The metadata snapshot deleted every SCUM sqlite query job while persisting, so
the recurring database poll lost its job records as soon as anything wrote the
snapshot and left its stdout/stderr streams behind. Those polls are the only
producer for the platform scum_user and scum_vehicle tables, and the service
already retires older terminal polls with their streams, so the snapshot no
longer drops them.
Startup now removes job log streams whose job no longer exists, keeping the
autonomous lifecycle streams the Run posts without a platform job. That clears
the streams left by the removed poll producers instead of carrying them in
every snapshot.
Each dispatched poll is a durable job with two job log streams, so an
unbounded poll history would grow the platform store by roughly two
thousand jobs a day per server. A new poll now retires older terminal jobs
of the same template together with their job log streams, leaving at most
two rows per template while nothing is in flight.
Run pushes typed SCUM facts through POST /api/v1/run/scum/facts, but the
production router rejected that path before the signature middleware because
runServiceRequest listed individual Run channel prefixes, and MySQL kept SCUM
rows inside the whole metadata snapshot instead of platform tables.
- /api/v1/run/ is now the signed machine channel space while
/api/v1/run/endpoints keeps normal bearer/admin authorization.
- MySQL gets real scum_user, scum_user_trajectory, scum_vehicle,
scum_vehicle_trajectory and scum_vehicle_lock tables with parameterized
per-row repositories instead of full snapshot rewrites. Snapshot-shaped
tables from the unreleased interim build are replaced, and SCUM rows still
inside a metadata snapshot are migrated once.
- Facts ingest verifies the target server plugin type and converges stale
online users to offline after SCUMUserOfflineAfter.
- The plugin page and browser read one bounded /scum/surface response instead
of five list calls per refresh.
Call-count budget for one server: per 5s facts batch, one SELECT plus one
INSERT/UPDATE per reported user and vehicle, one INSERT per moved trajectory
sample or new lock row, one bounded stale-user SELECT, and a trajectory
retention DELETE at most once per hour. One browser refresh issues one surface
request every 15s instead of five list requests.
npc0-hue