Écouteur IOCP non productif

Est-ce que quelqu'un sait ce que signifie "Écouteur IOCP non productif"?

Un de nos serveurs SQL vient de contenir un vidage de vérification:

=====================================================================                                            
       BugCheck Dump                                                                                             
=====================================================================                                            

This file is generated by Microsoft SQL Server                                                                   
version 9.00.5292.00                                                                                             
upon detection of fatal unexpected error. Please return this file,                                               
the query or program that produced the bugcheck, the database and                                                
the error log, and any other pertinent information with a Service Request.                                       


Computer type is AT/AT COMPATIBLE.                                                                               
Bios Version is DELL   - 1                                                                                       
Phoenix ROM BIOS PLUS Version 1.10 1.5.2                                                                         
Current time is 23:01:04 09/07/12.                                                                               
48 Unknown CPU 9., 2 Mhz processor (s).                                                                          
Windows NT 6.1 Build 7601 CSD Service Pack 1.                                                                    

Memory                               
MemoryLoad = 81%                     
Total Physical = 524278 MB           
Available Physical = 97549 MB        
Total Page File = 524276 MB          
Available Page File = 94472 MB       
Total Virtual = 8388607 MB           
Available Virtual = 7846765 MB       
**Dump thread - spid = 0, PSS = 0x0000000000000000, EC = 0x0000000000000000                                      
***Stack Dump being sent to C:\Program Files\Microsoft SQL Server\MSSQL.1\MSSQL\LOG\SQLDump0008.txt              
* *******************************************************************************                                
*                                                                                                                
* BEGIN STACK DUMP:                                                                                              
*   09/07/12 23:01:04 spid 0                                                                                     
*                                                                                                                
* Non-yielding IOCP Listener                                                                                     
*                                                                                                                
* *******************************************************************************             </pre>                   


SQLDump0008.log contains:

<pre>
 No user action is required.
2012-09-07 18:30:11.28 spid782     Recovery of any in-doubt distributed transactions involving Microsoft Distributed Transaction Coordinator (MS DTC) has completed. This is an informational message only. No user action is required.
2012-09-07 20:58:54.53 spid196     The alert for 'average delay' has been raised. The current value of '509' surpasses the threshold '100'.
2012-09-07 20:59:24.74 spid477     The alert for 'average delay' has been raised. The current value of '299' surpasses the threshold '100'.
2012-09-07 21:44:06.53 spid23s     Database mirroring is inactive for database 'ToDoLists'. This is an informational message only. No user action is required.
2012-09-07 21:44:06.59 spid456     The alert for 'average delay' has been raised. The current value of '518' surpasses the threshold '100'.
2012-09-07 21:44:57.98 spid425     Error: 18056, Severity: 20, State: 27.
2012-09-07 21:44:57.98 spid425     The client was unable to reuse a session with SPID 425, which had been reset for connection pooling. This error may have been caused by an earlier operation failing. Check the error logs for failed operations immediately before this error message.
2012-09-07 21:44:57.98 spid808     Error: 18056, Severity: 20, State: 27.
2012-09-07 21:44:57.98 spid808     The client was unable to reuse a session with SPID 808, which had been reset for connection pooling. This error may have been caused by an earlier operation failing. Check the error logs for failed operations immediately before this error message.
2012-09-07 21:44:58.01 spid155     Error: 18056, Severity: 20, State: 27.
2012-09-07 21:44:58.01 spid155     The client was unable to reuse a session with SPID 155, which had been reset for connection pooling. This error may have been caused by an earlier operation failing. Check the error logs for failed operations immediately before this error message.
2012-09-07 21:44:58.03 spid486     Task (Worker 0x00000001B93B21C0) was forced to yield 2 times: 
2012-09-07 21:44:58.04 spid65s     Database mirroring is inactive for database 'Tracking'. This is an informational message only. No user action is required.
2012-09-07 21:44:58.06 spid486     Task (Worker 0x0000000CB9B341C0) was forced to yield 8 times: 
2012-09-07 21:44:58.09 spid486     Task (Worker 0x0000000655A9E1C0) was forced to yield 3 times: 
2012-09-07 21:44:58.10 spid486     Task (Worker 0x00000006C03BE1C0) was forced to yield 8 times: 
2012-09-07 21:44:58.11 spid65s     Error: 1404, Severity: 16, State: 6.
2012-09-07 21:44:58.11 spid65s     The command failed because the database mirror is busy. Reissue the command later.
2012-09-07 21:44:58.11 spid486     Task (Worker 0x0000000C819D01C0) was forced to yield 2 times: 
2012-09-07 21:44:58.49 spid140     The alert for 'average delay' has been raised. The current value of '191' surpasses the threshold '100'.
2012-09-07 21:45:00.66 spid46s     SQL Server has encountered 6 occurrence(s) of cachestore flush for the 'Object Plans' cachestore (part of plan cache) due to some database maintenance or reconfigure operations.
2012-09-07 21:45:17.25 spid83s     SQL Server has encountered 6 occurrence(s) of cachestore flush for the 'SQL Plans' cachestore (part of plan cache) due to some database maintenance or reconfigure operations.
2012-09-07 21:45:17.25 spid54s     SQL Server has encountered 6 occurrence(s) of cachestore flush for the 'Bound Trees' cachestore (part of plan cache) due to some database maintenance or reconfigure operations.
2012-09-07 21:45:17.28 spid45s     The mirrored database "Tracking" is changing roles from "PRINCIPAL" to "MIRROR" due to Role Syncronization.
2012-09-07 21:45:17.61 spid46s     Bypassing recovery for database 'Tracking' because it is marked as a mirror database, which cannot be recovered. This is an informational message only. No user action is required.
2012-09-07 21:45:29.21 spid45s     Database mirroring is active with database 'Tracking' as the mirror copy. This is an informational message only. No user action is required.
2012-09-07 21:50:56.94 spid196s    SQL Server has encountered 5 occurrence(s) of cachestore flush for the 'Object Plans' cachestore (part of plan cache) due to some database maintenance or reconfigure operations.
2012-09-07 21:50:57.14 spid196s    SQL Server has encountered 5 occurrence(s) of cachestore flush for the 'SQL Plans' cachestore (part of plan cache) due to some database maintenance or reconfigure operations.
2012-09-07 21:50:57.14 spid196s    SQL Server has encountered 5 occurrence(s) of cachestore flush for the 'Bound Trees' cachestore (part of plan cache) due to some database maintenance or reconfigure operations.
2012-09-07 23:00:09.42 spid438     Error: 18056, Severity: 20, State: 27.
2012-09-07 23:00:09.42 spid438     The client was unable to reuse a session with SPID 438, which had been reset for connection pooling. This error may have been caused by an earlier operation failing. Check the error logs for failed operations immediately before this error message.
2012-09-07 23:01:04.26 Server      Using 'dbghelp.dll' version '4.0.5'
2012-09-07 23:01:04.29 Server      **Dump thread - spid = 0, PSS = 0x0000000000000000, EC = 0x0000000000000000
2012-09-07 23:01:04.29 Server      ***Stack Dump being sent to C:\Program Files\Microsoft SQL Server\MSSQL.1\MSSQL\LOG\SQLDump0008.txt
2012-09-07 23:01:04.29 Server      * *******************************************************************************
2012-09-07 23:01:04.29 Server      *
2012-09-07 23:01:04.29 Server      * BEGIN STACK DUMP:
2012-09-07 23:01:04.29 Server      *   09/07/12 23:01:04 spid 0
2012-09-07 23:01:04.29 Server      *
2012-09-07 23:01:04.29 Server      * Non-yielding IOCP Listener
2012-09-07 23:01:04.29 Server      *
2012-09-07 23:01:04.29 Server      * *******************************************************************************
2012-09-07 23:01:04.29 Server      * -------------------------------------------------------------------------------
2012-09-07 23:01:04.29 Server      * Short Stack Dump
2012-09-07 23:01:04.33 spid73      The alert for 'average delay' has been raised. The current value of '304' surpasses the threshold '100'.
2012-09-07 23:01:04.34 Server      Stack Signature for the dump is 0x00000000000002E8

