Wait-Statistiken: Worauf der Server wartet

Befehl

WITH w AS (
    SELECT wait_type, waiting_tasks_count,
           wait_time_ms / 1000.0                         AS wait_s,
           (wait_time_ms - signal_wait_time_ms) / 1000.0 AS resource_s,
           signal_wait_time_ms / 1000.0                  AS signal_s
    FROM   sys.dm_os_wait_stats
    WHERE  waiting_tasks_count > 0
      AND  wait_type NOT LIKE N'SLEEP[_]%'
      AND  wait_type NOT IN (
           N'BROKER_EVENTHANDLER', N'BROKER_RECEIVE_WAITFOR', N'BROKER_TASK_STOP', N'BROKER_TO_FLUSH',
           N'BROKER_TRANSMITTER', N'CHECKPOINT_QUEUE', N'CLR_AUTO_EVENT', N'CLR_MANUAL_EVENT',
           N'DIRTY_PAGE_POLL', N'DISPATCHER_QUEUE_SEMAPHORE', N'FT_IFTS_SCHEDULER_IDLE_WAIT',
           N'HADR_FILESTREAM_IOMGR_IOCOMPLETION', N'HADR_WORK_QUEUE', N'LAZYWRITER_SLEEP',
           N'LOGMGR_QUEUE', N'ONDEMAND_TASK_QUEUE', N'PVS_PREALLOCATE', N'REQUEST_FOR_DEADLOCK_SEARCH',
           N'SNI_HTTP_ACCEPT', N'SOS_WORK_DISPATCHER', N'SP_SERVER_DIAGNOSTICS_SLEEP',
           N'SQLTRACE_BUFFER_FLUSH', N'SQLTRACE_INCREMENTAL_FLUSH_SLEEP', N'WAITFOR',
           N'XE_DISPATCHER_WAIT', N'XE_TIMER_EVENT', N'XE_LIVE_TARGET_TVF', N'PREEMPTIVE_XE_DISPATCHER',
           N'QDS_PERSIST_TASK_MAIN_LOOP_SLEEP', N'QDS_ASYNC_QUEUE', N'QDS_SHUTDOWN_QUEUE',
           N'QDS_CLEANUP_STALE_QUERIES_TASK_MAIN_LOOP_SLEEP', N'PWAIT_ALL_COMPONENTS_INITIALIZED',
           N'PWAIT_EXTENSIBILITY_CLEANUP_TASK', N'STARTUP_DEPENDENCY_MANAGER', N'WAIT_XTP_CKPT_CLOSE')
)
SELECT TOP (20) wait_type, waiting_tasks_count,
       CAST(wait_s AS DECIMAL(18,1))                                AS wait_s,
       CAST(100.0 * wait_s / SUM(wait_s) OVER () AS DECIMAL(5,1))   AS pct,
       CAST(resource_s AS DECIMAL(18,1))                            AS resource_s,
       CAST(signal_s AS DECIMAL(18,1))                              AS signal_s,
       CAST(1000.0 * wait_s / waiting_tasks_count AS DECIMAL(18,2)) AS avg_wait_ms
FROM   w
ORDER BY wait_s DESC;

-- Values accumulate since:
SELECT sqlserver_start_time FROM sys.dm_os_sys_info;

-- User requests waiting right now:
SELECT wt.session_id, wt.wait_type, wt.wait_duration_ms, wt.resource_description, wt.blocking_session_id
FROM   sys.dm_os_waiting_tasks wt
  JOIN sys.dm_exec_sessions s ON s.session_id = wt.session_id
WHERE  s.is_user_process = 1;

Was es tut

Filtert harmlose Hintergrund-Waits heraus und schlüsselt die gesamte Wartezeit des Servers nach Wait-Typ mit Prozentanteil und dem Anteil Resource vs. Signal (CPU-Warteschlange) auf; zeigt außerdem Benutzeranfragen, die gerade warten.

Wann einsetzen

Als erster Schritt einer Performance-Analyse, um den Engpass einzuordnen (I/O, CPU, Sperren, Log, Speicher, Parallelität); für Baseline-Vergleiche vorher/nachher.

Vorsicht

Die Werte kumulieren seit dem Neustart; für ein aktuelles Problem zwei Snapshots erstellen und das Delta vergleichen. DBCC SQLPERF('sys.dm_os_wait_stats', CLEAR) setzt die Zähler für den ganzen Server zurück und stört Monitoring-Tools; vermeide es. Interpretation: PAGEIOLATCH_* = Plattenlesevorgänge, WRITELOG = Log-Platte, LCK_M_* = Blockierung, CXPACKET/CXCONSUMER = Parallelität, RESOURCE_SEMAPHORE = Memory Grants, SOS_SCHEDULER_YIELD + hoher Signal-Anteil = CPU-Druck. Reine Leseabfrage; sicher.