Plan değişimi P446

(@instance_uuid uniqueidentifier,@window_days int)DELETE r
FROM dbo.agent_job_runs AS r
JOIN dbo.agent_jobs AS j ON j.agent_job_uuid = r.agent_job_uuid
WHERE j.instance_uuid = @instance_uuid
  AND r.run_at < DATEADD(day, -@window_days, SYSUTCDATETIME())

Plan change case #P446 | DB: DB-9 | Improvement | Confidence: Medium (review pending)
Query: (@instance_uuid uniqueidentifier,@window_days int)DELETE r FROM dbo.agent_job_runs AS r JOIN dbo.agent_jobs AS j ON j.agent_job_uuid = r.agent_job_uuid WHERE j.instance_uuid = @instance_uuid AND r.run_at < DATEADD(day, -@window_days, SYSUTCDATETIME())
Change: 2026-09-28 11:00 UTC, plan 483 → 521, pattern: Alternating (6 switches)
Impact (Query Store, per execution): duration 1024.7 → 0.1 ms (-100%); CPU 12.4 → 0.1 ms (-99%); reads 144 → 2 (-99%); rows 0 → 0
Plan difference: Access only in the old plan: dbo.agent_jobs Clustered Index Seek [CIX_agent_jobs]; dbo.agent_job_runs Clustered Index Seek [CIX_agent_job_runs]; Access only in the new plan: dbo.agent_job_runs Index Seek [IX_agent_job_runs_window]; dbo.agent_job_runs Clustered Index Seek [CIX_agent_job_runs] (lookup); dbo.agent_jobs Index Seek [PK_agent_jobs]; Join only in the old plan: Nested Loops (Inner Join); Join only in the new plan: Nested Loops (Inner Join) ×3
Candidate cause: FORCED — The new plan is forced in Query Store (sp_query_store_force_plan); Query Store does not record when it was forced.
Candidate cause: GROWTH — From the old plan's period to the change, dbo.agent_job_runs rows changed +162% and size +130% (2026-09-27 → 2026-09-28).
Candidate cause: PARAM — The two plans alternate in the same period (6 switches): plan choice depends on the parameter value rather than an event (parameter sensitivity).
Counter-evidence: HIGH_VARIANCE — The old plan's duration standard deviation is 1198% of the average: the average is not reliable.
Counter-evidence: METADATA_GAP — Metadata collection started on 2026-09-27 and the change was on 2026-09-28: index, statistics and module events could not be evaluated (no data, not no events).
Confidence rationale: Medium: more than one candidate (FORCED, GROWTH, PARAM) — competing explanations. Weak counter-evidence: HIGH_VARIANCE, METADATA_GAP.
Note: this is a candidate cause, not proof of the root cause.