| *** ysandeep|out is now known as ysandeep | 01:19 | |
| *** ysandeep is now known as ysandeep|afk | 06:08 | |
| anskiy | mgariepy: in my test lab I've had to increase keystone_httpd_mpm_max_spare_threads to 2, others look fine with 1 thread/process | 07:35 |
|---|---|---|
| anskiy | question! I'm getting a little bit dissapointed with gnocchi (can tell exactly why, if you want), so I was looking to give monasca a try, but I see that role for it was archived in 2019. Why is it so or what happened to it? | 07:38 |
| *** ysandeep|afk is now known as ysandeep|lunch | 08:00 | |
| noonedeadpunk | anskiy: I'm interestred why :) | 08:31 |
| noonedeadpunk | Also I would say that monasca and gnocchi/ceilometer are for quite different purpose | 08:32 |
| noonedeadpunk | regarding monasca - we archived it due to lack of interest and tons of external dependencies we didn't want to maintain on our own. There was patch to revive it, but maintainer never came back to revise them and fix | 08:33 |
| opendevreview | wangjiaqi proposed openstack/openstack-ansible-os_mistral master: Use TOX_CONSTRAINTS_FILE https://review.opendev.org/c/openstack/openstack-ansible-os_mistral/+/851173 | 09:07 |
| anskiy | noonedeadpunk: I'm trying to get CPU usage as percent on a scale from 1 to 100 (for any amount for vcpus). It goes like this: `gnocchi aggregates '(/ (aggregate mean (/ (metric cpu rate:mean) (metric vcpus mean))) 60000000000)'` (60000000000 is poll interval times nanoseconds -- unit). | 09:23 |
| anskiy | vcpus mean is collected via event from instance_usage_audit and minimum resolution for it is 1 hour. `cpu` on the other hand is collected and aggregated however you want (archive-policy + poll interval). | 09:25 |
| noonedeadpunk | I'm pretty sure I did that waaay easier | 09:25 |
| noonedeadpunk | though I don't think I used mean for that.... | 09:25 |
| anskiy | so, when you have two series, and one of them lacks points, on the resulting aggregate you'll get points only on intersection | 09:26 |
| noonedeadpunk | It's quite a while when I touched gnocchi, but it was working perfectly for billing purposes and we had nice graphs in grafana. Now it's working for 5 years I guess... | 09:26 |
| anskiy | you can `fill`, but for it you could only use static values | 09:26 |
| noonedeadpunk | I don't have this deployment handy though to say how we did that | 09:27 |
| anskiy | well, earlier, you could calculate that percentage on transform in sink, but now that ability is gone. | 09:27 |
| anskiy | so, either I will retrieve both series, fill the shorter one with missing values for points and aggregate them by myself | 09:29 |
| anskiy | or, I would patch ceilometer to add new metric, like this: https://stackoverflow.com/questions/66450164/rocky-ceilometer-memory-mteric-add-memory-usage-data-collection-granularity-is-n/66502236#66502236. | 09:30 |
| anskiy | or, I could try monasca :) | 09:30 |
| anskiy | for which ceilometer has a official publisher now | 09:32 |
| noonedeadpunk | Yes, I totally was transforming smth in sink | 09:44 |
| noonedeadpunk | anskiy: well, you can also publish to prometheus> | 09:44 |
| noonedeadpunk | Because monasca is obscure imo... | 09:44 |
| noonedeadpunk | as they've dropped rabbitmq in favor of Kafka, you also need Apache Storm and smth else for it to work | 09:46 |
| noonedeadpunk | Influx? | 09:46 |
| noonedeadpunk | So bunch of Java crap | 09:46 |
| anskiy | influxdb can't clusterize in OSS -_- | 09:47 |
| noonedeadpunk | Well, you can get Cassandra instead :D | 09:47 |
| noonedeadpunk | But with prometheus you actually don't need ceilometer to be fair, as you can jsut get same stuff with libvirt exporter | 09:48 |
| noonedeadpunk | there's also openstack exporter available as well... | 09:48 |
