Monday, 2025-09-29

*** mhen_ is now known as mhen01:37
*** mhen_ is now known as mhen07:41
opendevreviewBalazs Gibizer proposed openstack/nova master: Rally job for eventlet-removal  https://review.opendev.org/c/openstack/nova/+/96013008:20
opendevreviewBalazs Gibizer proposed openstack/nova master: Rally job for eventlet-removal  https://review.opendev.org/c/openstack/nova/+/96013008:46
opendevreviewBalazs Gibizer proposed openstack/nova master: Rally job for eventlet-removal  https://review.opendev.org/c/openstack/nova/+/96013009:43
opendevreviewJonas Schäfer proposed openstack/nova master: Preserve vTPM state between power off and power on  https://review.opendev.org/c/openstack/nova/+/95565710:00
opendevreviewKamil Sambor proposed openstack/nova master: Switch nova-conductor to use global executor  https://review.opendev.org/c/openstack/nova/+/96247810:11
opendevreviewJonas Schäfer proposed openstack/nova master: Preserve vTPM state between power off and power on  https://review.opendev.org/c/openstack/nova/+/95565711:12
*** tridde is now known as Trident11:26
opendevreviewBalazs Gibizer proposed openstack/nova master: Rally job for eventlet-removal  https://review.opendev.org/c/openstack/nova/+/96013011:28
gibidansmith: fyi we can catch threading slowdowns now with rally https://review.opendev.org/c/openstack/nova/+/960130/30#message-1f12ccedc335aa0e59d469a478012a41d76fd67c12:14
gibi(this was a simulated slowdown by injecting sleep to nova)12:14
opendevreviewKamil Sambor proposed openstack/nova master: Switch nova-conductor to use global executor  https://review.opendev.org/c/openstack/nova/+/96247812:40
opendevreviewMerged openstack/nova stable/2025.1: Regression test for ImagePropsWeigher due to missing context  https://review.opendev.org/c/openstack/nova/+/96180513:09
dansmithgibi: nice.. hopefully that will be stable enough to be useful.. even within a single worker performance has varied a lot in the past (due to noisy neighbors) but even if we get a smoke signal that will be good13:58
gibiif the performance of the worker fluctuates in a ~10 mintues window then yeah we will get false postive signals14:04
gibithe window size depends one length of our rally scenarios14:05
sean-k-mooney-ptowhile we could have noise neighber effects i think its a lease worth trying this for a while and see if it becomes a problem14:17
sean-k-mooney-ptogibi: presumable the rally job wont be voting at least not in the short term14:18
gibiwon't be voting for a while 14:18
sean-k-mooney-ptoso if it a perodic or its non voting14:18
sean-k-mooney-ptoi think this ok for now14:18
sean-k-mooney-ptogibi: do you think it need to run in check longterm or would you jsut rung this in weekly perodic?14:19
gibiright now it is a cheap job so I would run it non-voting. But if we add more scenarios (I planning to ) then it might be expensive enough to run it only periodically14:23
gibior if we move it to a label with many subnodes14:23
sean-k-mooney-ptoack14:24
sean-k-mooney-ptothis is the kind fo job where it would be nice if there as a persitent mysql db where we coudl log some data across runs14:24
sean-k-mooney-ptoi.e. to track over all trends across time14:25
sean-k-mooney-ptobut ya i was thining the same start we non voting and hten move to perodic + experimental if its becomes expensive14:26
sean-k-mooney-ptohavign it in experminetal so you can spot check specific changes is ueful if we go the perodic route14:27
*** sean-k-mooney-pto is now known as sean-k-mooney14:28
sean-k-mooneygibi: the last day before i went on pto i spend a little time tryign to see if we could get any better infro form atop or the logs on overall load. i made some progress but the binary format and way that tool works is a littel painful14:29
gibisean-k-mooney: I actually noticed that when a service is stopped systemd already nicely logs the memory swap high watermark and the cpu time of the service14:40
gibiso I went with that14:40
sean-k-mooneyyes i saw that also14:40
sean-k-mooneyhowever we only have the logs form the second run correct14:40
sean-k-mooneyunless thats also in the journal dump?14:41
gibiwe have journal logs from both runs, and I do an extra service restart now at the end to get the resource usage after the second run as well. Then I pull the data out from the journal into eg https://storage.bhs.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/zuul_opendev_logs_eb1/openstack/eb1c16166a9e42b5be0987f5d8ca0d01/results/resource-usage-devstack%40n-sch.service.txt14:41
sean-k-mooneyah cool14:43
jssfrdansmith (and maybe others following https://review.opendev.org/c/openstack/nova/+/955657 ): I'm going to be AFK starting october 1st until november 25th. Is there anything else I need to do before that to get it over the line?14:46
dansmithjssfr: sorry, just been really busy with other stuff.. I think it mostly looks good to me now but I was just going to do some tests locally that I haven't had a chance to do yet14:47
dansmithbut others may feel more comfortable just looking at it14:47
jssfrrelated question: is it possible to take over another person's gerrit patch?14:48
jssfrI might find a substitute who can continue to look into it while I'm away.14:48
dansmithyep, easy and common14:48
jssfris there anything that needs to be taken into account when doing that? or can another gerrit user just take my patch and run `git review`?14:50
opendevreviewMerged openstack/nova stable/2025.1: Add admin context when filling metadata on ImagePropsWeigher  https://review.opendev.org/c/openstack/nova/+/96180614:50
dansmithno, anyone can git review it right now14:50
jssfrthanks14:50
clarkb`git review -d changenumber` make edits then `git review` to push back14:51
bauzasdansmith: given we have yet another issue with the ImageMetaWeigher https://bugs.launchpad.net/nova/+bug/2125935 I wonder why I can't just target a cell context for fill_metadata this way :16:48
bauzaswith nova_context.target_cell(ctxt, cell_mapping):16:49
bauzas   insts.fill_metadata()16:49
bauzasI guess I need to change the object context itself, right?16:49
dansmithwell, that's targeting a context but you're not using it16:49
dansmithall the instances are from the same cell right? If so, you need to target the context, then pass that to InstanceList.__init__() and then fill metadata16:49
dansmiththe usage is "with target_cell(...) as child_ctx: "16:50
bauzasyeah, given we look at all the instances from the same host, they are on the same context16:52
bauzass/context/cell16:52
bauzasgotcha, I got that16:52
bauzasthanks16:53
dansmithcool, yeah then just need to target the one you give to the InstanceList16:53
bauzas        with nova_context.target_cell(ctxt, cell_mapping) as cell_ctxt:17:00
bauzas            insts = objects.InstanceList(cell_ctxt,17:00
bauzas                objects=host_state.instances.values())17:00
bauzas            insts.fill_metadata()17:00
bauzasthat thing doesn't give me the metadata yet17:00
* bauzas continues17:01
dansmithdo the fake instances have sysmeta set on the object to an empty dict?17:14
bauzasyup17:21
bauzasI'm lost, I need to get some fresh air, ending my long day 17:27
dansmithif they do then the fill will just skip them all, so need to drop those from the object to make it work17:35
opendevreviewmelanie witt proposed openstack/nova master: FUP for vTPM live migration  https://review.opendev.org/c/openstack/nova/+/96230919:47
opendevreviewmelanie witt proposed openstack/nova master: Opt-in to new TPM secret security via resize  https://review.opendev.org/c/openstack/nova/+/96205219:47
opendevreviewmelanie witt proposed openstack/nova master: DNM vtpm tempest  https://review.opendev.org/c/openstack/nova/+/95747719:47

Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!