fix(hub): wait longer for a new project's coordinator

apps/hub/src/projects/coordinator-startup.test.ts

@@ -0,0 +1,55 @@
+import { afterEach, expect, setSystemTime, test } from "bun:test";
+import type { DB } from "@intx/db";
+import { workflowRun } from "@intx/db/schema";
+import type { SidecarAllocationStatus } from "@intx/types";
+import { awaitCoordinatorReady } from "./coordinator-startup.ts";
+
+const run = {
+  id: "run",
+  anchorRunId: "run",
+  address: "run@tenant",
+  status: "running",
+  expiresAt: null,
+  cancellationRequestedAt: null,
+};
+
+// Each poll reads the next allocation status, then moves the clock on by `stepMs`.
+function nativeStates(statuses: SidecarAllocationStatus[], stepMs: number) {
+  let polls = 0;
+  const db = {
+    select: () => ({
+      from: (table: unknown) => ({
+        where: async () => {
+          if (table === workflowRun) return [run];
+          const status = statuses[Math.min(polls, statuses.length - 1)];
+          polls += 1;
+          setSystemTime(new Date(Date.now() + stepMs));
+          return [{ status, nextAttemptAt: null }];
+        },
+      }),
+    }),
+  };
+  return db as unknown as DB["db"];
+}
+
+afterEach(() => {
+  setSystemTime();
+});
+
+test("a coordinator ready after more than 120 s still starts", async () => {
+  await awaitCoordinatorReady(
+    nativeStates(["provisioning", "allocated"], 121_000),
+    "tenant",
+    "run",
+  );
+});
+
+test("a start still pending past 270 s fails", async () => {
+  await expect(
+    awaitCoordinatorReady(
+      nativeStates(["provisioning"], 271_000),
+      "tenant",
+      "run",
+    ),
+  ).rejects.toMatchObject({ code: "coordinator_start_failed" });
+});

apps/hub/src/projects/coordinator-startup.ts

@@ -22,7 +22,10 @@ import { ProjectError } from "./errors.ts";
 import type { Actor } from "../lib/actor.ts";
 
 const START_FAILED = "coordinator_start_failed";
-const READY_WAIT_MS = 120_000;
+// Longer than the native sidecar ensure and connection stages (120 s each) normally take, and
+// shorter than the web's 5-minute request ceiling, so a failed start reaches the person who
+// created the project instead of archiving it after they stopped waiting.
+const READY_WAIT_MS = 270_000;
 const READY_POLL_MS = 250;
 
 type BoundRun = { runId: string; ended: boolean };
@@ -188,7 +191,7 @@ async function attach(
   return true;
 }
 
-async function awaitCoordinatorReady(
+export async function awaitCoordinatorReady(
   db: DB["db"],
   tenantId: string,
   runId: string,