mirror of
https://github.com/rcourtman/Pulse.git
synced 2026-10-25 04:59:43 +00:00
rcourtman
1b081c6475
Alert lifecycle changes (fired, acknowledged, snoozed, resolved) are
journaled with the alert's source-native resource ID.
`MonitorAdapter.RecordChange` only mapped full Docker container IDs, so
Proxmox node, guest and storage alerts and host-agent alerts landed
under IDs such as `instance-node`, `instance:node:vmid` or
`agent:<host>`. Resource facets, resource intelligence
(`recent_changes`) and Assistant resource context read history by
canonical ID, so they never showed those alerts. Incident timelines were
unaffected because they read by alert identifier.
**Change**
- `RecordChange` resolves non-Docker references to the canonical
resource and records a history binding, so the reference keeps reaching
the same history after a guest migration or inventory removal.
- Only durable identities bind: canonical ID, retired era, source ID,
node-scoped guest reference, `agent:<host ID>`. Names and hostnames
never bind (a VM named `pulse-system` cannot capture system alerts). A
reference that two resources answer to binds to neither. Docker keeps
its full-ID rule.
- Rows already journaled under a source reference join canonical history
without being rewritten. Each registry generation binds the unbound
alert references it can resolve, both those found at startup and those
written before inventory named the resource, retrying each for 30
minutes.
**Evidence** (private mock stack, same data dir, origin/main first, then
this branch)
- main: 51 of 52 alert references were source-shaped and there were zero
bindings. `GET /api/resources/{id}/facets` for affected nodes and guests
returned only `state_transition`.
- this branch: the first registry generation logged `bound 17 alert
history references`. Every non-Docker whole-resource reference is now
bound: node metric and connectivity alerts, guests, Proxmox storage,
`agent:host-linux-1` and `pmg-main`. Facets and
`/api/ai/intelligence?resource_id=` now include
`alert_fired`/`alert_resolved` for the VM, node and storage checked.
- Still unbound, as before: ZFS pool/device and agent-disk sub-resource
references (no resource of their own) and Docker host/service/short-ID
references.
**Where it shows**
In the UI, only the full-presentation `ResourceDetailDrawer` requests
this history. Proxmox node and Agents rows open the `table-row` drawer,
and guest and storage rows open their own drawers; none of these fetch
it (confirmed live: the node History tab makes no facets or intelligence
request). So the user-visible gain is in Assistant and Patrol resource
context and the facets and intelligence APIs, not in a Proxmox drawer.
**Tests**
`TestHistoryIdentityMonitorAdapterResolvesProxmoxAlertReferences` and
`TestHistoryIdentityBindsLegacyAlertRowsFromRegistryGenerations`
(registered `history_identity_test.go`), plus
`TestProxmoxAlertTimelineUsesCanonicalHistoryIdentity` through the real
alert-manager callback. The monitor test fails on main. Mutation checks
confirm the new tests catch alias-based resolution, a missing post-scan
retry and ignored ambiguity.