When the same physical Ceph cluster is reported by both the Proxmox API
poller and a Pulse host-agent (same FSID), v5 stored it under two pool-ID
namespaces and ran alert evaluation on the raw, un-deduplicated list,
producing duplicate/flapping Ceph pool alerts and a per-pool override that
appeared to revert between identities.
v6 closes this without the v5 'agent:'-prefix patch: both ingest paths
reconcile by FSID before alert evaluation. State.UpsertCephCluster
(host-agent) and State.UpdateCephClustersForInstance (Proxmox) both funnel
through upsertCephClusterInSlice, which collapses same-FSID clusters into a
single reconciled entry, and both call sites run checkCephPoolStorage on the
STORED cluster. The per-pool override resolves across source aliases via
storageThresholdLookupIDs/cephPoolStorageSourceAliasID, and
clearStorageAliasAlerts removes any alias-keyed alert.
This adds a regression test asserting the observable outcome for the exact
dual-source scenario (host-agent + Proxmox API, same FSID, one pool at ~61%
with a 50% per-pool override): exactly one active Ceph pool usage alert,
override honored, no second/flapping alert across either ingest order or
repeated interleaved cycles. No behavior change.
rcourtman