Tuesday, 2025-11-04

opendevreviewOpenStack Proposal Bot proposed openstack/openstack-ansible master: Imported Translations from Zanata  https://review.opendev.org/c/openstack/openstack-ansible/+/96543804:07
opendevreviewDmitriy Rabotyagov proposed openstack/openstack-ansible master: Update rally SHA  https://review.opendev.org/c/openstack/openstack-ansible/+/96530808:52
kleiniDuring upgrade to E, I ran into a broken glusterfs, too. I tried fixing through backup, container delete and recreate and wondered, why glusterfs always get back into same broken state until I found /var/lib/gluster being bind mount from underlying host.12:52
noonedeadpunkI was just looking into this12:52
noonedeadpunksomehow gluster seems quite unique in terms of bindings12:53
noonedeadpunkhttps://codesearch.openstack.org/?q=_container_bind_mounts&i=nope&literal=nope&files=&excludeFiles=&repos=12:53
noonedeadpunklike ceph, mysql, zookeeper, glance - all just bind mount to "/openstack/{{ inventory_hostname }}"12:53
kleiniJust wanted to add that "plus 1" to the above gluster playbook discussion.12:53
noonedeadpunkand only gluster to "/openstack/glusterd/{{ inventory_hostname }}"12:54
noonedeadpunkso if we delete the container, this directory does not get cleaned up12:54
noonedeadpunkyeah... I wouldn't expect though it breaking during upgrade...13:00
kleiniIn my case the step to manully remove one brick might not have finished successfully (as mentioned in distro upgrade) and add_brick then kept failing. I was not able to fix it manually.13:14
opendevreviewMerged openstack/openstack-ansible-rabbitmq_server master: Change repository for deb to deb1.rabbitmq.com  https://review.opendev.org/c/openstack/openstack-ansible-rabbitmq_server/+/96519613:16
noonedeadpunkso I just tried to drop one repo container and then spawn a new one, and it worked pretty much nicely for me. but I did not drop anything yet.13:20
jrosserisnt this something to do with it preserving the brick on the deleted node then re-adding that to a re-provisioned container?13:55
jrosseriirc its much more complicated to remove the old brick and re-add a new one, than to just re-add the one from before13:55
kleinihttps://docs.openstack.org/openstack-ansible/2025.1/admin/upgrades/distribution-upgrades.html#deploying-infrastructure-hosts <- chapter 3c mentions to remove the repo node aka brick13:56
opendevreviewDmitriy Rabotyagov proposed openstack/openstack-ansible-plugins master: Allow to skip pacemaker/corosync setup  https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96606514:06
noonedeadpunkyeah, I think that's exactly the case...14:08
noonedeadpunkok, so that's a distro upgrade, not openstack upgrade14:08
opendevreviewDmitriy Rabotyagov proposed openstack/openstack-ansible-plugins master: Allow to skip pacemaker/corosync setup  https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96606514:16
kleinisorry, it is a distro upgrade not the upgrade to E. That was wrong in my initial statement.14:43
noonedeadpunk++ thanks for clarification, makes much more sense now. And I was able to reproduce the behavior14:45
noonedeadpunknot sure how to fix it though :D14:45
noonedeadpunkgluster output format is not really helping either14:46
noonedeadpunkso we can collect current uuids, and we can collect peer uuids potentially. with some intersection, we can get ones that should be removed from peering14:47
noonedeadpunkbut we need to map uuids to hostnames as well...14:47
noonedeadpunkand also remove bricks14:48
noonedeadpunkand also handle glusterfs_bootstrap_node logic, to ensure that we don't use currently missing gluster member for creating peers.14:53
noonedeadpunkoops15:06
noonedeadpunktimezones15:06
noonedeadpunk#startmeeting openstack_ansible_meeting15:06
