Repository navigation
v2.19.1-beta.1
Pre-release
Pre-release
Fixes for what 2.19.0 got wrong, found by a second audit of the tree
before this tag.
Fixed
- The Settings page saves again with a long stop timeout. 2.19.0 let
DOCKER_STOP_TIMEOUTandHEALTHCHECK_MAX_STARTINGgo up to 24 hours andDISK_WARN_PERCENTdown to 1, but the form kept the old limits. WithDOCKER_STOP_TIMEOUT=5400the browser refused to submit the page at all, and a save cut 5400 back to 3600. Form, save and environment now read one table. - A broken pin list stops manual updates too. Since 2.19.0 the check treats an unreadable pin list as "everything pinned", but the update buttons, "Update all" and
/updatedidn't ask. They refuse now and say why, and/checksays it checked nothing instead of "all up to date". - Telegram holds a 502 instead of losing it. A 502, 503 or 408 is retried and the message is held until Telegram answers again. A 504 may already have been delivered, so it's resent at most once: two copies at worst, where I measured 93 before.
- The Discord bot channel holds a 502 or 429 too. It only held a message when Discord didn't answer at all.
- Right after a reboot the disk check and the monitor start straight away. On a host up for a few seconds the first monitor pass came one full interval late.
- One slow cgroup probe no longer marks a Podman host as v1 until restart. Every recreate after that carried
--memory-swappinessagain (#50).
Internal
- A tag on a red tree no longer publishes an image: the publish workflow runs the whole suite first.
- Checks that could never go red now can, or say what they really hold. Maintenance mode, the update lock and the self-kill guard have procedures of their own.
- 230 procedures, 6080 checks. The mutation ratchet is at 289 guarded promises, nothing unguarded, no debt.