Bare-metal Kura nodes (Dedibox, OVH, Scaleway Elastic Metal) now auto-recover from silent kernel lockups instead of staying NotReady until a human IPMI-resets the box. Newly (re)provisioned nodes arm a systemd hardware watchdog and kernel panic sysctls, so a detected lockup reboots the box in roughly 30 seconds and preserves the warm cache on its local PV. As a backstop for existing nodes, a per-fleet MachineHealthCheck reinstalls any box that stays NotReady for 10+ minutes, with a guardrail of unhealthyRange: [0-1] so a single dead box is fixed without mass-reinstalling a whole region on a cluster-wide blip. The result: production deploys no longer fail their canary stage because one frozen bare-metal node held a DaemonSet slot indefinitely.
Hive
Bare-metal Kura nodes now auto-recover from silent kernel lockups
Published
Jul 02, 2026 · 09:52 UTC
Repository
tuist/tuist