perf(audit-log): bound GetExecutionTreeAsync edge scan to a root-anchored time window (partition elimination)

This commit is contained in:
Joseph Doherty
2026-07-09 07:10:35 -04:00
parent f1957606ff
commit 5d0fcd4e66
3 changed files with 103 additions and 0 deletions
+11
View File
@@ -209,6 +209,17 @@ tag subscription) are also wired: trigger-driven runs carry `ParentExecutionId =
NULL` (top-level roots), and any nested `CallScript`/`CallShared` they perform
chains as above. The schema is unchanged — no further tag-cascade work is deferred.
**Execution-tree traversal bound.** `GetExecutionTreeAsync` first walks up
`ParentExecutionId` to the chain root, then walks down via a recursive CTE. The
down-walk's edge scan is bounded to a window anchored at the root's **first event**:
`[rootFirst 1 h, rootFirst + 7 d)` (configurable constants). Execution trees span
minutes, not years — an inbound request and everything it spawns fire inside one
operational burst — so the window lets SQL Server eliminate every `AuditLog`
partition outside it, turning a full-table `DISTINCT` scan into a seek on
`IX_AuditLog_Execution` (arch-review 04, P2). A genuine descendant stamped **beyond
the 7-day span is excluded by design**; a row-less stub root (its first event purged)
has no anchor and falls back to the unbounded scan (correctness over speed).
## The Site-Local `AuditLog` (SQLite)
A SQLite database file on each site node, alongside the Store-and-Forward
@@ -665,6 +665,16 @@ VALUES
// practice.
private const int ExecutionChainMaxDepth = 32;
// Execution trees span minutes, not years — an inbound request and every script it
// spawns all fire inside one operational burst. Anchoring the Edges CTE scan to a
// narrow window around the root's first event lets SQL Server eliminate every
// AuditLog partition outside it, turning a full-table DISTINCT scan into a seek on
// IX_AuditLog_Execution over one or two months (arch-review 04, P2). The slack absorbs
// minor clock skew before the root; the max span is the documented traversal bound —
// a genuine descendant stamped beyond it is excluded by design.
private static readonly TimeSpan ExecutionTreeWindowSlack = TimeSpan.FromHours(1);
private static readonly TimeSpan ExecutionTreeMaxSpan = TimeSpan.FromDays(7);
/// <inheritdoc />
public async Task<IReadOnlyList<ExecutionTreeNode>> GetExecutionTreeAsync(
Guid executionId,
@@ -713,6 +723,31 @@ VALUES
rootExecutionId = parent.Value;
}
// --- Between phases: bound the down-walk to a root-anchored window ---
// The root's first event timestamp anchors a [rootFirst - slack, rootFirst + maxSpan)
// window used to prune the Edges CTE scan to a handful of partitions. A row-less stub
// root (purged/no-action parent) has no MIN — leave the scan unbounded in that
// degenerate case (correctness over speed).
DateTime? rootFirstOccurred;
await using (var winCmd = conn.CreateCommand())
{
winCmd.CommandText =
"SELECT MIN(OccurredAtUtc) FROM dbo.AuditLog WHERE ExecutionId = @root;";
var pWinRoot = winCmd.CreateParameter();
pWinRoot.ParameterName = "@root";
pWinRoot.Value = rootExecutionId;
winCmd.Parameters.Add(pWinRoot);
var winResult = await winCmd.ExecuteScalarAsync(ct).ConfigureAwait(false);
rootFirstOccurred = winResult is null or DBNull ? null : (DateTime)winResult;
}
// Build the optional window predicate for the Edges CTE. When the root has rows the
// scan is confined to the window (partition elimination); otherwise it stays unbounded.
var edgesWindowPredicate = rootFirstOccurred is null
? string.Empty
: "AND OccurredAtUtc >= @winStart AND OccurredAtUtc < @winEnd";
// --- Phase 2: walk down from the root via a recursive CTE ---------
// Edges : a non-recursive, DISTINCT (ExecutionId, ParentExecutionId)
// edge set distilled from AuditLog. Recursing over edges
@@ -743,6 +778,7 @@ VALUES
SELECT DISTINCT ExecutionId, ParentExecutionId
FROM dbo.AuditLog
WHERE ExecutionId IS NOT NULL
{edgesWindowPredicate}
),
Chain AS (
-- Anchor: the root execution id, seeded as a literal so
@@ -797,6 +833,19 @@ VALUES
pRoot.Value = rootExecutionId;
downCmd.Parameters.Add(pRoot);
if (rootFirstOccurred is { } rootFirst)
{
var pWinStart = downCmd.CreateParameter();
pWinStart.ParameterName = "@winStart";
pWinStart.Value = rootFirst - ExecutionTreeWindowSlack;
downCmd.Parameters.Add(pWinStart);
var pWinEnd = downCmd.CreateParameter();
pWinEnd.ParameterName = "@winEnd";
pWinEnd.Value = rootFirst + ExecutionTreeMaxSpan;
downCmd.Parameters.Add(pWinEnd);
}
await using var reader = await downCmd.ExecuteReaderAsync(ct).ConfigureAwait(false);
while (await reader.ReadAsync(ct).ConfigureAwait(false))
{
@@ -1008,6 +1008,49 @@ public class AuditLogRepositoryTests : IClassFixture<MsSqlMigrationFixture>
}
}
[SkippableFact]
public async Task ExecutionTree_Bounds_EdgeScan_To_Root_Window()
{
Skip.IfNot(_fixture.Available, _fixture.SkipReason);
// arch-review 04 (P2): the down-walk Edges CTE is anchored to a
// [rootFirst - 1h, rootFirst + 7d) window around the root's first event, enabling
// partition elimination instead of a full-table DISTINCT scan. A genuine descendant
// stamped BEYOND that 7-day span is excluded by design; a normal in-window child is
// still returned.
var siteId = NewSiteId();
await using var context = CreateContext();
var repo = new AuditLogRepository(context);
var rootExec = Guid.NewGuid();
var childExec = Guid.NewGuid();
var farDescendantExec = Guid.NewGuid();
var unrelatedExec = Guid.NewGuid();
var t0 = new DateTime(2026, 10, 5, 9, 0, 0, DateTimeKind.Utc);
// Root + an in-window child (T0 + 5 min) — both must be returned.
await repo.InsertIfNotExistsAsync(NewEvent(siteId, occurredAtUtc: t0, executionId: rootExec));
await repo.InsertIfNotExistsAsync(NewEvent(siteId, occurredAtUtc: t0.AddMinutes(5), executionId: childExec, parentExecutionId: rootExec));
// A genuine descendant of the child, but stamped 30 days out — beyond the 7-day
// traversal bound, so it is documented-excluded from the tree.
await repo.InsertIfNotExistsAsync(NewEvent(siteId, occurredAtUtc: t0.AddDays(30), executionId: farDescendantExec, parentExecutionId: childExec));
// An unrelated execution far in time sharing no ancestry — never part of the chain
// (asserts the window bound does not accidentally pull in strangers).
await repo.InsertIfNotExistsAsync(NewEvent(siteId, occurredAtUtc: t0.AddDays(30), executionId: unrelatedExec));
var tree = await repo.GetExecutionTreeAsync(rootExec);
var ids = tree.Select(n => n.ExecutionId).ToHashSet();
Assert.Contains(rootExec, ids);
Assert.Contains(childExec, ids);
Assert.DoesNotContain(farDescendantExec, ids);
Assert.DoesNotContain(unrelatedExec, ids);
Assert.Equal(2, tree.Count);
}
[SkippableFact]
public async Task GetExecutionTree_StubParentNode()
{