Joseph Doherty
|
04d267d1ea
|
Phase 2 PR 13 — port GalaxyRuntimeProbeManager state machine + wire per-platform ScanState probing. PR 8 gave operators the gateway-level transport signal (MxAccessClient.ConnectionStateChanged → OnHostStatusChanged tagged with the Wonderware client identity) — enough to detect when the entire MXAccess COM proxy dies, but silent when a specific or goes down while the gateway stays alive. This PR closes that gap: a pure-logic GalaxyRuntimeProbeManager (ported from v1's 472-LOC GalaxyRuntimeProbeManager.cs, distilled to ~240 LOC of state machine without the OPC UA node-manager entanglement) lives in Backend/Stability/, advises <TagName>.ScanState per WinPlatform + AppEngine gobject discovered during DiscoverAsync, runs the Unknown → Running → Stopped state machine with v1's documented semantics preserved verbatim, and raises a StateChanged event on transitions operators should react to. MxAccessGalaxyBackend instantiates the probe manager in the constructor with SubscribeAsync/UnsubscribeAsync delegate pointers into MxAccessClient, hooks StateChanged to forward each transition through the same OnHostStatusChanged IPC event the gateway signal uses (HostName = platform/engine TagName, RuntimeStatus = 'Running'|'Stopped'|'Unknown', LastObservedUtcUnixMs from the state-change timestamp), so Admin UI gets per-host signals flowing through the existing PR 8 wire with no additional IPC plumbing. State machine rules ported from v1 runtimestatus.md: (a) ScanState is on-change-only — a stably-Running host may go hours without a callback, so Running → Stopped is driven only by explicit ScanState=false, never by starvation; (b) Unknown → Running is a startup transition and does NOT fire StateChanged (would paint every host as 'just recovered' at startup, which is noise and can clear Bad quality set by a concurrently-stopping sibling); (c) Stopped → Running fires StateChanged for the real recovery case; (d) Running → Stopped fires StateChanged; (e) Unknown → Stopped fires StateChanged because that's the first-known-bad signal operators need when a host is down at our startup time. MxAccessGalaxyBackend.DiscoverAsync calls _probeManager.SyncAsync with the runtime-host subset of the hierarchy (CategoryId == 1 WinPlatform or 3 AppEngine) as a best-effort step after building the Discover response — probe failures are swallowed so Discover still returns the hierarchy even if a per-host advise fails; the gateway signal covers the critical rung. SyncAsync is idempotent (second call with the same set is a no-op) and handles the diff on re-Discover for tag rename / host add / host remove. Subscribe failure rolls back the host's state entry under the lock so a later probe callback for a never-advised tag can't transition a phantom state from Unknown to Stopped and fan out a false host-down signal (the same protection v1's GalaxyRuntimeProbeManager had at line 237-243 of v1 with a captured-probe-string comparison under the lock). MxAccessGalaxyBackend.Dispose unsubscribes the StateChanged handler before disposing the probe manager to prevent dangling invocation-list references across reconnects, same discipline as PR 8's ConnectionStateChanged and PR 6's SubscriptionReplayFailed. Tests (12 new GalaxyRuntimeProbeManagerTests): Sync_subscribes_to_ScanState_per_host verifies tag.ScanState subscriptions are advised per Platform/Engine; Sync_is_idempotent_on_repeat_call_with_same_set verifies no duplicate subscribes; Sync_unadvises_removed_hosts verifies the diff unadvises gone hosts; Subscribe_failure_rolls_back_host_entry_so_later_transitions_do_not_fire_stale_events covers the rollback-on-subscribe-fail guard; Unknown_to_Running_does_not_fire_StateChanged preserves the startup-noise rule; Running_to_Stopped_fires_StateChanged_with_both_states asserts OldState and NewState are both captured in the transition record; Stopped_to_Running_fires_StateChanged_for_recovery verifies the recovery case; Unknown_to_Stopped_fires_StateChanged_for_first_known_bad_signal preserves the first-bad rule; Repeated_Good_Running_callbacks_do_not_fire_duplicate_events verifies the state-tracking de-dup; Unknown_callback_for_non_tracked_probe_is_dropped asserts a foreign callback is silently ignored; Snapshot_reports_current_state_for_every_tracked_host covers the dashboard query hook; IsRuntimeHost_recognizes_WinPlatform_and_AppEngine_category_ids asserts the CategoryId filter. Galaxy.Host.Tests Unit suite 75 pass / 0 fail (12 new probe + 63 pre-existing). Galaxy.Host builds clean (0 errors / 0 warnings). Branches off v2.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-04-18 07:27:56 -04:00 |
|