These are chat archives for ManageIQ/manageiq/performance

24th
Feb 2016
Alex Krzos
@akrzos
Feb 24 2016 00:24
@kbrock I looked for catalogs under services and my db doesn't have any?
I'll send you my ip addresses of my appliances with the databases
maybe i'm in the wrong account
(admin)
also by 8 slow use cases I read 7 things in the spreadsheet so I think I'm not aware of all "slow" use cases aside from the vm_infra/explorer
Keenan Brock
@kbrock
Feb 24 2016 01:23
@akrzos 1) I don't have any catalogs either. 2) I can't count - you are right with 7
Jared Deubel
@jdeubel
Feb 24 2016 15:15
@jrafanie @akrzos Do you guys have any insight into how to tune the refresh_threshold in the advanced config?
Where did we get that refresh_threshold:100 ... was that just a good number that we picked or did we have a reason to pick that?
Joe Rafaniello
@jrafanie
Feb 24 2016 15:16
I don't remember what that setting does
Jared Deubel
@jdeubel
Feb 24 2016 15:18
I would assume that each environment is different and it would have to be changed as I would think its faster to run that at 200 then kicking off a full refresh
Have we done any testing on this?
@Fryguy do you have any insight on this ^
Jason Frey
@Fryguy
Feb 24 2016 15:26
you can see how long a single targeted refresh takes
and then see how long a full refresh takes
then just divide
the number 100 was picked a little randomly but also a little based on various customer logs we had at the time
it definitely hasn't been updated, and we would be up for a better base recommendation
I thought @akrzos looked into that number a long time ago, but I can't remember
Dennis Metzger
@dmetzger57
Feb 24 2016 15:30
added that to the perf todo list
Jared Deubel
@jdeubel
Feb 24 2016 15:34
@dmetzger57 @Fryguy let me take a look
Alex Krzos
@akrzos
Feb 24 2016 15:44
@Fryguy , @jdeubel I do have some numbers regarding that
also depends on environment size
@dmetzger57 for bug 1296192, I'll bump the miq schedule worker and re-run some tests
Jared Deubel
@jdeubel
Feb 24 2016 15:47
@akrzos I would think that plays a huge role. So would I just see what the targeted refresh times are and then * 100 to see if its less then the full refresh time? Or is that not accurate
Dennis Metzger
@dmetzger57
Feb 24 2016 15:48
@akrzos thank you in advance regarding 1296192
Alex Krzos
@akrzos
Feb 24 2016 15:50
@jdeubel Let me grab you the latest numbers
Jared Deubel
@jdeubel
Feb 24 2016 15:50
Ok thanks
Alex Krzos
@akrzos
Feb 24 2016 15:50
@dmetzger57 also for the ui perf issue, I don't seem to have any service catalog items to order
Jared Deubel
@jdeubel
Feb 24 2016 15:53
@akrzos @dmetzger57 The UI perf issue...we just got a new DB but its not showing any services in there .. which is there primary compliant
Alex Krzos
@akrzos
Feb 24 2016 16:01
Whats that link to the table markdown?
Found it
provider-size Delta Refresh 1 25 50 75 100
vmware-small 6.1108 1.5395 5.348 11.8338 13.2047 17.5967
vmware-medium 44.7753 5.8776 12.4942 17.6182 26.3721 22.5993
vmware-large 147.1746 18.4882 21.7599 24.2993 34.5319 37.3083
vmware-xlarge 702.5532 57.4833 67.3931 75.6916 72.8032 78.3191
@jdeubel ^
5.5.2.1
so for xlarge providers scaling that number can result in faster refreshes
small is only a 100vms so I stopped collecting data after that
can't refresh 200 vms when there is only 100 :smile:
I could try and run some number on higher vm counts
on the providers that have it
Alex Krzos
@akrzos
Feb 24 2016 16:23
@dmetzger57 I am setting up some tests to run for the Schedule worker now
I'll test with 5.5.2.4, is that good?
Dennis Metzger
@dmetzger57
Feb 24 2016 16:24
@akrzos that’ll work
Alex Krzos
@akrzos
Feb 24 2016 16:46
@dmetzger57 Ok I got some tests running now, just gotta confirm I got everything setup correctly but I'm bumping the threshold to 1GiB so it shouldn't recycle during each 2 hour workload
Jared Deubel
@jdeubel
Feb 24 2016 18:14
@akrzos Thanks for that information.
@akrzos is the deltra refresh a full provider refresh time and then 1 25, etc.. are the refresh_threshold count?
Alex Krzos
@akrzos
Feb 24 2016 20:45
@jdeubel Initial refreshes aren't shown on that table because that would include all of the initial populating of the database that takes more time
subsequent "Delta" refreshes are a refresh on a provider that has already been inventoried. IMO it is more accurate to compare those numbers than the initial refresh to the number of vms refreshed at each threshold
Alex Krzos
@akrzos
Feb 24 2016 20:52
@dmetzger57 so my tests are running, first test completed over two hours and shows the MiqScheduleWorker topping out at 253MiB
MiqScheduleWorker-34581.png
Dennis Metzger
@dmetzger57
Feb 24 2016 20:57
@akrzos sneaking 3MiB over the default threshold
Alex Krzos
@akrzos
Feb 24 2016 21:07
@dmetzger57 Yup
That was a provisioning workload on VMware