ServiceRadar v1.4.61

Managed add-on updates no longer freeze because one agent
stopped reporting, and object-store retention now actually
runs on the deployed core so a full bucket can reclaim space.

Whats New

1.4.61

• Add-ons: a decommissioned or unreporting agent no longer
  pauses a managed-add-on rollout for the rest of the fleet.
  Unreporting hosts are excluded from new rollouts the same
  way incompatible hosts already were, and a paused rollout
  can be reaped when the live hosts already have the
  candidate (PR #404, PR #410).
• core_elx: object-store retention is registered on the
  deployed release. The chart has been setting
  OBJECT_STORE_RETENTION_* since the feature shipped, but
  serviceradar_core_elx never scheduled the worker, so
  mirrored agent-release and native add-on artifacts were
  never reclaimed (PR #403).
• datasvc / web-ng: a full object store (JetStream DiscardNew
  at MaxBytes) returns ResourceExhausted naming the bucket
  and its cap, instead of an opaque gRPC Internal (PR #403).

Upgrade Notes

• If a managed add-on source is still sitting paused from an
  unreporting agent, the next reconcile should supersede it
  when the live hosts already have the candidate. If it stays
  paused, cancel that rollout so a new one can start without
  those hosts.
• Object-store retention now runs on core (default 03:00 UTC).
  A bucket that is already full still refuses new uploads
  until space is reclaimed.