Le message d'alerte de «délai moyen» se rapporte à la mise en miroir de la base de données et est une alerte déclenchée lorsque le temps nécessaire pour valider les transactions dépasse le temps spécifié. La is_event_loggedcolonne est 0 pour l'alerte «délai moyen».

Vous pouvez voir la sortie à partir sys.configurationsde Qu'est-ce qui peut provoquer un délai d'attente de la mise en miroir puis un basculement? .

IOCP est un port d'achèvement d'E / S. Un écouteur IOCP sans rendement signifie que le thread qui gère les routines d'achèvement des E / S a pris un temps (relativement) long à faire quelque chose, et SQLOS bien qu'il puisse être bloqué / suspendu / peu importe.

Le serveur SQL fait beaucoup d'E / S ASYNC. La façon dont cela fonctionne est quand il soumet la demande d'E / S au système d'exploitation, il dit "Faites cette E / S de manière asynchrone. Voici un pointeur de fonction à appeler quand c'est fait."

La fonction qui est appelée est l'écouteur d'achèvement d'E / S.

Considérons une page lue sur le disque. Un thread exécutant une sélection doit lire une page qui n'est pas en mémoire. Il prend un PAGEIOLATCH, envoie le IO asynchrone aux fenêtres pour lire la page et se met en veille.

Lorsque le système d'exploitation termine l'IO, il appelle la fonction IOCP qui signale l'IO comme «terminé». Peu de temps après, un thread sql termine son quantum de 4 ms et vérifie les entrées / sorties à gérer. Il le marque comme terminé et signale au thread émetteur de se réveiller. Le thread SELECT est planifié, libère le PAGEIOLATCH et la vie est bonne.

Maintenant, la quantité de travail que l'IOCP fait varie selon le type d'E / S impliquées. Je crois qu'avec DB Mirroring, cela fait plus de travail que s'il ne faisait que lire une page dans le pool de tampons.

Si vous êtes un programmeur travaillant sur un serveur SQL et que vous souhaitez optimiser le code de mise en miroir de base de données, vous pourriez être tenté de mettre plus de travail dans le chemin du code de mise en miroir IOCP par rapport au chemin du code du thread système SQLOS.

Ou peut-être que l'IOCP doit copier les données dans un tampon de mise en miroir de taille fixe, et il reste en boucle jusqu'à ce qu'il soit terminé.

Ou peut-être que <> se produit et que la fonction IOCP semble "bloquée".

Je ne m'inquiéterais pas de cela si cela se produisait lors d'un basculement et que beaucoup d'activité LOG se produisait. Si cela se produit régulièrement, cela peut nécessiter une enquête plus approfondie.