October 10, 2019 at 8:29 pm
Everyone has seen this, new DBA at new company has 3rd party EMR product that we have AG setup for read only node. Week one on board users screaming. performance for RO node is awful, 2017 Std Edition 12 procs 82 gig of memory. Ran blitzcache and wow 3 sql statements that require over 250 million logical reads (yes that number is correct). CPU gets buried but PLE is not bad but with these 3 statements which have all in common. And yes with the 3rd party not easy to change, other than jumping out window anyone have any great ideas?
Compilation Timeout, Forced Parameterization, Parallel, Parameter Sniffing, Long Running Query, Plan created last 4hrs, Row estimate mismatch, >500mb spills
October 11, 2019 at 10:31 am
Everyone has seen this, new DBA at new company has 3rd party EMR product that we have AG setup for read only node. Week one on board users screaming. performance for RO node is awful, 2017 Std Edition 12 procs 82 gig of memory. Ran blitzcache and wow 3 sql statements that require over 250 million logical reads (yes that number is correct). CPU gets buried but PLE is not bad but with these 3 statements which have all in common. And yes with the 3rd party not easy to change, other than jumping out window anyone have any great ideas?
Compilation Timeout, Forced Parameterization, Parallel, Parameter Sniffing, Long Running Query, Plan created last 4hrs, Row estimate mismatch, >500mb spills
Capture the code and check the execution plan or post here.
Muthukkumaran Kaliyamoorthy
https://www.sqlserverblogforum.com/
October 11, 2019 at 5:59 pm
Looks multiple problems, query is bad but ran on test system I/o issues in report server. ran blitzio and got some scary results with stall time over 40ms
October 13, 2019 at 4:44 am
Sounds like a stupid question but what condition are the index and column statistics in?
--Jeff Moden
Change is inevitable... Change for the better is not.
October 14, 2019 at 11:14 am
This was removed by the editor as SPAM
October 14, 2019 at 12:47 pm
stats and indexes good, looks like this may be a storage issue
October 14, 2019 at 1:49 pm
We went through that same issue because they went cheap on the RO node storage. They also forgot to do things like change the settings for parallelism, etc.
--Jeff Moden
Change is inevitable... Change for the better is not.
Viewing 7 posts - 1 through 6 (of 6 total)
You must be logged in to reply to this topic. Login to reply