Give Cloudflare test polls a 10 s limit

[cardmagic]

Summary

delivers the complete failure envelope for 'retry' in test/cloudflare/effect-payloads.test.ts failed in CI run 36884020000, attempt 1 (PR #59, head 295b908). The poll failed with Matcher did not succeed in time after the received list stayed []. Attempt 2 of the same run passed, and the push run on the same commit passed.

Cause: the Cloudflare tests wait for durable work with expect.poll. Of the 24 polls in test/cloudflare, 23 used the Vitest default limit of 1 s. The retry case needs five alarm-driven delivery attempts and a callback turn. In the failed attempt, every Cloudflare test file ran 5 to 12 times slower than usual, so that chain took more than 1 s:

File	Usual CI time	Failed attempt
realtime	~0.75 s	5.3 s
runtime	~0.65 s	5.8 s
recovery	~1.6 s	8.6 s
contract	~0.24 s	4.1 s
effect-payloads	~1.4 s	7.9 s

The engine is not at fault. The same file also ran slowly before this change, for example 4.3 s on main (8d8140e) and 10.0 s on feat/result-lookup (2c9dd13).

Fix: set expect.poll.timeout to 10 s for the whole Cloudflare suite in vitest.cloudflare.config.ts. A poll still returns as soon as its condition holds, so a normal run does not take longer. The 15 s test timeout and the explicit 15 s poll in recovery.test.ts do not change.

Undici audit fix

The first CI run of this PR failed at pnpm audit --audit-level=high, not at a test. Two high Undici advisories (WebSocket subprotocol denial of service, TLS certificate validation bypass) now affect undici >=7.0.0 <7.29.1, which reaches this repository only through Miniflare. main last ran CI on September 23, before the advisories, so main fails this audit too.

48aec85 overrides [email protected] with 7.29.1 in pnpm-workspace.yaml and the lockfile. This is the same change as e128a95 on feat/portable-observability (#59). #59 also records it in the changelog, so this PR does not; a second entry would conflict when #59 merges.

Compatibility

Test configuration and a development-tooling dependency override only. No runtime, API, or published package effect. Not applicable to the Ruby gem, which has no Cloudflare backend.

Validation

• Reproduction: a temporary 250 ms delay in each 'retry' delivery attempt in test/cloudflare/worker.ts copies a slow runner. With the default limit, the test fails with Matcher did not succeed in time after 1052 ms. With the 10 s limit, it passes in 1365 ms. Both results are on main and on PR Align observability and runtime contracts #59's head.
• A lost callback still fails clearly: with a 12 s delay per attempt, the test fails with Matcher did not succeed in time after 10037 ms, before the 15 s test timeout.
• pnpm run test:cloudflare: 58 passed on main; 59 passed three times on PR Align observability and runtime contracts #59's head.
• pnpm run check and prettier --check pass.
• Audit: without the override, pnpm audit --audit-level=high exits 1 with the same two high Undici advisories as CI. With the override, it reports no high vulnerabilities.
• With the override: pnpm install --frozen-lockfile, format:check, check, build, test:cloudflare (59 passed), check:cloudflare, and pnpm test (528 passed, 32 skipped) pass.

[greptile-apps]

Confidence Score: 5/5

[Low risk] Adjusts test timeout settings for Cloudflare tests.

The PR appears safe to merge.

Summary

Extends the default expect.poll timeout for Cloudflare tests from 1 s to 10 s while retaining the 15 s test timeout and explicit per-poll overrides.

_{Reviews (1) · Last reviewed commit: "test: give Cloudflare polls a 10 s limit"}