Skip to main content

The Purge SQL Agent Job for MDW takes a long time to complete

I use the dbWarden alerts to inform me if a SQL job is taking longer to complete than normal and I got one this morning:
 
Click to enlarge

I noticed by looking at the history this purge job was gradually taking longer and longer to complete each day since I installed it again (see mylast post on this):

RunDate
RunTime
Duration
ExecutionStatus
JobName
04/10/2013
02:00:01
00:00:45
Succeded
mdw_purge_data_[MDW]
05/10/2013
02:00:00
00:13:27
Succeded
mdw_purge_data_[MDW]
06/10/2013
02:00:00
00:17:03
Succeded
mdw_purge_data_[MDW]
07/10/2013
02:00:01
00:30:58
Succeded
mdw_purge_data_[MDW]
08/10/2013
02:00:01
01:04:25
Succeded
mdw_purge_data_[MDW]
09/10/2013
02:00:01
01:41:39
Succeded
mdw_purge_data_[MDW]
10/10/2013
02:00:01
02:35:17
Succeded
mdw_purge_data_[MDW]
11/10/2013
02:00:00
03:34:21
Succeded
mdw_purge_data_[MDW]
12/10/2013
02:00:01
04:12:51
Succeded
mdw_purge_data_[MDW]
13/10/2013
02:00:01
04:54:04
Succeded
mdw_purge_data_[MDW]
14/10/2013
02:00:00
05:38:58
Succeded
mdw_purge_data_[MDW]
15/10/2013
02:00:01
07:07:13
Succeded
mdw_purge_data_[MDW]
16/10/2013
02:00:00
08:21:11
Cancelled
mdw_purge_data_[MDW]

The last job I cancelled so I could figure out what was going on. A quick Google took me to this page which recommended adding a couple of indexes. Before I did so I took a look at the size of both the tables mentioned in the article:

sp_spaceused  '[snapshots].[query_stats]'

name
rows
reserved
data
index_size
unused
query_stats
5,702,294
599072 KB
587616 KB
11320 KB
136 KB

sp_spaceused '[snapshots].[notable_query_text]'

name
rows
reserved
data
index_size
unused
notable_query_text
26,383
34872 KB
33616 KB
272 KB
984 KB

...And the execution plan of the following query before I made any changes:

SELECT qt.[sql_handle]
INTO #tmp_notable_query_text_test
FROM snapshots.notable_query_text AS qt
WHERE NOT EXISTS (
   SELECT snapshot_id
   FROM snapshots.query_stats AS qs
   WHERE qs.[sql_handle] = qt.[sql_handle])

Execution Plan Before

Click to enlarge

You can clearly see that the scan of CIDX_query_stats costs by far the most of any operation in the query.
So I applied the following indexes as recommended in the article and looked at the execution plan of the query again:

USE [MDW]
GO

SET ANSI_PADDING ON
GO

CREATE NONCLUSTERED INDEX [IX_sql_handle]
    ON [snapshots].[query_stats]
          (
          [sql_handle] ASC
          )
       GO

CREATE NONCLUSTERED INDEX [IX_sql_handle]
    ON [snapshots].[notable_query_text]
          (
          [sql_handle] ASC
          )  
GO




Execution Plan After

Click to enlarge

Now various changes have happened to the query play including a Nested Loop operator replacing the Hash Match and an Index Seek replacing the Index Scan.

I ran the job manually and it completed much faster :

RunDate
RunTime
Duration
ExecutionStatus
JobName
16/10/2013
10:41:56
00:05:32
Succeded
mdw_purge_data_[MDW]

But much of the hard work would have already been done from the previous night’s schedule, so I will need to wait until tonight to see if the time had be reduced.

**Update**

It looks like the queries have done their job. The  duration has greatly improved:

RunDate
RunTime
Duration
ExecutionStatus
JobName
17/10/2013
02:00:00
00:21:00
Succeded
mdw_purge_data_[MDW]


A word of warning

Even though the MDW has been active for only two weeks and I am monitoring only one medium-busy server, nearly 14GB of data has been generated:

Click to enlarge




Comments

  1. Very useful. I have many customers that simply LOVE the Data Collector interface. Unfortunately, DC is full of problems (random errors in collection jobs, database growing too much, report errors).

    ReplyDelete

Post a Comment

Popular posts from this blog

How to create a custom Windows Event Log view and email trigger

The filtering on Windows event logs can be slow, clunky and although you can do it on fields like event ID, it seems that many event IDs are shared amongst many different errors – the event ID may match but the body of the error (therefore the actual error) may be completely unrelated. Fortunately, it is possible to filter on the contents of the body of the error message but it requires creating a custom XML query. Also, it would be handy to send out a notification email when this event gets logged. Read on to find out how to work this magic…. This example is looking for a  Warning  event  1309  for  ASP.NET 4.0.30319.0  on a web server. If you were to just filter the log on the criteria above today it would return 435 results because it is a fairly general error ID. If I filter it using XML for SqlException (what I’m really interested in) only 5 results are returned. So the first step is go to the Application Log and choose  Create Custom View…  Select the  XML  tab, check  Edit

How to configure the SSAS service to use a Domain Account

NB Updating SPNs in AD is not for the faint hearted plus I got inconsistent results from different servers. Do so at your own risk! If you need the SSAS account on a SQL Server to use a domain account rather than the local “virtual” account “NT Service\MSSQLServerOLAPService”. You may think you just give the account login permissions to the server, perhaps give it sysadmin SQL permissions too. However, if you try and connect to SSAS  remotely  you may get this error: Authentication failed. (Microsoft.AnalysisService.AdomdClient) The target principal name is incorrect (Microsoft.AnalysisService.AdomdClient) From Microsoft: “A Service Principle Name (SPN) uniquely identifies a service instance in an Active Directory domain when Kerberos is used to mutually authenticate client and service identities. An SPN is associated with the logon account under which the service instance runs. For client applications connecting to Analysis Services via Kerberos authentication, th

Fun and games with the Management Data Warehouse (MDW and Data Collectors)

The SQL Server Management Data Warehouse (when you first come across it) seems to promise so much if the verbiage from Microsoft and some other websites is to to believed. But when you install it you may find that it is not as useful as it could be. This is a shame but we are currently only on v2 of the product with SQL 2012 so one hopes it will improve in subsequent versions. However, it probably is worth playing with if you have never used it before - at least you can show your boss some reports on general server health when he asks for it and you have nothing else in place. There is one big problem with it though if you decide that you don't want to use it any more, uninstalling it is not supported! Mad, I know. But as usual some very helpful people in the community have worked out, what seems to me, a pretty safe way of doing it. I had a problem with my MDW. The data collector jobs were causing a lot of deadlocking on some production servers and impacting performance. I