[antosubash]

Summary

Closes #880.

RedisInvalidationTransport.publish runs under a 1 s asyncio.timeout, while the request that made the write waits for its response. The client opened its command connection only on first use. So the first publish after boot, or after a Redis outage, also paid the TCP connect inside that second. Under load the connect was cancelled (seen in CI, #865) and the invalidation was lost; other workers kept the stale value until its TTL.

• After every successful SUBSCRIBE, at boot and after each reconnect, the listener now sends one PING on the command client (_warm_publish_connection). That opens the pooled connection the next publish reuses.
• The PING is bounded by SUBSCRIBE_TIMEOUT_SECONDS. A failure is logged at DEBUG and never restarts the listener. start() still never waits on Redis.
• Unchanged: the 1 s budget, the bus's log-and-continue handling of a failed publish, and settings.

Design: docs/superpowers/specs/2026-11-07-invalidation-warm-publish-design.md. Rejected alternatives:

• a separate connect budget (it doubles how long a request can be held);
• a background sender queue with retry (it changes the transport's failure contract and loses queued messages at shutdown).

Verification

• New test_background_tasks_invalidation_warmup.py:
  • fake client: one ping after subscribe, another after a reconnect, none before a subscribe succeeds;
  • fake client: a ping that fails or hangs leaves the listener delivering;
  • real Redis: CLIENT LIST shows the warmed connection, and a publish runs on that same connection without opening a new one. Connections are found by a per-test client_name, because CI workers share one Redis.
  • Without the production change, the three warm-up tests fail.
• make lint green.
• make test-py: 8,204 passed, 28 skipped, with the Redis tests running.
• make test-js: 900 passed.
• Reviews:
  • Codex: clean.
  • Sonnet correctness: clean.
  • Sonnet maintainability: fixed a docstring from test(background_tasks): fix two CI flakes in eager outbox and real-Redis invalidation tests #881 that this PR made false, plus spec wording. Its point about the test file having its own fake was rejected: the shared fake's file is at 293 of 300 lines, and there is no shared test helper module to move it to.

https://claude.ai/code/session_01M9neheZZEe3sVpDi2S3zT4

[cloudflare-workers-and-pages]

Deploying simple-module-python with    Cloudflare Pages

Latest commit:	d9e745f
Status:	✅  Deploy successful!
Preview URL:	https://d01f862f.simple-module-python.pages.dev
Branch Preview URL:	https://fix-issue-880-invalidation-c.simple-module-python.pages.dev

View logs

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
🔒 Security Review	✅ Completed 2026-11-07T14:08:31.828689Z	d9e745f	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.