A database blip while the portal was starting used to leave it running and serving nothing, which cost nearly three hours of downtime. Startup now ends the process when a plugin cannot come up, so the pod restarts instead of sitting dead, and liveness fails once the backend has never reached ready inside its deadline.
A short database absence no longer costs a restart. Each retry opens a fresh connection, so a database that returns within thirty seconds is waited out rather than failed.
What's Changed
Full Changelog: v0.1.3...v0.1.4
