feat(auditlog): GetExecutionTreeAsync recursive execution-chain query

This commit is contained in:
Joseph Doherty
2026-05-21 18:21:49 -04:00
parent d35551efc2
commit 255dd95cd9
9 changed files with 462 additions and 0 deletions

View File

@@ -555,4 +555,197 @@ VALUES
BacklogTotal: 0L,
AsOfUtc: anchorUtc);
}
// Hard ceiling on chain depth for both the upward walk and the downward
// recursive CTE. The ParentExecutionId graph is a tree (acyclic by
// construction — each execution is minted fresh, its parent always
// pre-exists), so this is purely a guard against corrupt/pathological data:
// a cycle must surface as a bounded error rather than hang the server.
// Chains are shallow (1-2 levels typical) so the guard is never reached in
// practice.
private const int ExecutionChainMaxDepth = 32;
/// <summary>
/// Audit Log ParentExecutionId (Task 8) — returns the whole execution chain
/// containing <paramref name="executionId"/>, regardless of entry point.
/// </summary>
/// <remarks>
/// <para>
/// Two phases. <b>Walk up:</b> an iterative
/// <c>SELECT TOP 1 ParentExecutionId … WHERE ExecutionId = @cur AND ParentExecutionId IS NOT NULL</c>
/// climbs from the supplied node to the root — the last execution id with no
/// parent. The loop is capped at <see cref="ExecutionChainMaxDepth"/>
/// iterations; a purged/missing parent simply ends the climb early. <b>Walk
/// down:</b> a recursive CTE seeded at the root joins
/// <c>child.ParentExecutionId = parent.ExecutionId</c> to enumerate every
/// descendant, bounded by <c>OPTION (MAXRECURSION 32)</c> — corrupt cyclic
/// data raises a <see cref="SqlException"/> (msg 530) rather than spinning.
/// </para>
/// <para>
/// The chain's full execution-id set is the union of the rows'
/// <c>ExecutionId</c> and their <c>ParentExecutionId</c>, so an execution
/// referenced only as a parent — a "stub" that emitted no rows of its own —
/// is included. The final projection LEFT JOINs that id set back to
/// <c>AuditLog</c> and <c>GROUP BY</c>s, so a stub yields a node with
/// <c>RowCount = 0</c> and empty/null aggregates. The query is SELECT-only
/// (the audit writer role grants no UPDATE/DELETE — reads are unrestricted).
/// </para>
/// </remarks>
public async Task<IReadOnlyList<ExecutionTreeNode>> GetExecutionTreeAsync(
Guid executionId,
CancellationToken ct = default)
{
var conn = _context.Database.GetDbConnection();
var openedHere = false;
if (conn.State != System.Data.ConnectionState.Open)
{
await conn.OpenAsync(ct).ConfigureAwait(false);
openedHere = true;
}
try
{
// --- Phase 1: walk up to the root ---------------------------------
// Climb ParentExecutionId until a node has no parent (root) or the
// parent execution has no rows of its own (purged/stub — the climb
// cannot continue past a row-less node). The depth cap guards
// against a cycle in corrupt data; a tree never reaches it.
var rootExecutionId = executionId;
for (var depth = 0; depth < ExecutionChainMaxDepth; depth++)
{
Guid? parent;
await using (var upCmd = conn.CreateCommand())
{
upCmd.CommandText =
"SELECT TOP 1 ParentExecutionId FROM dbo.AuditLog " +
"WHERE ExecutionId = @cur AND ParentExecutionId IS NOT NULL;";
var pCur = upCmd.CreateParameter();
pCur.ParameterName = "@cur";
pCur.Value = rootExecutionId;
upCmd.Parameters.Add(pCur);
var result = await upCmd.ExecuteScalarAsync(ct).ConfigureAwait(false);
parent = result is null or DBNull ? null : (Guid)result;
}
if (parent is null)
{
// No parent row for the current node — it is the root (or a
// row-less stub at the top of what survives). Stop climbing.
break;
}
rootExecutionId = parent.Value;
}
