Tuesday, 2023-03-14

opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: WIP: Cleanup ceph install on cs9  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714701:54
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: WIP: Cleanup ceph install on cs9  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714704:37
opendevreviewTakashi Kajinami proposed openstack/manila master: Remove TripleO job  https://review.opendev.org/c/openstack/manila/+/87733505:26
opendevreviewTakashi Kajinami proposed openstack/manila master: Remove TripleO job  https://review.opendev.org/c/openstack/manila/+/87733505:29
carthacaperiodic tasks are giving me headaches https://bugs.launchpad.net/oslo.service/+bug/2011587, related bug in manila is https://bugs.launchpad.net/manila/+bug/2008497 - I appreciate any thoughts on this14:15
gouthamrhey carthaca - the log for one is annoying if we're not picking up the decorated function's name: https://opendev.org/openstack/oslo.service/src/commit/43413775fca9431bcb3b7aae91c58f7e885f27a3/oslo_service/loopingcall.py#L188 15:15
gouthamrcarthaca: to check the scale of things, does the backend have a lot of replicas? 15:18
carthacagouthamr: no, not many replicas - I see a 700 sec delay for a backend that has roughly 30 share instances... ah but 1200 deleted ones, let me check the share servers15:25
carthacagouthamr: nope, just 3 share servers overall15:26
carthacagouthamr: all our shares are replication enabled (replication_type dr), but real replicas i.e. replication_state!='active' are only 10 out of those 120015:36
gouthamrcarthaca: the task ran longer than the interval by 700 seconds? 15:42
carthacayes, and my interval is configured at 300, so it took 1000 seconds15:43
gouthamrcarthaca: :| wow that's bad.. 15:49
gouthamr there are 11 periodic tasks in the share manager, and 1 of them is disabled by default (think there's another one we can disable by default too) 15:50
gouthamri wonder if the db query (https://github.com/openstack/manila/blob/74ae712d58a01daf2a6a1ca76ea8a43cfeaff4e7/manila/share/manager.py#L2672-L2679) is inefficient15:51
gouthamrthen there is work being done to repair replicas in the netapp driver during the update call: https://github.com/openstack/manila/blob/29db644db5a60a93356a7837da3d9f7551bb95f4/manila/share/drivers/netapp/dataontap/cluster_mode/lib_base.py#L2555 15:52
gouthamrwe could try factoring those out into separate threads15:53
carthacare db query: the filtering should move to the db layer probably15:58
carthacabut the other place, the netapp driver on replica update, is more likely to consume that many time15:58
carthacais it wise to spawn threads within the periodic tasks? probably this would need a bigger rewrite, then16:01
* gouthamr re-reads the netapp code16:01
gouthamragree, feels like a minimal gain there if any16:03
gouthamr> is it wise to spawn threads within the periodic tasks? probably this would need a bigger rewrite, then16:04
gouthamrthere're separate periodic threads in the netapp driver16:05
gouthamrcurrently used for deferred cleanups; that pattern is good for reconciling driver's resources since won't overload the looping call that we use in the share manager 16:06
opendevreviewMerged openstack/manila stable/zed: update capacity filter during extend share  https://review.opendev.org/c/openstack/manila/+/87622716:07
opendevreviewMartin Kopec proposed openstack/devstack-plugin-ceph master: DNM testing out  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87738816:12
carthacagouthamr: seems like I just needed you to be my rubber duck, thanks16:20
carthacagouthamr: the db calls are the slow thing https://paste.opendev.org/show/bSdsJTqVpArecXhTLjiu/16:20
gouthamrcarthaca: :P happy to help16:20
gouthamrdid you prune the deleted shares?16:21
carthacagouthamr: 10 minutes to query all the stuff until it goes from 'Updating status of share replica instances' to 'Updating status of share share_replica 12..'16:21
gouthamrah :O16:22
carthacasame for periodic_share_replica_snapshot_update16:23
carthacaI'm already pruning after 7 days16:23
gouthamrack, the code attempts the same query there and looks for snapshots of non-active replicas16:24
gouthamrack carthaca; so our bug here would be to optimize the db queries in those two methods 16:27
gouthamrstill interested in someone looking at the oslo.service bug and telling us if we're missing some config opts 16:28
carthacaI fully agree16:29
carthacadeleted object should be already ignored https://github.com/openstack/manila/blob/74ae712d58a01daf2a6a1ca76ea8a43cfeaff4e7/manila/db/sqlalchemy/api.py#L194816:30
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: WIP: Cleanup ceph install on cs9  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714718:33
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: WIP: Cleanup ceph install on cs9  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714718:45
*** dviroel_ is now known as dviroel20:21
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: Cleanup installation and revive cephfs-nfs job  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714723:17
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: Cleanup installation and revive cephfs-nfs job  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714723:21
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: Remove branch specific code from repo  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87743523:42
opendevreviewGoutham Pacha Ravi proposed openstack/devstack-plugin-ceph master: Cleanup installation and revive cephfs-nfs job  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/87714723:58

Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!