Plan değişimi P410

(@keys nvarchar(4000),@from datetime,@to datetime)SELECT q.ServerID, q.DatabaseName, q.query_hash, q.StatDateUtc, SUM(q.TotalDurationMs) / NULLIF(SUM(q.Executions), 0)
FROM OPENJSON(@keys) WITH (s int '$.s', d sysname '$.d', h varchar(18) '$.h') AS k
JOIN perf.QueryDailyStats AS q ON q.ServerID = k.s AND q.DatabaseName = k.d AND q.query_hash = CONVERT(binary(8), k.h, 1)
WHERE q.StatDateUtc >= @from AND q.StatDateUtc <= @to
GROUP BY q.ServerID, q.DatabaseName, q.query_hash, q.StatDateUtc

Plan change case #P410 | DB: DB-11 | Improvement | Confidence: Medium (review pending)
Query: (@keys nvarchar(?),@from datetime,@to datetime)SELECT q.ServerID, q.DatabaseName, q.query_hash, q.StatDateUtc, SUM(q.TotalDurationMs) / NULLIF(SUM(q.Executions), ?) FROM OPENJSON(@keys) WITH (s int ?, d sysname ?, h varchar(?) ?) AS k JOIN perf.QueryDailyStats AS q ON q.ServerID = k.s AND q.DatabaseName = k.d AND q.query_hash = CONVERT(binary(?), k.h, ?) WHERE q.StatDateUtc >= @from AND q.StatDateUtc <= @to GROUP BY q.ServerID, q.DatabaseName, q.query_hash, q.StatDateUtc
Change: 2026-09-30 22:00 UTC, plan 25803 → 25811, pattern: Alternating (13 switches)
Impact (Query Store, per execution): duration 1310.6 → 3.4 ms (-100%); CPU 1292.3 → 3.4 ms (-100%); reads 460 → 911 (+98%); rows 253.37 → 231.48 (-9%)
Plan difference: Access only in the old plan: perf.QueryDailyStats Clustered Index Seek [PK_QueryDailyStats]; Access only in the new plan: perf.QueryDailyStats Index Seek [IX_QueryDailyStats_Query]; perf.QueryDailyStats Clustered Index Seek [PK_QueryDailyStats] (lookup); Join only in the old plan: Nested Loops (Inner Join); Join only in the new plan: Nested Loops (Inner Join) ×2
Candidate cause: PARAM — The two plans alternate in the same period (13 switches): plan choice depends on the parameter value rather than an event (parameter sensitivity).
Counter-evidence: ORDER_UNCERTAIN — STATS candidate ruled out: the time of the perf.QueryDailyStats.PK_QueryDailyStats event (2026-09-30 22:56) spans the change, so whether it happened before or after is unknown.
Counter-evidence: ORDER_UNCERTAIN — STATS candidate ruled out: the time of the perf.QueryDailyStats._WA_Sys_00000003_5D80D6A1 event (2026-09-30 22:56) spans the change, so whether it happened before or after is unknown.
Counter-evidence: ORDER_UNCERTAIN — STATS candidate ruled out: the time of the perf.QueryDailyStats._WA_Sys_00000004_5D80D6A1 event (2026-09-30 22:56) spans the change, so whether it happened before or after is unknown.
Counter-evidence: ORDER_UNCERTAIN — STATS candidate ruled out: the time of the perf.QueryDailyStats.IX_QueryDailyStats_Query event (2026-09-30 22:56) spans the change, so whether it happened before or after is unknown.
Counter-evidence: METADATA_GAP — Metadata collection started on 2026-10-01 and the change was on 2026-09-30: index, statistics and module events could not be evaluated (no data, not no events).
Confidence rationale: Medium: a single candidate (PARAM), not mechanical (no direct link to the plan is proven). Weak counter-evidence: METADATA_GAP.
Note: this is a candidate cause, not proof of the root cause.