fix(intergral/alerting): broadcast alerts to peers under HA partitioning

[jhawksley-intergral]

Two changes, both small.

CI trigger. The push workflows filtered on branches: [ "*" ], which does not match /. Since the pull_request triggers were dropped in aabd85c, every feat/… and fix/… branch has therefore had no CI at all, silently — no run has ever existed for a slash-named branch. Widened to "**". This is why the fix is bundled here: without it this branch could not build.

False resolved notifications under partitioning. With ha_scheduler_partitioning_enabled, each peer runs its own Alertmanager and only receives the alerts its own scheduler evaluates. When a topology change reassigns a firing rule, the previous owner is left holding a copy that nothing refreshes; it expires after nextEndsTime and that peer dispatches a resolved notification for a rule still firing on its new owner.

Upstream already has the mechanism for this — MultiOrgAlertmanager.BroadcastAlerts, built for ha_single_node_evaluation — so this just widens the existing gate to cover partitioning, letting the current owner refresh every peer’s copy.

Measured on a 5-node cluster with a receiver that actually delivers, hard-killing 2 peers: 5 false resolved notifications delivered before, 1 after, for alerts the API concurrently reported as firing. Duplicate firing notifications were checked and are not a concern — deliveries stay at one per repeat_interval, deduplicated by the gossiped notification log.

This is a mitigation, not a complete fix. Gossip is best effort and the residual case is not yet diagnosed, so a topology change can still emit an occasional false resolved. Flagging that explicitly because it bears on whether the 12.4.8 RC can be promoted. No config surface change — it activates from the existing partitioning flag.

[codehub-agent]

What this PR does

What this PR does

Widens the existing alert-broadcast gate to also trigger under HA scheduler partitioning, and fixes CI branch-name globs that silently disabled builds on slash-named branches.

• NewAlertsRouter's broadcast flag now activates on HASingleNodeEvaluation || HASchedulerPartitioningEnabled, so MultiOrgAlertmanager.BroadcastAlerts also runs when partitioning is enabled, refreshing firing-alert copies on peers that no longer own the rule.
• Mitigates false "resolved" notifications from stale, unrefreshed alert copies after topology changes (measured 5→1 false resolves per reassignment on a 5-node cluster); not a complete fix — gossip is best-effort and occasional false resolves can still occur.
• CI workflow branches filters changed from [ "*" ] to [ "**" ] in three workflow files, since * doesn't match branch names containing / (e.g. feat/…, fix/…), which had left those branches without any push-triggered CI.
• No config surface change; behavior is gated entirely by the pre-existing ha_scheduler_partitioning_enabled flag.

Risk: Low direct blast radius (5/83,962 symbols, x1.0 amplification, 0 execution roots) — change is a boolean-expression widening on an already-exercised code path, but the fix is explicitly described as a mitigation, not a complete fix, so residual false-resolved notifications can still occur under topology changes.

Flagged because it touches a contract (Historian) — callers elsewhere may be affected.

⚡ Blast radius

4 files · 12 symbols touched · 7 call sites elsewhere in the codebase

symbol	kind	fan-in	file
subscribeToFolderChanges	Function	2	pkg/services/ngalert/ngalert.go
shouldRun	Method	2	pkg/services/ngalert/ngalert.go
configureHistorianBackend	Function	2	pkg/services/ngalert/ngalert.go
init	Method	1	pkg/services/ngalert/ngalert.go
ApplyStateHistoryFeatureToggles	Function	1	pkg/services/ngalert/ngalert.go
createRemoteAlertmanager	Function	1	pkg/services/ngalert/ngalert.go

⚠️ Contracts touched (interfaces or fan-in ≥ 10): Historian — callers elsewhere may be affected.

Explore this PR's blast radius → · graph: workspace c3075765a84f · by CodeHub

💰 Agent economics

no agent sessions have reported to this workspace in the last 30 days, so there is no spend to attribute. Set up agent telemetry →

Full economics → · by CodeHub