| opendevreview | Merged openstack/tooz master: zookeeper: fix argparse related to ssl https://review.opendev.org/c/openstack/tooz/+/927370 | 05:23 |
|---|---|---|
| opendevreview | Dirk Mueller proposed openstack/osprofiler master: Unpin for newer elasticsearch versions https://review.opendev.org/c/openstack/osprofiler/+/927749 | 06:19 |
| opendevreview | Dirk Mueller proposed openstack/osprofiler master: Update bandit to 1.7.x https://review.opendev.org/c/openstack/osprofiler/+/927776 | 06:23 |
| opendevreview | Dirk Mueller proposed openstack/osprofiler master: Bump hacking to latest https://review.opendev.org/c/openstack/osprofiler/+/927777 | 06:28 |
| *** elodilles_pto is now known as elodilles | 07:26 | |
| opendevreview | Takashi Kajinami proposed openstack/oslo.db master: Remove deprecated async property https://review.opendev.org/c/openstack/oslo.db/+/927785 | 07:30 |
| opendevreview | Takashi Kajinami proposed openstack/oslo.db master: Remove deprecated async property https://review.opendev.org/c/openstack/oslo.db/+/927785 | 07:30 |
| opendevreview | Takashi Kajinami proposed openstack/oslo.db master: Remove deprecated async property https://review.opendev.org/c/openstack/oslo.db/+/927785 | 07:33 |
| kevko | Hi, please, can u please advise me if i am correct if I say that in kolla deployment (containerized without /dev/shm shared between containers) we need to setup lock_path from oslo_concurrency to some writable path by a service and also mount /dev/shm of docker host so we will share it between containers ? or /dev/shm doesn't need to be shared ? | 12:02 |
| kevko | Or in other words, can every container have his own /dev/shm and set lock_path to /dev/shm/oslo_lock for example ...because it seems if i am using queue_manager from oslo.messaging ..oslo.concurrency is not deleting the locks ... | 12:03 |
| opendevreview | Takashi Kajinami proposed openstack/oslo.messaging master: Use more sensible "default" for [oslo_messaging_rabbit] processname https://review.opendev.org/c/openstack/oslo.messaging/+/927810 | 12:49 |
| tkajinam | kevko, for queue manager what I understand is that the lock is used per hostname+process, so unless you run two same service(eg nova-scheduler) in a same host you don't need to share lock directories | 12:49 |
| tkajinam | there might be a case where lock_path is used for locking among services but I've not really examined the full usage... if you want to keep it in safe side then I'd suggest creating a directory per service | 12:51 |
| tkajinam | and share it among containers for the specific service | 12:51 |
| kevko | tkajinam: this is exactly what i think it should be | 13:14 |
| kevko | tkajinam: but what is the affect if lock file was created and i restart container ..and old lock file is there ? shouldn't be lock_path set to /dev/shm for example ? | 13:15 |
| kevko | tkajinam: question also is ... if i want to fix this approach globally ... (not only for queuee_manager .. and i really don't know on which all places this type of locking mechanism is used - i suppose not many - but still ...) ..i mean the oslo_concurrency lock_path ... how it should be ? create tmpfs for examplemb (as it is by default in docker) | 13:19 |
| kevko | and mount it for every container as common /dev/shm ? + implement startup script which will delete locks regarding queue_manager for specific service which is starting up ? | 13:19 |
| kevko | simply question .. can be this file mechanism used across the services (i know queue_manager creating lock as hostname+process name ...but it can be different in other places) ...if yes ..it should be shared ... | 13:21 |
| tkajinam | kevko, Idk if /dev/shm is the good path to be used for lock_path. historically rhel systems use /var/lib/<service>/tmp while debian family uses /var/lock/<service> | 14:37 |
| tkajinam | the mechanism can technically be used across services though I don't think many services may have such multi-process locking really. Especially if lock should be shared among nodes then tooz is the preferred way to handle such case | 14:38 |
| tkajinam | the only exception I'm aware of was lock_path used for os-brick but that work is still being implemented and can be ignored now, afair | 14:39 |
| kevko | tkajinam: thank you Takashi | 16:07 |
| kevko | tkajinam: As I dive deeper into these things, I'm increasingly surprised. Some containers have /dev/shm shared with the host, while others don't. As a result, some containers have a clean /dev/shm after a restart, while others don't. I would say that this could cause some issues. | 16:08 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!