opendevreview | renliang proposed openstack/kolla-ansible master: Fix incorrect docker storage driver document link https://review.opendev.org/c/openstack/kolla-ansible/+/927696 | 03:07 |
---|---|---|
opendevreview | Ivan Vnučko proposed openstack/kolla-ansible master: Add backend TLS encryption between RabbitMQ management and HAProxy https://review.opendev.org/c/openstack/kolla-ansible/+/919086 | 07:20 |
*** MatejFeder|Outofofficeuntil020 is now known as MatejFeder[m] | 07:43 | |
kevko | So, let's backport elastic support :) https://www.elastic.co/blog/elasticsearch-is-open-source-again | 07:44 |
bbezak | :) | 07:45 |
SvenKieske | it's nice that it's AGPL, but then again opensearch has security plugins like auth stuff open sourced, afaik elastic has those still as proprietary? | 07:53 |
SvenKieske | good morning btw :) | 07:53 |
SvenKieske | kevko: maybe you can look at my notes for the rabbitmq streams vs quorum queues stuff? I tried to write the current state up in: https://etherpad.opendev.org/p/KollaWhiteBoard#L72 would be interested in your opinions. | 07:54 |
SvenKieske | e.g. did anybody already use streams anywhere? | 07:55 |
bbezak | kevko: could you please take a look into those two changes in the chain? https://review.opendev.org/c/openstack/kolla/+/927461 | 08:09 |
kevko | SvenKieske: I tried streams and it didn't work 100 % I turned off | 09:09 |
kevko | I will check both folks ..but we are moving offices ..so it will be in hour or two | 09:10 |
opendevreview | Matúš Jenča proposed openstack/kolla master: Change Manila container user to root https://review.opendev.org/c/openstack/kolla/+/927722 | 09:17 |
kevko | SvenKieske: yeah, but I think we can provide a way how to choose between opensearch/elastic | 09:30 |
SvenKieske | fine with me I guess, if it doesn't add a thousand lines ;) | 09:34 |
kevko | SvenKieske: bbezak: btw, we lastly forgot to discuss https://review.opendev.org/c/openstack/kolla/+/926744 vs https://review.opendev.org/c/openstack/kolla-ansible/+/926632 << My opinion is to just make images simpler than add some special role to fetch additional information to just render into some templates ... symlinks are normal way how to | 09:38 |
kevko | handle such stuff | 09:38 |
opendevreview | Matúš Jenča proposed openstack/kolla-ansible master: Enable backend TLS for Manila https://review.opendev.org/c/openstack/kolla-ansible/+/927725 | 09:56 |
dougszu | Hey folks, do we still want to restore Zookeeper? (for Tooz) | 09:58 |
dougszu | If so, there is this for starters: https://review.opendev.org/c/openstack/kolla/+/920405 | 09:59 |
opendevreview | Michal Arbet proposed openstack/kolla-ansible master: Add ansible-core as a dependency https://review.opendev.org/c/openstack/kolla-ansible/+/922369 | 12:32 |
kevko | SvenKieske: hmm, i am squashing my changes into one - for use of queue_manager .... i think we need to share /dev/shm between containers and set also oslo_concurrency lock | 14:27 |
SvenKieske | kevko: looking forward to that, I'm unfortunately very much blocked for most of the week with customer work and travel, so don't expect fast responses by me :) | 14:28 |
kevko | SvenKieske: hmm, btw i think the current implementation is not good from the beginning in kolla... | 14:36 |
kevko | SvenKieske: classic deployment (not containerized) take into account that processes (openstack service - nova, neutron as example) have shared /dev/shm for intercommunication (for now queue_manager as example ..accessing /dev/shm) ...but containerized deployment which kolla is don't do that. In kolla every container has shared memory of 64MB size | 14:40 |
kevko | ..and host /dev/shm is not accessible ..and also processes between (nova-scheduler, nova-api as example) don't have /dev/shm same ... So i think we need to mount /dev/shm:/dev/shm OR create tmpfs for kolla_dev_shm and mount to all containers as shared /dev/shm ... | 14:40 |
SvenKieske | mhm, what do the other deployments do? why would that be necessary? in a distributed system you can't take it for granted you can communicate with other processes via local shared memory? I must be missing something | 14:47 |
kevko | SvenKieske: https://github.com/openstack/oslo.messaging/blob/master/oslo_messaging/_drivers/amqpdriver.py#L45-L99 | 14:47 |
kevko | SvenKieske: https://github.com/openstack/oslo.messaging/blob/master/releasenotes/notes/rabbit_queue_manager-363209285cbbe257.yaml | 14:49 |
