[Inch-high]

Patch v0.8.3 — fix the /Docker "Check for updates" button finishing too quickly

Symptom

On a 40+ container host, "Check for updates" looked like it finished in ~2 seconds — out of sync with the real per-image registry walk, which takes much longer.

Root cause (front-end, deterministic)

The backend is correct: stock DockerTemplates::getAllInfo($reload=true) calls reloadUpdateStatus($image) per container (the real HTTPS digest fetch), confirmed against the live host. The bug was in the status-poll loop in md-docker-page.ts:

if (status.running) {
  this._checkPollHandle = window.setTimeout(tick, ...);
  return;                       // returns THROUGH the finally...
}
...
} finally {
  this._checkingUpdates = false;   // ...which fired on the FIRST poll, even while running
}

A return inside try still runs finally, so _checkingUpdates (a Lit @state()) was reset after the first 2s tick. The button reverted to idle while the worker kept walking in the background. A secondary issue: POLL_MAX_MS was a hard 60s total cap, shorter than a real 40+ container walk, so it could also cut the post-check snapshot refresh off mid-walk and read stale updated flags.

Fix

• Remove the misplaced finally; clear _checkingUpdates only on genuine terminal paths (completed / poll error / watchdog).
• Conclude only once the run actually finished — observed running:true, or a finishedAt newer than a pre-launch baseline (keeps small/fast hosts snappy). Extracted as a pure checkUpdatesCompleted() helper.
• Turn POLL_MAX_MS into an inactivity watchdog, reset on each running:true, so a long-but-alive walk isn't cut off; a worker that never establishes (or wedges) its lock still bails.

Backend left untouched.

Tests

• New checkUpdatesCompleted() + 5 unit cases in check-updates.test.ts.
• Full TS suite green (456 passed); Biome clean.

Verified live

Deployed to a 43-container Unraid host. Triggered "Update all" on the 2 flagged containers: cloudflareddns (running) updated + restarted; Nginx-Proxy-Manager (stopped, autostart off) updated + correctly left stopped. unraid-update-status.json flipped both false→true and webui-info was rewritten — badges clear as expected.

Release

Bumped package.json, package-lock.json, and the .plg <!ENTITY version> to 0.8.3, with a matching ###v0.8.3 changelog entry in the .plg <CHANGES> block. Tag v0.8.3 to publish after merge.

🤖 Generated with Claude Code