Stratum v1.0.0-rc.53

Release candidate. Fixes agent supervision, which could be lost silently and stay lost.

Fixed

• The watchdog is restored and can now be updated. Two faults in the same area, found together on a live node.

The watchdog stopped alongside the agent but nothing started it again, so restarting a node's agent as two steps — stop, then start — left the agent running with no supervision until the next reboot. Nothing reported it: the agent was healthy, and the only trace was a log line. The watchdog is now tied to the agent in both directions and comes back whenever the agent does. This part arrives through the installer, so re-run it on a node to pick it up.

Separately, updating a node could never refresh the watchdog itself: replacing a running program in place is refused by the kernel, so every update reported an asset-refresh failure and quietly left the old watchdog in place. Files are now replaced atomically, which also removes the possibility of an interrupted update leaving a half-written file behind.

Notes

Fixing the first fault is what made the second one permanent rather than occasional — a watchdog that was frequently not running was also a watchdog that could usually be overwritten. Both are addressed here.

Feedback on this release candidate is welcome — report issues through your account.