Files
pulse/internal
rcourtman 8b6c93d3e5 Keep restarts from re-recording every resource as discovered
The first registry diff after Pulse starts compares against an empty
generation, so every restart or update recorded each resource the
fleet already had as "resource discovered". On a 1,776-resource mock
fleet one restart added 1,776 such rows, which buries a night's real
changes when an agent asks list_recent_changes what happened overnight,
for example after an auto-update.

Before its first diff the monitor adapter now reads each resource's
last recorded state. A known resource's first appearance in the run
records nothing when that state still holds, a transition from it
marked "resource state changed while Pulse was stopped" when it
differs, and a normal discovery when it was last recorded as removed.
The while-stopped transition keeps the boundary availability reports
need for a resource that recovered or failed during the gap. The same
restart now records 35 rows, each a real difference.

Removals during the gap are still not recorded, and a resource whose
history aged out or whose canonical ID changed across the restart is
discovered once more.
2026-10-29 19:30:04 +00:00
..