[agentHits]

Stacked on #35. Retarget to AgentHits-Dev after #35 is merged.

Problem

After "Update Server" -> "Confirm", the Web Server Update dialog only showed a spinner with one sentence. The admin could not tell whether the image was still downloading, whether the download failed, or how long Dokploy had been down while it restarted.

What

Server

• New packages/server/src/services/web-server-update.ts: startServerUpdate(command) runs the update script in the background (as before, via spawnAsync("sh", ["-c", ...])) and tracks it in memory: phase (idle / pulling / updating / done / failed), startedAt, pulledAt, finishedAt, a short error, docker pull layer counts (total / downloaded / extracted) and the last 5 output lines. A second request while an update is running does not start another script.
• getAgentHitsUpdateCommand / getOfficialUpdateCommand print a marker right after docker pull ... || exit 1, so the server knows when the pull finished. Pull-first, || exit 1 and the env args are unchanged.
• settings.updateServer uses startServerUpdate; it still returns true.
• New admin query settings.getServerUpdateStatus, classified as "read".
• The status lives only in the process. It is lost on restart by design: a fresh idle status is also how the UI recognises the new process.

UI

• UpdateStatusPanel (components/dashboard/settings/web-server/update-status-panel.tsx) is a pure view with props { progress, now }. It shows four steps (Downloading the new image, Restarting Dokploy, Starting the new version, Done). Each step has a done / active / pending / failed state, elapsed times, layer progress, a live downtime counter and a failure box with the last output lines.
• The phase logic is pure in update-progress.ts. The dialog polls getServerUpdateStatus while the old server answers, switches to /api/health while Dokploy is down, then waits for the API to answer again (or 60 s of healthy /api/health) before the existing toast and reload.
• Failures: the pull or docker service update fails, the server does not go down within 20 minutes, or it does not come back within 10 minutes after it stopped. The dialog then shows what failed and a Close button.
• If the restart happens between two polls (for example, timers throttled in a background tab), an idle status after a seen job counts as done.

Checks

• apps/dokploy: bunx tsc --noEmit passes. packages/server: bunx tsc --noEmit passes.
• bunx vitest run --config __test__/vitest.config.ts __test__/settings/ __test__/security/: 33 files, 279 tests pass. New: web-server-update-status.test.ts (output parsing, phases, failures, no double start) and update-progress.test.ts (phase transitions, limits, step states). Updated: the router boundary tests, and the update command tests, which now also run the marker echo line through sh.
• bunx biome check --write on the changed paths is clean.

Not checked

• The panel has not been checked in a browser yet.
• No real update was run against a server, so the docker pull output format (layer lines) has not been checked with the Docker version on the host. If the format differs, only the layer counter stays empty; phases still work through the marker and the exit code.
• After the restart the UI does not check that the new process runs a newer version. A crash and restart of the old version during the pull would also read as "done". The page reload then shows the real state.

🤖 Generated with Claude Code