The host does not guarantee IHostedService.StopAsync is driven before the DI
container is disposed — WebApplicationFactory's teardown reaches Dispose first
— so cancelling the internal CTS from StopAsync threw ObjectDisposedException
and aborted the host's whole shutdown sequence. Four services shared the same
copy-pasted lifecycle and the same two races: StopAsync cancelling an already-
disposed CTS, and StartAsync reading _cts.Token lazily inside the Task.Run
lambda, which faults the loop task the host awaits when Dispose wins that race.
Each service now captures the token on the caller's thread, tolerates a
disposed CTS, and cancels-before-disposing so the loop is always signalled and
its pending Task.Delay sees a cancelled token rather than a dead source.
SiteAuditBacklogReporter also gains the outer OperationCanceledException guard
its sibling SiteAuditRetentionService already carried (arch-review 04 R2, R7),
without which a shutdown landing mid-probe threw TaskCanceledException out of
Host.StopAsync.
Surfaced while verifying the Gitea #15 test-harness fix: in Host.Tests the
aborted teardown skipped the fixture's env-var restore, contaminating every
later test in the run.
Refs: Gitea #15
AddSiteEventLogHealthMetricsBridge registered via AddHostedService(factory-lambda),
which sets ImplementationFactory and leaves ImplementationType null. The prior
ImplementationType == guard was therefore silently dead — a second call would spin
up a second SiteEventLogFailureCountReporter. Fix: add a private
SiteEventLogHealthMetricsBridgeMarker singleton and guard on its ServiceType instead.
Also corrects the cycle-path comment in both ServiceCollectionExtensions.cs and
SiteEventLogFailureCountReporter.cs: StoreAndForward.csproj does reference
SiteEventLogging.csproj, so the transitive path HealthMonitoring → StoreAndForward →
SiteEventLogging is real, but adding a direct HealthMonitoring → SiteEventLogging
reference would NOT create a cycle (SiteEventLogging has no back-edge to HealthMonitoring).
The Func<long> seam is a coupling-avoidance measure, not a cycle-breaker.
Adds AddSiteEventLogHealthMetricsBridgeTests.AddSiteEventLogHealthMetricsBridge_IsIdempotent_DoesNotDoubleRegister_HostedService
as a regression test (builds provider and asserts exactly one reporter via GetServices<IHostedService>().OfType<T>()).
Perf re-baseline (HotPathLatencyTests): empirical p95 on Apple M-series Release
build: 4KB DetailsJson slow path ≈14 µs, small-DetailsJson no-redactors ≈2 µs,
true no-op fast path ≈0 µs. Thresholds updated: 200 µs / 30 µs / 5 µs (≈15×
headroom for contested CI runners). Old thresholds (50 µs / 10 µs) were set for
the pre-C3 typed-field path; canonical JSON parse+rewrite is empirically faster.
Adds a third test (Filter_Apply_NoDetailsJson_FastPath) that asserts same-instance
return on the DetailsJson-null + within-cap fast path. Env-var overrides retained.
CollapseAuditLogToCanonicalMigrationTests (new): three MSSQL-gated [SkippableFact]
tests verifying Action/Category/Outcome projection, NULL Actor, DetailsJson codec
round-trip, and all six persisted computed columns (Kind/Status/SourceSiteId/
ExecutionId/ParentExecutionId) for ApiOutbound, InboundAuthFailure, and Failed-
status rows.
AddAuditLogTableMigrationTests: rename CreatesFiveNamedIndexes →
CreatesNineNamedIndexes; expand coverage from 5 original indexes to all 9 named
non-clustered indexes present after CollapseAuditLogToCanonical (adds
IX_AuditLog_Execution, IX_AuditLog_ParentExecution, IX_AuditLog_Node_Occurred,
UX_AuditLog_EventId).
Dead-cref cleanup: zero references to the deleted IAuditPayloadFilter /
DefaultAuditPayloadFilter / SafeDefaultAuditPayloadFilter types remain in any
.cs file (source or test). 26 occurrences across 13 files replaced with correct
references to IAuditRedactor / ScadaBridgeAuditRedactor / SafeDefaultAuditRedactor
or reworded as plain prose.
Residual sweep: no unused transitional code found beyond the acknowledged
"C3 transitional shim" comments on IngestedAtUtc stamping (active code, not dead).
Joseph Doherty