1_000),
     // Background sweep that resolves snapshot units left pending by an interrupted transition
     // (see snapshotRecoveryWorker.server.ts). Inert without RUN_ENGINE_SNAPSHOT_STORE_REDIS_URL.
     // Off unless set to "1": enable it on one background replica. Several replicas with it on are

apps/webapp/app/v3/runStore.server.ts

@@ -219,6 +219,7 @@ export const snapshotStoreRuntime = singleton("SnapshotStoreRuntime", () => {
   const metrics = createSnapshotStoreMetrics(metricsRegister);
   const connection = createSnapshotConnection(endpoint, {
     metrics: metrics.store,
+    completedTtlMs: env.RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS,
     onError: (error) => logger.error("Snapshot connection error", { error }),
   });
   return createSnapshotStoreRuntime({

apps/webapp/app/v3/utils/snapshotStoreConnection.server.ts

@@ -1,10 +1,10 @@
 import { createRedisClusterClient, type Cluster } from "@internal/redis";
 import { RedisSnapshotStore, type SnapshotStoreMetrics } from "@internal/run-store";
 
+const DEFAULT_COMPLETED_TTL_MS = 4 * 24 * 60 * 60 * 1_000;
 const COMMAND_TIMEOUT_MS = 1_000;
 const CONNECT_TIMEOUT_MS = 500;
 const POLICY_CHECK_INTERVAL_MS = 30_000;
-const COMPLETED_TTL_MS = 4 * 24 * 60 * 60 * 1_000;
 
 /** Infrastructure only. Rollout modes belong to the existing polled flags, not this URL. */
 export function snapshotClusterOptions(endpoint: string) {
@@ -45,7 +45,12 @@ export function snapshotClusterOptions(endpoint: string) {
  */
 export function createSnapshotConnection(
   endpoint: string,
-  options: { metrics?: SnapshotStoreMetrics; onError?: (error: Error) => void } = {}
+  options: {
+    metrics?: SnapshotStoreMetrics;
+    onError?: (error: Error) => void;
+    /** How long a finished run's snapshot state stays in the store. */
+    completedTtlMs?: number;
+  } = {}
 ) {
   const clientOptions = snapshotClusterOptions(endpoint);
   let client: Cluster | undefined;
@@ -114,7 +119,7 @@ export function createSnapshotConnection(
     getStore: () =>
       (store ??= new RedisSnapshotStore({
         client: getClient(),
-        completedTtlMs: COMPLETED_TTL_MS,
+        completedTtlMs: options.completedTtlMs ?? DEFAULT_COMPLETED_TTL_MS,
         beforeCommand,
         metrics: options.metrics,
       })),

apps/webapp/test/env.server.test.ts

@@ -115,4 +115,28 @@ describe("webapp environment secrets", () => {
     expect(env.GITHUB_APP_CLIENT_ID).toBe("client-id");
     expect(env.GITHUB_APP_CLIENT_SECRET).toBe("client-secret");
   });
+
+  it("defaults the snapshot store completed-state TTL to 4 days", async () => {
+    process.env = { ...requiredEnv };
+
+    const { env } = await import("../app/env.server");
+
+    expect(env.RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS).toBe(4 * 24 * 60 * 60 * 1_000);
+  });
+
+  it("reads the snapshot store completed-state TTL from the environment", async () => {
+    process.env = { ...requiredEnv, RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS: "3600000" };
+
+    const { env } = await import("../app/env.server");
+
+    expect(env.RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS).toBe(3_600_000);
+  });
+
+  it("rejects a non-positive snapshot store completed-state TTL", async () => {
+    process.env = { ...requiredEnv, RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS: "0" };
+
+    await expect(import("../app/env.server")).rejects.toThrow(
+      "RUN_ENGINE_SNAPSHOT_STORE_COMPLETED_TTL_MS"
+    );
+  });
 });