[StrandedTurtle]

Summary

Fixes

• "Update all" stopped part-way, and progress vanished after navigating away. Updates were driven by the dashboard page:

  • The batch was a loop in the browser, so leaving the page or locking the phone killed it.
  • Containers whose card wasn't on screen (e.g. hidden by search) were silently skipped.

  Now every update, revert and tag switch is a server-side job (server/src/jobs.js):

  • Jobs run one at a time within a stack, with at most 3 running at once.
  • "Update all" is one request (POST /api/update-all); containers it can't queue are reported as skipped, never silently dropped.
  • GET /api/jobs plus one app-wide event stream let any page or device show progress. The header shows "Updating · 2/5 done", and cards re-attach to their live log, which is kept for 15 minutes.
  • In "Updates only", a finished card stays visible with its result instead of disappearing.

• Release build stall: tag and main builds now use separate build-cache scopes (the shared scope stalled :latest on v1.8.0), and the publish job times out at 25 minutes instead of 6 hours.

Features

• Update logs saved in History ("Show update log" on each entry).
• Container logs on every card (the container's own output, e.g. why it crash-loops).
• Auto-revert if unhealthy: per-container opt-in. A broken update is reverted in the same job, with history and a notification.
• "Update all" summary notification: one message when the batch finishes (toggle in Settings).
• Download size next to each update: only the layers you don't already have, read from the registry. Platform selection now prefers the server's own CPU (arm64 on a Pi, previously always amd64).
• Faster dashboard:
  • container listing runs 8 at a time with shared image lookups (was one-by-one)
  • client reloads are coalesced, so a burst of finishing updates triggers one reload, not one per container

Testing

• 198 server tests pass, with new tests for the queue, log tail, history logs, download size, platform pick, log decoding, mapLimit and batch messages. The client builds.
• Live on Docker 29 with a local registry:
  • "Update all" over 4 containers with no browser connected: the 2 services in one stack ran one after another, the others in parallel; a missing container was reported as skipped; the summary webhook arrived.
  • Auto-revert: a crash-looping version was automatically reverted to the previous one, with history and a notification.
  • Exact download sizes (~3.0 MB new layer, not the whole image) and the container-logs API.
• Real browser (Playwright): tapped Update all, went straight to Settings (header showed progress), came back mid-batch (cards re-attached to their live logs), everything finished and the summary banner showed. No console errors.

---

Generated by Claude Code