fix(update): self-update rolls back healthy releases when the dashboard is off #1309

Closed
opened 2026-10-27 13:39:52 +00:00 by sam · 0 comments
Owner

With [web].listen unset or "off" archied serves no /healthz, so releaseupdate.CommandInstaller exports no ARCHIE_HEALTH_URL and the watchdog falls back to the dashboard default (127.0.0.1:8484). Nothing answers it, the probe times out and a healthy release is rolled back -- the same failure archie-core-1r4g fixed for the on-dashboard case, in the one configuration that has no HTTP surface to probe.

Reachable today only from Telegram, since a dashboard-initiated update implies the dashboard is up. An operator can work around it by exporting ARCHIE_HEALTH_URL, which the installer deliberately leaves intact.

Needs a decision, not just code: either verify liveness without HTTP (systemd unit state) or give the report an honest third outcome for 'not verifiable'. Both renderers word any non-passed check as a failure requiring intervention (internal/channels/telegram/update.go formatPendingReport, internal/webui/update_report.go summarizeUpdateReport), and that wording is deliberate -- an unverified success must not read as a confirmed one.

Close reason

Fixed in 7138b8d, not by choosing between the two options this bead posed. archied now serves its own /healthz on [health].listen (default 127.0.0.1:8485), always on, independent of the dashboard, and the watchdog probes that. The dashboard-off case has a health surface like every other case.


Imported from beads archie-core-exbz · GitHub #815 · type bug · priority P2 · created 2026-09-09 · closed 2026-09-09

With [web].listen unset or "off" archied serves no /healthz, so releaseupdate.CommandInstaller exports no ARCHIE_HEALTH_URL and the watchdog falls back to the dashboard default (127.0.0.1:8484). Nothing answers it, the probe times out and a healthy release is rolled back -- the same failure archie-core-1r4g fixed for the on-dashboard case, in the one configuration that has no HTTP surface to probe. Reachable today only from Telegram, since a dashboard-initiated update implies the dashboard is up. An operator can work around it by exporting ARCHIE_HEALTH_URL, which the installer deliberately leaves intact. Needs a decision, not just code: either verify liveness without HTTP (systemd unit state) or give the report an honest third outcome for 'not verifiable'. Both renderers word any non-passed check as a failure requiring intervention (internal/channels/telegram/update.go formatPendingReport, internal/webui/update_report.go summarizeUpdateReport), and that wording is deliberate -- an unverified success must not read as a confirmed one. ### Close reason Fixed in 7138b8d, not by choosing between the two options this bead posed. archied now serves its own /healthz on [health].listen (default 127.0.0.1:8485), always on, independent of the dashboard, and the watchdog probes that. The dashboard-off case has a health surface like every other case. --- _Imported from beads `archie-core-exbz` · GitHub #815 · type bug · priority P2 · created 2026-09-09 · closed 2026-09-09_
sam added the mediumbug labels 2026-10-27 13:39:52 +00:00
sam self-assigned this 2026-10-27 13:39:52 +00:00
sam closed this issue 2026-10-27 13:39:52 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: sam/archie#1309