opendevmeetMeeting started Tue Nov  4 15:06:43 2025 UTC and is due to finish in 60 minutes.  The chair is noonedeadpunk. Information about MeetBot at http://wiki.debian.org/MeetBot.15:06
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.15:06
opendevmeetThe meeting name has been set to 'openstack_ansible_meeting'15:06
noonedeadpunk#topic rollcall15:06
noonedeadpunko/15:06
noonedeadpunk#topic office hours15:09
noonedeadpunkWe had a quite productive PTG session last week, I am still lagging behind a little bit in terms of sending out the summary15:10
noonedeadpunkI am expecting to send it out today by EOD15:10
noonedeadpunkWe had also plenty of progress regarding Debian 13 support15:11
noonedeadpunkall votes are there, CI is passing15:11
noonedeadpunkIf everything goes up to the plan - we have all chances to land patches today or tomorrow15:12
noonedeadpunkPKI progress - not much so far15:14
noonedeadpunkI am still struggling to review the patch to the pki role adding the vault driver15:15
noonedeadpunkit somehow always end up somewhere at the end of the list15:15
noonedeadpunkAnd we're coming to the end of the development cycle for 2025.2. Ideally, we need to branch next week, according to the ML sent by releases team yesterday15:17
noonedeadpunkso we're running on time wrt vault driver :*15:18
noonedeadpunk*out of time15:18
noonedeadpunkwe also need to make new minor releases this week for all supported branches15:19
noonedeadpunk#link https://security.openstack.org/ossa/OSSA-2025-002.html15:19
noonedeadpunkso once patches are merged to keystone, I will propose a bump and a release15:31
noonedeadpunkWould be nice if we could backport and merge eligible to backport things15:32
noonedeadpunkand we probably should start with https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96594115:33
noonedeadpunkas it's likely blocking this one: https://review.opendev.org/c/openstack/openstack-ansible-os_neutron/+/96514215:34
noonedeadpunkas well as https://review.opendev.org/c/openstack/ansible-role-uwsgi/+/96534215:35
noonedeadpunkthat's pretty much it from my side.15:36
jrossero/ hello15:46
noonedeadpunko/15:50
noonedeadpunkShould we briefly discuss gluster thing?15:50
NeilHanlonah crap, timezone change got me :( 15:51
opendevreviewMerged openstack/openstack-ansible-rabbitmq_server master: Bump RabbitMQ version to 4.1 series  https://review.opendev.org/c/openstack/openstack-ansible-rabbitmq_server/+/96519715:51
NeilHanlongluster sure15:51
noonedeadpunkyup...15:51
noonedeadpunkso kind the thing is that we're not handling host being re-installed inside of the gluster role15:52
noonedeadpunkbut also during container re-creation we leave stale data on the host 15:52
noonedeadpunkthus I'm thinking about 2 possible actions here15:53
noonedeadpunkfirst - try to handle clean-up or better way to re-add containers/hosts to the cluster after OS re-install for instance15:53
noonedeadpunkand second - actually move /openstack/gluster/<hostname> to /openstack/<hostname>/gluster15:54
noonedeadpunkso that when you drop the container, gluster data was also cleaned up15:54
mgariepyit was working ok on 20.04 > 22.04 upgrade tho.15:54
mgariepyho i did a dist-upgrade on the hosts. 15:55
noonedeadpunkI think it works for us, but it's somehow unobvious behavior\15:55
noonedeadpunkat least we get already couple of confused users in IRC withing couple of weeks15:55
noonedeadpunkmeaning - we have smth to improve :)15:56
mgariepyyep.15:56
noonedeadpunkI have a sandbox to play with, and will try to come up with smth15:59
noonedeadpunkI wish gluster output could be in some parsable format, but I failed to find anything16:00
jrosseri think that i'd intended the gluster state to persist on the host if the container was deleted16:00
jrosserthere is a uuid and other state in the data on each host which is shared across the cluster16:01
