ci: Retry host shard on Vite dynamic-import flakes too

[habdelra]

Summary

The host-shard CI retry only matched ChunkLoadError, the webpack-era flake signature. After the host build moved to Vite, the same transient class of failure surfaces as Failed to fetch dynamically imported module (and occasionally NetworkError when attempting to fetch resource), which never tripped the retry and instead failed the shard outright.

A recent example: Host Tests (4, 20) failed on monaco.contribution-DyRU9U2K.js failing to fetch dynamically. The grep for ChunkLoadError returned zero matches, the retry skipped, the shard turned red.

Changes

• Broaden the retry trigger from ChunkLoadError to a regex covering all three transient chunk-fetch signatures.
• Print the first 5 matched lines so the cause shows up directly in the GitHub Actions run log instead of buried in the per-shard testem log artifact.
• Tee the retry attempt's output and emit one of two banners after the retry:
  • ::notice:: "recovered on retry" when the shard turns green — confirms the original failure was a flake.
  • ::error:: "failed twice" when both runs fail — flags the failure as deterministic so investigators don't waste time treating it as flake. If the retry hit the same transient pattern again, that's also called out separately as a possible serving-side issue worth investigating.

The realm-shim layer already enumerates these same signatures for its in-process retry (packages/runtime-common/package-shim-handler.ts). This brings the CI shard-retry layer into alignment.

Test plan

• YAML parses (python3 -c "import yaml; yaml.safe_load(open('.github/workflows/ci-host.yaml'))")
• Inline shell parses (bash -n on extracted step)
• CI green
• On the next host shard that hits Failed to fetch dynamically imported module, the retry triggers and the new diagnostic lines appear in the run log

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3953df21cc

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

This PR updates the host-test CI shard retry logic in .github/workflows/ci-host.yaml so that it also retries on Vite-era dynamic import / chunk-fetch flake signatures (not just the older webpack ChunkLoadError), and surfaces useful diagnostics directly in the Actions log.

Changes:

• Expand the retry trigger to match ChunkLoadError, Failed to fetch dynamically imported module, and NetworkError when attempting to fetch resource.
• Print the first 5 matching lines in a grouped log section to make the flake cause visible in the job output.
• On retry success, promote the retry log to /tmp/test-output.log so downstream parsing uses the successful run; on retry failure, emit explicit “failed twice” messaging and (if applicable) show retry-run matches.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Host Test Results

    1 files      1 suites   1h 46m 35s ⏱️
2 634 tests 2 619 ✅ 15 💤 0 ❌
2 653 runs  2 638 ✅ 15 💤 0 ❌

Results for commit 0fa718f.

[backspace]

I wish we could add resilience that would let us retry loading vs having to restart the whole job (which breaks Percy), but getting back to what we had with Webpack is baseline at least

[habdelra]

Thanks for the approve. Agree on the longer-term shape — an in-process retry around the dynamic import('monaco-editor') (and any other lazy chunks) would mirror what package-shim-handler.ts's withResolveRetry already does for the realm shim layer (#4531) and would dodge the Percy / wall-clock cost of restarting the whole shard. Happy to file a follow-up branch that lifts that helper out and applies it here; this PR just restores the webpack-era baseline so the bleeding stops.

[habdelra]

lol, sorry my claude PR comment monitoring is very eager