Wednesday, 2025-08-13

opendevreviewTakashi Kajinami proposed openstack/aetos master: Support client cert/key for TLS connection with Prometheus  https://review.opendev.org/c/openstack/aetos/+/95698301:05
opendevreviewMerged openstack/aetos master: Validate input of [DEFAULT] auth_mode  https://review.opendev.org/c/openstack/aetos/+/95665006:36
opendevreviewCallum Dickinson proposed openstack/ceilometer master: Add vcpus pollster  https://review.opendev.org/c/openstack/ceilometer/+/95721409:38
opendevreviewCallum Dickinson proposed openstack/ceilometer master: Add memory pollster  https://review.opendev.org/c/openstack/ceilometer/+/95721610:09
opendevreviewCallum Dickinson proposed openstack/ceilometer master: Add memory pollster  https://review.opendev.org/c/openstack/ceilometer/+/95721611:15
opendevreviewJaromír Wysoglad proposed openstack/aetos master: Add configuration for project label name  https://review.opendev.org/c/openstack/aetos/+/95722412:20
tkajinamhmm I suspect that current power_state pollster may conflict with resize (or cold migration) which keeps the original domain at source until the operation is confirmed12:30
tkajinamhttps://review.opendev.org/c/openstack/ceilometer/+/95721412:30
tkajinammaybe even live migration can interfere12:38
tkajinamjwysogla, ^^^ you might want to discuss this with Juan12:38
tkajinamThe potential problem I'm pointing is that current power_state pollster publishes metrics regardless of domain status. During resize or migration we expect two domains defined in source and dest so we may get two samples from both these two nodes12:40
opendevreviewTakashi Kajinami proposed openstack/ceilometer master: Disable power.state meter  https://review.opendev.org/c/openstack/ceilometer/+/95723112:50
jwysoglaI'll try, but I'll need to do some studying as I have 0 knowledge on the topic. Not sure if Juan is much better.12:52
jwysoglaAlso I won't see Juan until September because of our overlapping PTOs, but I'll try to point him to this conversation.12:53
tkajinamok12:53
tkajinamjwysogla, if you can leave this link (maybe in internal chat or something) for him then that might be helpful https://bugs.launchpad.net/ceilometer/+bug/212056012:58
mrungethis is tricky. I wonder if there is a better solution to the issue. 12:59
tkajinamthe only way I can think of atm is to publish only specific state such as running and probably crashed.13:05
tkajinamthough I'm not sure that does not break the original motivation to add that meter13:05
tkajinamhttps://libvirt.org/html/libvirt-libvirt-domain.html#virDomainState13:06
tkajinampublishing shutoff is definitely tricky. also afair domain enters paused state during migration so we properly exclude that case or avoid publishing that state13:07
tkajinamthe others might be ok13:07
tkajinamfor total cpu/ram we can avoid this problem by publishing these meters only when the domain is active but power state requires the other state by its nature... so again it's tricky 13:10
jwysoglaok, I think I understand the issue. I agree with disabling it by default until we figure out how to have the metric in a not confusing way in the scenarios you pointed out. Regarding possible solutions and original motivation, I think it'd be good to wait for Juan as he knows more about the metric.13:19
tkajinamyeah13:31
opendevreviewMerged openstack/python-observabilityclient master: Fix PrometheusAPIClient set_client_cert  https://review.opendev.org/c/openstack/python-observabilityclient/+/95718513:56
opendevreviewMerged openstack/ceilometer master: Disable power.state meter  https://review.opendev.org/c/openstack/ceilometer/+/95723118:51
opendevreviewJaromír Wysoglad proposed openstack/aodh master: Rbac restrictions to prometheus alarms  https://review.opendev.org/c/openstack/aodh/+/95727919:23
opendevreviewJaromír Wysoglad proposed openstack/aodh master: Rbac restrictions to prometheus alarms  https://review.opendev.org/c/openstack/aodh/+/95727919:24

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