| opendevreview | Takashi Kajinami proposed openstack/ceilometer master: wip: Restore SNMPInspector https://review.opendev.org/c/openstack/ceilometer/+/958368 | 05:08 |
|---|---|---|
| opendevreview | Merged openstack/ceilometer master: Get more flavor attributes from libvirt metadata (if available) https://review.opendev.org/c/openstack/ceilometer/+/943365 | 07:27 |
| opendevreview | Takashi Kajinami proposed openstack/ceilometer master: wip: Restore SNMPInspector https://review.opendev.org/c/openstack/ceilometer/+/958368 | 08:30 |
| opendevreview | Takashi Kajinami proposed openstack/ceilometer master: wip: Restore SNMPInspector https://review.opendev.org/c/openstack/ceilometer/+/958368 | 08:36 |
| jlarriba | tkajinam thanks for updating the commit message in https://review.opendev.org/c/openstack/ceilometer/+/958368 to reflect the rationale of reintroducing SNMP information, but this still seems strange to me | 08:44 |
| jlarriba | tkajinam ceilometer is not meant to provide hardware metrics anymore for the compute nodes, it just provides measurements for the VMs that are running in each node | 08:46 |
| tkajinam | jlarriba, yeah, though again that's what watcher expects now | 09:13 |
| tkajinam | to measure memory usage of hosts | 09:13 |
| tkajinam | ceilometer provides IPMI metrics so having SNMP metrics too does not really extend its scope too much. I'm just checking the amount of work we may need to restore it | 09:13 |
| tkajinam | a lot of things have been changed in pysnmp recently | 09:14 |
| jlarriba | ceilometer provides ipmi metrics in a different instance, the ceilometer-compute does not provide ipmi metrics | 09:50 |
| jlarriba | i think it extends its scope, yeah. Because why provide only memory, and not cpu information about the node? or disk | 09:51 |
| jlarriba | i think all of that is better done outside of ceilometer, with a generic agent as none of that has anything to do with openstack | 09:52 |
| tkajinam | technically you can run a dedicated instance of polling-agent to pull snmp metrics. | 10:14 |
| tkajinam | It can pull cpu metrics. the wired situation in watcher using only memory metrics based on snmp is caused by incomplete implementation in nova which exposes only cpu metrics | 10:15 |
| tkajinam | I think that nova feature was implemented for some tricky scheduler filter which they don't want to extend anymore so adding memory metrics to nova may not be accepted | 10:15 |
| tkajinam | jlarriba, are you saying that we should even drop ipmi agent ? | 10:16 |
| tkajinam | which is nothing to do with openstack really | 10:16 |
| jlarriba | tkajinam yes, it might be subject for deprecation in the future, depending on how the power monitoring open source space evolves | 10:37 |
| jlarriba | if users can reliably get good power consumption via a different, generic agent, I would vouch for stop maintaining ipmi as it has nothing to do with openstack, really | 10:38 |
| jlarriba | tkajinam watcher could rely on memory metric retrieved by a different, generic monitoring agent that provides those metrics | 10:39 |
| tkajinam | yeah | 10:50 |
| opendevreview | Merged openstack/ceilometer master: Publish image meta from notifications and pollsters (if available) https://review.opendev.org/c/openstack/ceilometer/+/942879 | 11:01 |
| opendevreview | Tobias Urdin proposed openstack/ceilometer master: Handle EPIPE Broken Pipe from Libvirt RPC https://review.opendev.org/c/openstack/ceilometer/+/958039 | 11:36 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!