jrosserif you destroy a node and reprovision it, quite easy to end up with gluster being N+1 nodes but one missing16:02
jrosserthis is quite easy to play with on a `infra` AIO because i think it makes 3 repo containers16:02
noonedeadpunkyeah, so I'm thinking if based of uuids we can drop the stale version of it16:02
noonedeadpunkas when container re-spawned it get's new UUID16:02
jrosseryes, then you have to somehow determine the one that should not be there and get rid of it16:03
noonedeadpunkso get all current uuids, get peers, intersect, drop missing16:03
jrosserindeed that could work16:03
jrosserbut needs to be super careful if host down or something16:03
noonedeadpunkone problem I don't understand how to figure out, is mapping of old uuid to hostname16:03
noonedeadpunkas you need to drop peer by hostname specifically, you somehow can't do that by uuid16:03
noonedeadpunkI was thinking to do action only if uuid diff exists. if host is down, it should not be a problem I guess16:04
noonedeadpunkas it won't produce weird thing...16:05
jrossergluster get-state16:08
jrosserewwwwwww16:08
jrosser`gluster peer status`16:09
jrosserbut i don't an example in the state with a deleted/recreated node just now16:09
noonedeadpunkyeah, it's not really parsable16:09
noonedeadpunkit looks smthj like this: https://paste.openstack.org/show/b6exkK3d7xbTlfwApghn/16:10
noonedeadpunk`State: Peer Rejected (Connected)`16:10
noonedeadpunkbut I was thinking of smth like this: https://paste.openstack.org/show/bD4Xhlsxp6Vfhwnwf8WG/16:11
jrosseryes we can slurp these16:12
noonedeadpunk#endmeeting16:13
opendevmeetMeeting ended Tue Nov  4 16:13:11 2025 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)16:13
opendevmeetMinutes:        https://meetings.opendev.org/meetings/openstack_ansible_meeting/2025/openstack_ansible_meeting.2025-11-04-15.06.html16:13
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/openstack_ansible_meeting/2025/openstack_ansible_meeting.2025-11-04-15.06.txt16:13
opendevmeetLog:            https://meetings.opendev.org/meetings/openstack_ansible_meeting/2025/openstack_ansible_meeting.2025-11-04-15.06.log.html16:13
noonedeadpunkbut like mapping of uuid to the hostname is biggest pita right now...16:13
jrossermaybe we should set a local fact as part of the deployment16:16
noonedeadpunkthat would help, sure16:17
jrosseras we know the uuid for each node from glusterd.info when it gets created16:17
noonedeadpunkbut if we're redeploying the thing...16:17
noonedeadpunkwe are loosing the fact16:17
jrosser`this_uuid` `all_uuids` perhaps16:18
jrosserbut not sure16:18
noonedeadpunkah16:19
noonedeadpunklol16:19
noonedeadpunkI can slurp file from peers16:19
noonedeadpunkhttps://paste.openstack.org/show/b08e52WF71QCKaK5oMDI/16:20
noonedeadpunkand pretty much same with bricks https://paste.openstack.org/show/bdKxUJXsfSTiZBOpXuH1/16:21
noonedeadpunkit's annoying though16:21
noonedeadpunksounds more like a module :D16:21
jrosseryes i was thinking that16:22
jrosserand was just (disappointedly) looking at the gluster collection16:23
jrosser`gluster_info` or something16:23
spatelHello noonedeadpunk 16:41
noonedeadpunk[e]o/16:42
spatel:)16:42
spatelafter long time 16:42
spatelI was super busy in new k8s deployment 16:43
spatelhow are you guys doing?16:43
noonedeadpunkyeah. been a while indeed :)16:48
noonedeadpunkso how you like helm charts these days ? :p16:48
noonedeadpunkwas really a nice summit and ptg16:48
noonedeadpunkbut now too many things on plate as well16:48
spatelnoonedeadpunk hey! I am learning helm chart 17:21
spatelour plan is to move out from openstack and build all containertized app 17:22
spatelI am building k8s on bare metal and playing with to see how it goes 17:22
noonedeadpunkomfg17:23
