Central singleton (M6-T4 Bundle C) that drives the daily AuditLog partition
purge. On a configurable timer (default 24 hours) the actor:
1. Queries IAuditLogRepository.GetPartitionBoundariesOlderThanAsync for
monthly boundaries whose latest OccurredAtUtc is older than
DateTime.UtcNow - AuditLogOptions.RetentionDays.
2. For each eligible boundary calls SwitchOutPartitionAsync, which runs
the drop-and-rebuild dance around UX_AuditLog_EventId.
3. Publishes AuditLogPurgedEvent(boundary, rowsDeleted, durationMs) on
the actor-system EventStream so the Bundle E central health collector
and ops surfaces can subscribe without coupling to this actor.
Co-changes:
* SwitchOutPartitionAsync returns long (rows deleted) — sampled BEFORE the
switch via COUNT_BIG over the per-partition filter so the count
reflects what the switch removed, not a post-purge scan of a table that
no longer exists. All stub implementations updated.
* AuditLogPurgeOptions: IntervalHours (default 24), IntervalOverride for
tests, Interval property resolving either.
* AuditLogPurgedEvent: record with MonthBoundary, RowsDeleted, DurationMs.
Behavior:
* Continue-on-error per boundary — one partition that throws does NOT
abandon the rest of the tick.
* DI scope opened per tick (IAuditLogRepository is a SCOPED EF Core
service); mirrors SiteAuditReconciliationActor and AuditLogIngestActor.
* SupervisorStrategy Resume keeps the singleton alive across leaked
exceptions.
* EventStream capture BEFORE the first await — Context is unsafe after
await in async receive handlers (same pattern as Sender-capture in
AuditLogIngestActor.OnIngestAsync).
Tests:
* Tick_Fires_OnDailyInterval — visible timer side effect.
* Tick_OldPartitions_SwitchedOut — both seeded boundaries purged.
* Tick_NewerPartitions_Untouched — empty enumerator → no switches.
* Tick_PublishesPurgedEvent_WithRowCount — AuditLogPurgedEvent carries
RowsDeleted and DurationMs.
* Tick_SwitchThrows_OtherPartitionsStillProcessed — continue-on-error.
* Threshold_UsesAuditLogOptionsRetentionDays — non-default 30-day window
computed from UtcNow - RetentionDays.
* EndToEnd_RealPartition_RowsRemoved_PurgedEventPublished — TestKit +
MsSqlMigrationFixture: real partitioned table, Jan-2026 row purged,
Apr-2026 row kept, AuditLogPurgedEvent observed via probe.
Replaces M1's NotSupportedException stub with the production drop-DROP-INDEX
→ CREATE-staging → SWITCH PARTITION → DROP-staging → CREATE-INDEX dance
documented in alog.md §4. UX_AuditLog_EventId is intentionally non-aligned
with ps_AuditLog_Month so single-column EventId uniqueness can be enforced
cheaply for InsertIfNotExistsAsync; SQL Server rejects ALTER TABLE SWITCH
while a non-aligned unique index is present, so the implementation drops
it, switches the partition data into a GUID-suffixed staging table on
[PRIMARY], drops staging (discarding the rows), and rebuilds the unique
index — all inside an explicit transaction with a CATCH that guarantees
the unique index is rebuilt regardless of failure point.
Also adds GetPartitionBoundariesOlderThanAsync to IAuditLogRepository: a
CROSS APPLY over sys.partition_range_values + per-partition MAX(OccurredAtUtc)
to enumerate retention-eligible months for the M6 purge actor (next commit).
Tests verify:
* Old partition's rows are removed; other months untouched
* UX_AuditLog_EventId is rebuilt after a successful switch
* InsertIfNotExistsAsync's first-write-wins idempotency still holds after switch
* On engineered SWITCH failure (inbound FK from a probe table), SqlException
propagates AND UX_AuditLog_EventId is still present (CATCH branch ran)
* GetPartitionBoundariesOlderThanAsync returns only boundaries whose partition's
MAX(OccurredAtUtc) is strictly older than the threshold; empty partitions
excluded
Two concurrent sessions can both pass the IF NOT EXISTS check and then both attempt the INSERT against UX_AuditLog_EventId; the loser surfaced as SqlException 2601 (or 2627 for PK violations) and aborted the audit write. First-write-wins idempotency is the documented contract, so the race outcome is semantically a no-op — catch the two duplicate-key error numbers and log at Debug, let any other SqlException bubble.
Tests:
- InsertIfNotExistsAsync_ConcurrentDuplicateInserts_ProduceExactlyOneRow: 50 parallel inserters with the same EventId end with exactly one row and no escaped exceptions.
- QueryAsync_Keyset_SameOccurredAtUtc_TiebreaksOnEventId: four rows sharing the same OccurredAtUtc page deterministically through the (OccurredAtUtc, EventId) keyset cursor — exercises the e.OccurredAtUtc == after && e.EventId.CompareTo(afterId) < 0 branch and verifies EF Core 10's Guid.CompareTo translation against SQL Server uniqueidentifier order (deferred Bundle D reviewer recommendation).
AuditLogRepository now takes an optional ILogger<AuditLogRepository> (NullLogger default, mirrors InboundApiRepository); DI registration unchanged.
EF Core implementation of IAuditLogRepository:
- InsertIfNotExistsAsync: single IF NOT EXISTS ... INSERT via
ExecuteSqlInterpolatedAsync, bypasses the change tracker. Enum
values converted to string in C# (columns are varchar(32) via
HasConversion<string>).
- QueryAsync: AsNoTracking, predicate-per-non-null-filter, keyset
paging on (OccurredAtUtc DESC, EventId DESC) — EF Core 10
translates Guid.CompareTo to a uniqueidentifier < comparison
natively (verified against MSSQL 2022).
- SwitchOutPartitionAsync: throws NotSupportedException naming M6;
the non-aligned UX_AuditLog_EventId unique index blocks
ALTER TABLE SWITCH PARTITION until the drop-and-rebuild dance
ships with the purge actor.
DI: AddScoped<IAuditLogRepository, AuditLogRepository>() added after
the NotificationOutboxRepository registration; existing DI smoke test
extended with an IAuditLogRepository assertion.
Integration tests (8 new) use the Bundle C MsSqlMigrationFixture and
scope by a per-test SourceSiteId guid so they neither collide nor
require cleanup.
Bundle D of the Audit Log #23 M1 Foundation plan.
