Monday, 2024-05-27

opendevreviewRajat Dhasmana proposed openstack/openstacksdk master: WIP: Add support for default volume types  https://review.opendev.org/c/openstack/openstacksdk/+/92046407:03
opendevreviewRajat Dhasmana proposed openstack/python-openstackclient master: WIP: Add support for default volume type commands  https://review.opendev.org/c/openstack/python-openstackclient/+/92046507:04
dtantsurI wonder what the chances are that https://opendev.org/openstack/openstacksdk/commit/2b4aeff6d3972761da9c4a6a8b696103664a758b causes memory leaks in the code that worked before10:48
fricklerdtantsur: do you actually see some memory leak or is this just a theoretical question?10:57
dtantsurfrickler: I do see ironic-inspector leaking memory. Taking periodical snapshows from Python's tracealloc led me here: https://paste.opendev.org/show/b9HRCbkxqtPrZ3yU0TCH/10:58
dtantsurbut after looking at lines, I'm not longer sure it's related to __del__10:58
dtantsurI'm now wondering if we somehow make CONF grow infinitely10:59
dtantsurAt the very least, it looks like we're modifying global defaults..11:01
dtantsurmmm, or not really, it has copy() everywhere11:01
* dtantsur is puzzled11:01
fricklerdtantsur: which version are you using? there was a leak fixed a short time later https://review.opendev.org/c/openstack/openstacksdk/+/89078111:09
dtantsurfrickler: it's possible that I don't have the newest version, let me check11:10
dtantsuralas, I have this change already11:10
opendevreviewJens Harbott proposed openstack/openstackclient master: zuul: Drop retired repos from required-projects  https://review.opendev.org/c/openstack/openstackclient/+/92051212:45
opendevreviewDmitry Tantsur proposed openstack/openstacksdk master: Do not create a class in runtime on a potentially hot path  https://review.opendev.org/c/openstack/openstacksdk/+/92054113:26
dtantsurOkay, this is the first, unfortunately by far the smallest, part of the puzzle ^^13:26
dtantsurfrickler, so, is now calling close() on a Connection mandatory? I don't think it was always the case.14:01
dtantsurHonestly, using atexit smells badly. Are you sure you guys considered long-running users of openstacksdk?14:04
fricklerdtantsur: I've so far successfully avoided to dig into performance relevant details of python implementations, will defer to gtema and stephenfin 14:10
gtemadtantsur - token caching currently relies on calling close14:10
dtantsurgtema: it's off by default though?14:11
dtantsurI mean, the caching?14:11
gtemawell, yes - it is config option14:11
gtemaand off by default14:11
dtantsurAt the very least, you need a very, VERY ugly warning in the release notes14:11
dtantsurI don't think we're the only people who don't call close()14:11
gtemacurrently the cli does so (since that is the main user of caching)14:12
dtantsurgtema: related question: are Connections long lived and (green) thread safe? Can I just cache one for the lifetime of ironic-inspector?14:13
gtemawhat is this lifetime? I have certain usecases with connections living multiple months14:13
dtantsurgtema: it's an OpenStack service. Weeks, months, in a very unlikely case years :)14:14
gtemashould be thread safe14:14
dtantsurin any case, it's a pity that the change went in without a release note (or am I not looking carefully enough?)14:15
gtemathis is there since years14:16
dtantsurlast year, to be precise14:16
gtemaok, could be, time is relative14:17
gtema:)14:17
dtantsurgtema: also for your attention: https://review.opendev.org/c/openstack/openstacksdk/+/92054114:19
gtemaack14:19
opendevreviewMerged openstack/openstackclient master: zuul: Drop retired repos from required-projects  https://review.opendev.org/c/openstack/openstackclient/+/92051214:36
opendevreviewMerged openstack/openstackclient master: Remove clients of retired projects  https://review.opendev.org/c/openstack/openstackclient/+/91985721:00

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