noonedeadpunkwell. I guess good luck then :)17:23
opendevreviewMerged openstack/openstack-ansible master: Update rally SHA  https://review.opendev.org/c/openstack/openstack-ansible/+/96530817:29
spatellol17:44
spatelnoonedeadpunk how is everything your side.. anything new outside openstack?17:45
noonedeadpunkno, not at all... just waaaay more openstack with VMware migrations now...17:46
spatelHmm nice!! 17:55
spatelHow is your bgp deployment is in production with ovn?17:56
noonedeadpunkI think will know in a month when first upgrade time will come :D17:56
noonedeadpunkotherwise it works17:56
noonedeadpunkbut I'm not sure if load in the region is enough to evaluate how good does it work17:57
noonedeadpunkI still don't like a design tbh17:57
noonedeadpunkbut hopefully native bgp in ovn will make things better17:57
noonedeadpunkand it was really nice summit this time in Paris. It was almost twice from Vancouver.18:06
noonedeadpunkSo felt really nice, but also super busy over there18:06
spatelmy company not approving budget otherwise I would love to see you guys 18:16
spatelnoonedeadpunk I agreed about ovn-bgp stuff.. its kind of hacky when python script doing stuff 18:17
spateltotally agreed with ovn bgp native support 18:17
spatelIn my company I am trying to deploy ceph with BGP (its over engineering at this point but plan is to do L3 last mile in DC)18:18
spatelgoing to run FRR in each host to advertise IPs in fabric 18:19
noonedeadpunkwell, I actually have ceph in L3 in one of deployment18:33
noonedeadpunkthough it's OSPF for some reason (don't ask me why)18:34
noonedeadpunkand it was birdc18:34
noonedeadpunkand I got storage completely down for a while, as somehow ospf master was on controllers, and I messed up with prio on controllers during OS upgrade....18:35
mgariepythat's a good day18:39
noonedeadpunkyeah.... 18:41
noonedeadpunkso I'm shutting down a single controller node, and whole region goes black the same minute...18:41
mgariepyyep, from that single time you wont do that again :)18:42
opendevreviewMerged openstack/openstack-ansible-os_neutron stable/2024.1: Rename  OVN BGP Agent SHA pinning var  https://review.opendev.org/c/openstack/openstack-ansible-os_neutron/+/96514419:16
opendevreviewMerged openstack/openstack-ansible-os_neutron stable/2024.2: Rename  OVN BGP Agent SHA pinning var  https://review.opendev.org/c/openstack/openstack-ansible-os_neutron/+/96514319:16
opendevreviewMerged openstack/openstack-ansible master: Add debian trixie job definitions  https://review.opendev.org/c/openstack/openstack-ansible/+/95461619:52
opendevreviewMerged openstack/openstack-ansible master: Imported Translations from Zanata  https://review.opendev.org/c/openstack/openstack-ansible/+/96543819:52
opendevreviewDmitriy Rabotyagov proposed openstack/openstack-ansible-plugins master: Split glusterd tasks to task files  https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96611620:22
opendevreviewDmitriy Rabotyagov proposed openstack/openstack-ansible-plugins master: [DNM] Clean-up stale gluster peers  https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96611720:22
opendevreviewMerged openstack/openstack-ansible stable/2024.2: Ensure that gluster mount is started after service  https://review.opendev.org/c/openstack/openstack-ansible/+/96521120:23
noonedeadpunkSo I started pulling in together gluster thing: https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96611720:26
noonedeadpunkI _think_ this might be enough to do some cleaning-up....20:27
opendevreviewDmitriy Rabotyagov proposed openstack/openstack-ansible-plugins master: [DNM] Clean-up stale gluster peers  https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/96611720:29
jrosserblimey20:34
*** mnaser[m] is now known as mnaser20:46

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