December 29, 2016 at 7:34 am
Hoping somebody can help. We have a data flow task in an SSIS package that is called from a SQL job. The task takes data from a view on one server (view takes less thank a second to run from my local machine) to a table on another server. This task is part of a package that usually takes 2 to three minutes to run. Twice now the package has run for hours, getting hung on this data flow task. What could possibly be causing this? There is no deadlock, no process that this process is waiting for, etc. Any help is appreciated. This is production so I cannot run the task locally. Thanks.
December 29, 2016 at 10:55 pm
Possibly out of date stats on the source? Maybe parameter sniffing? Maybe someone had a long winded transaction on one of the underlying tables of the view?
--Jeff Moden
Change is inevitable... Change for the better is not.
January 3, 2017 at 4:33 pm
This is a case where the execution logs will be very useful. Are you executing from the SSIS catalog? If so, take a peek at the logs for that execution.
Tim Mitchell, Microsoft Data Platform MVP
Data Warehouse and ETL Consultant
TimMitchell.net | @Tim_Mitchell | Tyleris.com
ETL Best Practices
January 3, 2017 at 7:37 pm
Re-reading the original post, I'd also check and see what was going on in the form of blocking during those times on the remote server.
--Jeff Moden
Change is inevitable... Change for the better is not.
January 4, 2017 at 5:40 am
What we found is that one of the underlying tables used in one of the views the query uses had 20 million + rows. Rebuilding the indexes prior to running the query caused the error to go away.
January 4, 2017 at 8:00 am
karen.ferrara (1/4/2017)
What we found is that one of the underlying tables used in one of the views the query uses had 20 million + rows. Rebuilding the indexes prior to running the query caused the error to go away.
Don't forget that Rebuilding indexes causes statistics to update. There's a real high probability that updating statistics on the table would have done it without the ardor of rebuilding indexes on a large table. Try rebuilding stats first the next time and see what happens.
--Jeff Moden
Change is inevitable... Change for the better is not.
January 4, 2017 at 8:59 am
Thanks for the advice!
January 9, 2017 at 6:07 pm
If you are using a OLE DB destination adapter in the data flow then you can also tinker with the "Rows per patch:" or the "Maximum insert commit size:" options to see if you can get a little less overall run time if you are moving tons of data across the network.
----------------------------------------------------
Viewing 8 posts - 1 through 7 (of 7 total)
You must be logged in to reply to this topic. Login to reply