| anskiy | or elastic, yeah. Regarding that "java crap": there is still some (~45%) Java in monasca-api's github repo | 09:48 |
| anskiy | noonedeadpunk: I think you can't get away with only libvirt, as you need to enrich that data with, at least server UUID | 09:49 |
| anskiy | else you'd get only instance IDs | 09:50 |
| noonedeadpunk | and elastic as of today has quite interesting licensing per say. So basically we were relying on 3rd party roles for all that crap, and obviously they needed revision and some interested parties and we had non | 09:50 |
| noonedeadpunk | *none | 09:50 |
| noonedeadpunk | anskiy: well, yes, that's why ceilometer is kind of neat. But I think it could use prometheus still... Not sure if it's working though as I never tried that | 09:51 |
| anskiy | I see. So if you would ever need to deploy two openstacks for the price of one, I'll go with monasca | 09:51 |
| noonedeadpunk | but looking at coude I have quite some doubts https://opendev.org/openstack/ceilometer/src/branch/master/ceilometer/publisher/prometheus.py | 09:52 |
| anskiy | oh god | 09:52 |
| noonedeadpunk | So for monasca - we have this patch https://review.opendev.org/c/openstack/openstack-ansible-os_monasca/+/796616 What it totally misses - list a requirements of collections/roles that would deploy infra for it | 09:53 |
| noonedeadpunk | So I'm quite fine in reviving it given there's some interest, but I won't be able to put too much time/effort into it | 09:54 |
| anskiy | oh, it's not that old | 09:55 |
| noonedeadpunk | I think it was for Victoria | 09:55 |
| noonedeadpunk | so basically it's matter of finding roles, that would deploy kafka, influx in HA manner and adding some config for them to make thing work out of the box | 09:56 |
| anskiy | noonedeadpunk: I'm unsure of how to proceed, have to think about it. There is another option, tho: patch Gnocchi to get `fill(previousValue)` behaviour, but numpy and matrices -_- | 09:57 |
| anskiy | noonedeadpunk: for Influx it would be some ugly switchover thing, as I've mentioned earlier -- no clustering support in OSS | 09:57 |
| noonedeadpunk | Well. I'm jsut saying status of monasca in OSA :) | 09:58 |
| anskiy | and I really appreciate that! Thank you :) | 09:58 |
| *** ysandeep|lunch is now known as ysandeep | 10:01 | |
| opendevreview | Andrew Bonney proposed openstack/openstack-ansible-lxc_hosts master: Take account of lxc_apt_mirror in new debootstrap command https://review.opendev.org/c/openstack/openstack-ansible-lxc_hosts/+/851203 | 10:07 |
| andrewbonney | noonedeadpunk: I noticed when doing our Yoga upgrade there isn't a Yoga section in https://docs.openstack.org/releasenotes/openstack-ansible/index.html | 10:56 |
| andrewbonney | Similarly https://docs.openstack.org/openstack-ansible/latest/admin/upgrades/major-upgrades.html still references Xena, and doesn't include the Octavia upgrade script in the manual steps | 10:56 |
| noonedeadpunk | for the first - likely we missed to merge smth | 10:57 |
| noonedeadpunk | https://review.opendev.org/c/openstack/openstack-ansible/+/847244 | 10:57 |
| *** dviroel|out is now known as dviroel | 11:35 | |
| jrosser | fwiw the demo we did at the summit was libvirt exporter + openstack exporter -> prometheus | 11:45 |
| jrosser | but it depends who the intended view is for, operator or customer | 11:45 |
| opendevreview | Andrew Bonney proposed openstack/openstack-ansible-lxc_hosts master: Take account of lxc_apt_mirror in new debootstrap command https://review.opendev.org/c/openstack/openstack-ansible-lxc_hosts/+/851203 | 12:00 |
| *** ysandeep is now known as ysandeep|break | 12:19 | |
| opendevreview | Merged openstack/openstack-ansible master: Update master for stable/yoga https://review.opendev.org/c/openstack/openstack-ansible/+/847244 | 12:29 |
