feat(audit): M5.6 SourceNode sentinel backfill (purge-role) + CLI + runbook note (T5)

This commit is contained in:
Joseph Doherty
2026-06-16 22:02:21 -04:00
parent de2968b03d
commit 55630b48b6
12 changed files with 1399 additions and 10 deletions
@@ -716,6 +716,102 @@ VALUES
.ToListAsync(ct);
}
/// <inheritdoc />
public async Task<long> BackfillSourceNodeAsync(
string sentinel,
DateTime before,
int batchSize,
CancellationToken ct = default)
{
if (string.IsNullOrWhiteSpace(sentinel))
{
throw new ArgumentException("Sentinel must be a non-empty value.", nameof(sentinel));
}
if (batchSize <= 0)
{
throw new ArgumentOutOfRangeException(nameof(batchSize), batchSize, "Batch size must be > 0.");
}
var beforeUtc = DateTime.SpecifyKind(before.ToUniversalTime(), DateTimeKind.Utc);
// M5.6 (T5) SourceNode sentinel backfill. This is the ONE sanctioned UPDATE
// against dbo.AuditLog in the codebase. It touches ONLY rows where
// SourceNode IS NULL AND OccurredAtUtc < @before — rows that pre-date the
// M5.6 feature and whose node-of-origin is UNKNOWABLE. The sentinel (default
// "unknown") makes that explicit. ExecutionId/ParentExecutionId are PERSISTED
// COMPUTED columns derived from DetailsJson — mutating DetailsJson is forbidden
// under the append-only invariant, so those stay NULL on pre-feature rows.
//
// Maintenance path (NOT the writer role): runs on the same connection used for
// SwitchOutPartitionAsync (partition-switch DDL), which requires a role that
// holds UPDATE — the append-only scadabridge_audit_writer role has only
// INSERT + SELECT.
//
// Bounded + idempotent: UPDATE TOP (@batch) caps the log/lock footprint per
// statement; the loop exits when a batch updates 0 rows. Re-running after a
// crash simply resumes where it left off.
//
// The trailing AUDIT-PURGE-ALLOWED marker on the UPDATE line below is the
// single narrow exemption the append-only CI guard (AuditLogAppendOnlyGuardTests)
// recognises for an UPDATE; any other UPDATE targeting AuditLog still trips the guard.
const string updateBatchSql =
"UPDATE TOP (@batch) dbo.AuditLog SET SourceNode = @sentinel WHERE SourceNode IS NULL AND OccurredAtUtc < @before;"; // AUDIT-PURGE-ALLOWED: SourceNode sentinel backfill (M5.6 T5), maintenance path
long totalUpdated = 0;
var conn = _context.Database.GetDbConnection();
var openedHere = false;
if (conn.State != System.Data.ConnectionState.Open)
{
await conn.OpenAsync(ct).ConfigureAwait(false);
openedHere = true;
}
try
{
while (true)
{
ct.ThrowIfCancellationRequested();
await using var cmd = conn.CreateCommand();
cmd.CommandText = updateBatchSql;
var pBatch = cmd.CreateParameter();
pBatch.ParameterName = "@batch";
pBatch.Value = batchSize;
cmd.Parameters.Add(pBatch);
var pSentinel = cmd.CreateParameter();
pSentinel.ParameterName = "@sentinel";
pSentinel.Value = sentinel;
cmd.Parameters.Add(pSentinel);
var pBefore = cmd.CreateParameter();
pBefore.ParameterName = "@before";
pBefore.Value = beforeUtc;
cmd.Parameters.Add(pBefore);
var rows = await cmd.ExecuteNonQueryAsync(ct).ConfigureAwait(false);
if (rows <= 0)
{
break;
}
totalUpdated += rows;
}
}
finally
{
if (openedHere)
{
await conn.CloseAsync().ConfigureAwait(false);
}
}
return totalUpdated;
}
/// <summary>
/// Splits a <c>STRING_AGG</c> comma-joined value into a distinct, ordered
/// list. A null/empty aggregate (a stub node with no rows) yields an empty