[kasnder]

What

When there was no work, the analyser called /ping and /queue about every 30 seconds. Railway only sleeps a service after 5–10 minutes without outbound traffic, and responses to inbound requests count as outbound traffic, so the website never slept.

• analyser/processQueue.sh: after an empty or failed /queue, wait IDLE_SLEEP_MIN (60 s), doubling each time up to IDLE_SLEEP_MAX (1800 s). The wait goes back to the minimum once an app is assigned. Every curl call now retries transient failures (connection refused, 5xx including Railway's cold-start 502) for up to 60 s. Retrying an upload whose first attempt did land just gets a 409 for the spent claim.
• The analyser's last check-in was only held in memory (lastPing), so a woken process showed "Analyser offline" until the next ping. It is now also written to a one-row analyser_heartbeat table (migration 015) and read back once per process. A failed read is retried after a minute, and a failed write is logged without failing /ping.
• Documented in analyser/.env.example and raspberry-pi-analyser.md.

Trade-off: an app requested while the queue is empty can wait up to IDLE_SLEEP_MAX (30 min by default) before the analyser picks it up. The Pi's checkout needs updating for the script change to take effect.

Testing

• npm test: 165/165 pass, including the new test/analyserHeartbeat.test.js (the stored heartbeat is read once per process and /ping writes it).
• Ran migration 015 (applied twice) and the heartbeat upsert/read against a local Postgres 16. The CHECK (id) constraint rejects a second row.
• Ran the curl retry settings against a local server returning 502 twice and then 200 to a POST: it succeeded on the third attempt. A refused connection was retried.
• Ran idle_backoff with small limits: 1, 2, 4, 5, 5 s.

CI will fail at the npm audit step until #37 merges; the advisories are unrelated to this change.

https://claude.ai/code/session_01Fi3H98SycMK13rrr8ySSU1

---

Generated by Claude Code

[kasnder]

CI is red at the npm audit step only. npm test passes all 165 tests, including test/analyserHeartbeat.test.js.

The failure is not from this PR: it doesn't touch dependencies. The advisories (proxy-addr, brace-expansion, and braces via the dev dependency nodemon) were published after main last passed. #37 fixes them and its CI is green. To keep this PR separate, I haven't copied the lockfile change in here. Once #37 is merged I'll merge main into this branch so CI runs again. A re-run before then would fail the same way.

---

Generated by Claude Code

[kasnder]

Closing in favour of #39. That PR drops the polling entirely: the analyser waits for an ntfy wake-up or the next due time, and it keeps the stored heartbeat and curl retries from this PR.

---

Generated by Claude Code