| anskiy | jrosser: thank you. The problem with prometheus is that I don't have one now :) | 12:32 |
| anskiy | I've found changes for ceilometer for cpu_util and memory_util in percent from 2021 and 2018 respectively. Going to file an issue for gnocchi on desired fill behavior, and if I'll get quick answer, I could try to revive those stuck changes I've mentioned | 12:37 |
| opendevreview | Andrew Bonney proposed openstack/openstack-ansible master: rabbitmq: default to using TLS for management user interface https://review.opendev.org/c/openstack/openstack-ansible/+/851219 | 12:55 |
| *** ysandeep|break is now known as ysandeep | 12:58 | |
| jrosser | should we just bump ControlPersist= more than 60s and see if that affects our MODULE ERROR problem | 13:15 |
| mgariepy | yes | 13:23 |
| mgariepy | or fork the function in our connection plugin to fix the issue.. | 13:24 |
| mgariepy | it's not fun that ansible is not willing to fix or discuss a fix. | 13:24 |
| mgariepy | jrosser, how would you fix it ? just not testing for 255 exit status ? or just changing it to p.returncode != 0: ? | 13:29 |
| jrosser | that is a very good question | 13:29 |
| jrosser | maybe this code is just too specific to broken pipe https://github.com/ansible/ansible/blob/devel/lib/ansible/plugins/connection/ssh.py#L492-L513 | 13:30 |
| jrosser | and that any failure of any kind should trigger one retry there | 13:31 |
| mgariepy | should we then ask what is the numerous problem that matching 255 cause ? | 13:34 |
| jrosser | its pretty hard to follow as there is the outer retries loop and then the inner retry but only for the specific case of rc=255 | 13:34 |
| jrosser | sounded like they wanted to get rid of that anyway | 13:35 |
| jrosser | perhaps the whole thing can be simplified really a lot | 13:35 |
| mgariepy | you want to bump it to what ? | 13:53 |
| mgariepy | if it reproduce often at 2s, daily at 60s. i guess we shouldn't see it too often at 300s or something like this ? | 13:54 |
| jrosser | sounds reasonable | 14:18 |
| mgariepy | still it's kinda dirty.. | 14:18 |
| jrosser | yes agreed, just really would like some sort of dicussion other than computer-says-no on the issue | 14:19 |
| mgariepy | is it easier to poke them in the #ansible channel ? | 14:20 |
| jrosser | yeah it is | 14:21 |
| opendevreview | Andrew Bonney proposed openstack/openstack-ansible master: zuul: fix definition of centos 9 stream job https://review.opendev.org/c/openstack/openstack-ansible/+/851233 | 14:30 |
| opendevreview | Andrew Bonney proposed openstack/openstack-ansible-lxc_hosts master: Take account of lxc_apt_mirror in new debootstrap command https://review.opendev.org/c/openstack/openstack-ansible-lxc_hosts/+/851203 | 14:37 |
| *** ysandeep is now known as ysandeep|out | 14:54 | |
| *** dviroel is now known as dviroel|lunch | 14:57 | |
| *** dviroel|lunch is now known as dviroel | 15:52 | |
| spatel | I have windows.iso image can i convert it to qcow2 using qemu-img convert command? | 16:25 |
| mgariepy | you want to go through the windows installer via vnc ? | 16:37 |
| spatel | oh! so there is no direct way to convert iso to qcow2? | 16:56 |
| mgariepy | you probably want to create a cloud image from that iso instead no ? | 17:00 |
| spatel | I am new to windows so not sure, but when i google i found people doing qemu-img converter for iso to qcow2 | 17:03 |
| spatel | i can give it a try and see... | 17:03 |
| opendevreview | Merged openstack/openstack-ansible master: zuul: fix definition of centos 9 stream job https://review.opendev.org/c/openstack/openstack-ansible/+/851233 | 20:27 |
| *** dviroel is now known as dviroel|afk | 22:36 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!