SvenKieske | ok, so that is namespaced per host already and it's only relevant for the queue manager itself, got it. | 14:50 |
kevko | SvenKieske: yes - which also means streams ... | 14:51 |
SvenKieske | ok, so this really requires same /dev/shm/ per container per service per host.. | 14:51 |
kevko | SvenKieske: but, i can't confirm IF some other services in a code not using /dev/shm in a code ... | 14:51 |
kevko | SvenKieske: what i am trying to say that if code using /dev/shm (in this case oslo.messaging) and assumes it's shared ...it's bad from the kolla beginning ... because every container has own /dev/shm of 64m | 14:53 |
opendevreview | Matúš Jenča proposed openstack/kolla master: Gnocchi: Bump to 4.6.3 https://review.opendev.org/c/openstack/kolla/+/927741 | 14:53 |
SvenKieske | sure, that's the point of containers. But imho it's more problematic from the design pov of openstack as a whole: everybody says it's a "distributed system" and designed for scale etc and then someone comes around and just assumes distinct service processes can read each others ram..doesn't make a lot of sense tbh. | 14:54 |
SvenKieske | either say "it must be deployed on the same host" or "we are distributed for scale, we don't assume local access to different processes of the same service" you can't have both without pain | 14:55 |
SvenKieske | seems we chose pain :) | 14:56 |
kevko | SvenKieske: I din't implement shared memory in my testing deployment ... and I see this for example in heat_engine ... (queue_manager on ) | 14:57 |
kevko | SvenKieske: (heat-engine)[root@controller0 /]# cat /dev/shm/controller0_heat-engine_qmanager | 14:57 |
kevko | 1:36(heat-engine)[root@controller0 /]# | 14:57 |
SvenKieske | well skimming that queue manager code I would expect any openstack service using it will need /dev/shm then. | 14:58 |
kevko | after restart heat_engine it's the same ...but as I said before ...streams didn't work for me on my testing stack ...so I started wondering if it is not about this .... | 14:58 |
SvenKieske | interesting blog post: https://hynek.me/articles/docker-virtualenv/ | 15:00 |
kevko | SvenKieske: ^^ yep, as you said ... but I am now wondering if i add /dev/shm:/dev/shm to kolla-ansible .... OR create tmpfs for kolla usage and mount to all containers | 15:00 |
SvenKieske | you might want to check if /dev/shm sharing even works in podman, might be troublesome :) | 15:00 |
kevko | SvenKieske: hmm, it looks like only oslo.messaging using this https://codesearch.opendev.org/?q=%2Fdev%2Fshm&i=nope&literal=nope&files=&excludeFiles=&repos= | 15:02 |
kevko | *this approach | 15:02 |
SvenKieske | I guess most people communicate..well over rabbitmq? :D oslo can't do that for rabbit coordination itself I guess | 15:04 |
opendevreview | Matúš Jenča proposed openstack/kolla master: Gnocchi: Bump to 4.6.3 https://review.opendev.org/c/openstack/kolla/+/927741 | 15:04 |
kevko | SvenKieske: is it problem to just mount /dev/shm of host to /dev/shm of containers ? | 15:05 |
SvenKieske | I don't know, but there where problems with some other low level mount points in the past imho | 15:09 |
SvenKieske | I guess CI will tell us ;) | 15:09 |
kevko | SvenKieske: Haha, CI works at about 80 percent :) | 15:12 |
opendevreview | Martin Hiner proposed openstack/kolla-ansible master: Move to high level client in DockerWorker https://review.opendev.org/c/openstack/kolla-ansible/+/908295 | 15:38 |
opendevreview | Michal Arbet proposed openstack/kolla-ansible master: [DNM] Use host's shared memory https://review.opendev.org/c/openstack/kolla-ansible/+/927752 | 17:19 |
opendevreview | Merged openstack/kolla-ansible master: Drop prometheus-msteams support https://review.opendev.org/c/openstack/kolla-ansible/+/927001 | 17:52 |
opendevreview | Verification of a change to openstack/kolla master failed: Drop prometheus-msteams container images https://review.opendev.org/c/openstack/kolla/+/927000 | 18:02 |
frickler | meh, kevko jinxed it ;-) | 18:05 |
*** priteau_ is now known as priteau | 19:12 | |
opendevreview | Merged openstack/kolla master: Drop prometheus-msteams container images https://review.opendev.org/c/openstack/kolla/+/927000 | 20:56 |
mnasiadka | frickler, kevko: how about https://review.opendev.org/c/openstack/kolla/+/927461 ? ;-) | 22:31 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!