// --- Phase 2: walk down from the root via a recursive CTE ---------
// Chain : seeded at the root, recursively pulls every distinct
// ExecutionId whose rows carry a ParentExecutionId already
// in the chain. SELECT DISTINCT in the recursive member is
// rejected by SQL Server, so the recursion walks raw rows
// and the outer query de-duplicates.
// ChainIds: the chain's full execution-id set = every ExecutionId in
// Chain UNIONed with every non-null ParentExecutionId — the
// UNION pulls in stub parents that emitted no rows.
// Final : LEFT JOIN ChainIds back to AuditLog and GROUP BY so a
// stub surfaces with RowCount 0 and NULL aggregates.
var nodes = new List<ExecutionTreeNode>();
await using (var downCmd = conn.CreateCommand())
{
downCmd.CommandText = @"
WITH Chain AS (
SELECT CAST(@root AS uniqueidentifier) AS ExecutionId
UNION ALL
SELECT a.ExecutionId
FROM dbo.AuditLog a
INNER JOIN Chain c ON a.ParentExecutionId = c.ExecutionId
WHERE a.ExecutionId IS NOT NULL
),
ChainIds AS (
SELECT DISTINCT ExecutionId FROM Chain
UNION
SELECT DISTINCT a.ParentExecutionId
FROM dbo.AuditLog a
INNER JOIN Chain c ON a.ExecutionId = c.ExecutionId
WHERE a.ParentExecutionId IS NOT NULL
)
SELECT
ids.ExecutionId AS [ExecutionId],
MIN(a.ParentExecutionId) AS [ParentExecutionId],
COUNT(a.EventId) AS [RowCount],
(SELECT STRING_AGG(d.Channel, ',')
FROM (SELECT DISTINCT a2.Channel FROM dbo.AuditLog a2
WHERE a2.ExecutionId = ids.ExecutionId) d) AS [Channels],
(SELECT STRING_AGG(d.Status, ',')
FROM (SELECT DISTINCT a2.Status FROM dbo.AuditLog a2
WHERE a2.ExecutionId = ids.ExecutionId) d) AS [Statuses],
MIN(a.SourceSiteId) AS [SourceSiteId],
MIN(a.SourceInstanceId) AS [SourceInstanceId],
MIN(a.OccurredAtUtc) AS [FirstOccurredAtUtc],
MAX(a.OccurredAtUtc) AS [LastOccurredAtUtc]
FROM ChainIds ids
LEFT JOIN dbo.AuditLog a ON a.ExecutionId = ids.ExecutionId
GROUP BY ids.ExecutionId
OPTION (MAXRECURSION 32);";
var pRoot = downCmd.CreateParameter();
pRoot.ParameterName = "@root";
pRoot.Value = rootExecutionId;
downCmd.Parameters.Add(pRoot);
await using var reader = await downCmd.ExecuteReaderAsync(ct).ConfigureAwait(false);
while (await reader.ReadAsync(ct).ConfigureAwait(false))
{
var nodeExecutionId = reader.GetGuid(0);
Guid? parentExecutionId = reader.IsDBNull(1) ? null : reader.GetGuid(1);
var rowCount = reader.GetInt32(2);
var channels = SplitAggregate(reader.IsDBNull(3) ? null : reader.GetString(3));
var statuses = SplitAggregate(reader.IsDBNull(4) ? null : reader.GetString(4));
var sourceSiteId = reader.IsDBNull(5) ? null : reader.GetString(5);
var sourceInstanceId = reader.IsDBNull(6) ? null : reader.GetString(6);
DateTime? firstOccurred = reader.IsDBNull(7) ? null : reader.GetDateTime(7);
DateTime? lastOccurred = reader.IsDBNull(8) ? null : reader.GetDateTime(8);
nodes.Add(new ExecutionTreeNode(
ExecutionId: nodeExecutionId,
ParentExecutionId: parentExecutionId,
RowCount: rowCount,
Channels: channels,
Statuses: statuses,
SourceSiteId: sourceSiteId,
SourceInstanceId: sourceInstanceId,
FirstOccurredAtUtc: firstOccurred,
LastOccurredAtUtc: lastOccurred));
}
}
return nodes;
}
finally
{
if (openedHere)
{
await conn.CloseAsync().ConfigureAwait(false);
}
}
}
/// <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
/// list rather than a single empty string.
/// </summary>
private static IReadOnlyList<string> SplitAggregate(string? aggregate)
{
if (string.IsNullOrEmpty(aggregate))
{
return Array.Empty<string>();
}
return aggregate
.Split(',', StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries)
.Distinct(StringComparer.Ordinal)
.OrderBy(v => v, StringComparer.Ordinal)
.ToArray();
}
}