Friday, 2023-07-14

opendevreviewJulia Kreger proposed openstack/ironic master: DNM: Don't actually heartbeat with sqlite!  https://review.opendev.org/c/openstack/ironic/+/88850601:09
TheJuliaso, that might be a bit extreme, but there is a point where everyone should just use a full database.01:09
opendevreviewJulia Kreger proposed openstack/ironic master: DNM: Don't actually heartbeat with sqlite!  https://review.opendev.org/c/openstack/ironic/+/88850604:53
dtantsurTheJulia: I don't think it's extreme to disable conductor heartbeating for an architecture that is very explicitly single-conductor. That being said, retries will probably solve most of our issues.08:07
JayFI wonder if that conductor heartbeat getting stale will have side effects we didn't consider08:09
JayFIs metal3 single conductor or single process?08:27
JayFit's still separate api/cond, yeah08:27
dtantsursingle process08:45
dtantsurmetal3 does support the split mode though08:45
dtantsurbut that will require mariadb08:46
dtantsurone of our problems (our = metal3) is that we don't really prescribe any architecture, we support too many options08:47
dtantsurI'm trying to fix that with https://github.com/dtantsur/ironic-operator but no idea when I'm able to finish that08:48
dtantsurJayF: ^^08:48
JayFthat doesn't sound familiar at all08:48
JayFlol08:48
dtantsur:D08:48
dtantsur.. maybe I should decide it's my last day before the PTO, and I should just hack on this operator ...08:49
JayFYES08:49
JayFAs PTL of Ironic; I'm assigning you an urgent task08:49
JayFto go do whatever you think is best today ;) 08:49
dtantsur\o/08:49
JayFI assign myself that task every morning lol08:51
opendevreviewDmitry Tantsur proposed openstack/ironic master: Document caveats of running with SQLite  https://review.opendev.org/c/openstack/ironic/+/88852909:24
dtantsurJayF, TheJulia ^^^09:24
dtantsurI wonder if there is still value of having different guides for RH/Debian09:26
opendevreviewDmitry Tantsur proposed openstack/ironic master: Stop splitting installation docs per distros  https://review.opendev.org/c/openstack/ironic/+/88853109:38
dtantsurokay, I wanted to do it for a long time ^^09:38
JayFdtantsur: fwiw; if we wanted to support migrations on sqlite, if we wanted to pick a release to start I think we can09:55
JayFdtantsur: I have the knowledge on what can/can't be done and how to get around them on sqlite09:55
* JayF used to work on a sqlite-only project that had a ton of alembic migrations09:55
JayFdtantsur: > This is not a typo! A full path requires 4 slashes.09:56
JayFdtantsur: I needed this knowledge last month, I spent way too long learning that the hard way :D 09:56
dtantsurWe've all been there :D10:15
dtantsurJayF: re sqlite migrations: yes, I'd like to have this option.10:15
JayFdtantsur: post to the mailing list about it then and gather consensus, I'm happy to have the discussion 10:18
dtantsurJayF: I'll be out for the next 2 weeks, don't want to start any large discussions10:18
JayFonly real hilarious bit would be, adding more migrations testing right now seems like the stupidest possible thing we could do lol10:18
JayFdtantsur: that's the best time! Just set off the smoke bomb and run ;)10:18
dtantsur:D10:18
JayFseriously, I don't think it should be too controversial; maybe even -nv a job to validate it10:21
JayFonly real pain is sqlite migrations in alembic can sometimes end up  as "Do N^2 things for N rows"10:22
JayFso they are not fast, but they also are not typically large so it mostly is a wash10:22
JayF(for migrations unsupported by sqlite; alembic does a batch migration where it basically recreates the table under a new name and schema, copies over the data, and then replaces the old table with the new)10:22
dtantsurI think it's fair to say that sqlite migrations will always be offline ones, with downtime10:25
dtantsurand that the database sizes will be modest10:25
JayFyou know what10:32
JayFwe shoudl wait until post-sqla2.0 to do this10:32
JayFbecause otherwise we'll have a crapton of 1.4/2.0 conditionals in the same way as we do for the WAL event stuff, I think10:32
JayFthat would also put it at the start of a slurp, which I doubt sqlite users care about but still is sensible10:32
opendevreviewMerged openstack/bifrost master: Consolidate ubuntu/debian required_defaults  https://review.opendev.org/c/openstack/bifrost/+/88844410:33
opendevreviewDmitry Tantsur proposed openstack/ironic master: Document caveats of running with SQLite  https://review.opendev.org/c/openstack/ironic/+/88852910:36
dtantsurJayF: agreed10:36
JayFdtantsur: then go agree with my TC resolution to set a deadline on it, as was mostly consensus at the forum ;) https://review.opendev.org/c/openstack/governance/+/88708310:38
opendevreviewMerged openstack/bifrost master: sgabios-bin is a subpackage of sgabios  https://review.opendev.org/c/openstack/bifrost/+/88844510:45
opendevreviewMerged openstack/bifrost stable/xena: Skip unnecessary SDK get_machine calls  https://review.opendev.org/c/openstack/bifrost/+/88311810:45
dtantsurCould someone please merge https://review.opendev.org/c/openstack/bifrost/+/887613 ?12:21
dtantsurIt is very annoying to test inspection because of it12:21
JayF+2 if it's still blocking you just workflow it and we can revert it if anyone disagrees 😁12:58
TheJuliaJayF: mostly just hash ring resets which I think I've identified and... addressed13:23
TheJuliaAlso, good morning everyone!13:23
TheJuliawell, looks like retry logic and turning off heartbeating seems to be the magic trick13:29
TheJuliawe'll need to retry lock handling as well it looks like based upon my latest change13:29
TheJuliadtantsur: replied to most of your comments, basically agree, none of those patches were in final form13:36
TheJuliaI got a result where I watched the heartbeater somehow just line up perfectly to thrash everything into the ground which started me on the "just disable the heartbeater" path13:36
TheJuliaRemarkably, looking at the latest log, we still had an incoming request for a lock error, and had an earlier retry occur, so... I think we're on the "right" path (even though I'll restack that one first on master and go from there.13:37
dtantsurmeanwhile, I got a first successful Ironic installation using https://github.com/dtantsur/ironic-operator \o/13:59
dtantsurit's very limited and cannot be used for e2e deployments, but it's something!13:59
opendevreviewVerification of a change to openstack/bifrost master failed: Allow nginx to access images again on systems with SELinux  https://review.opendev.org/c/openstack/bifrost/+/88761314:53
dtantsurSee you on July 31st, folks o/ I'll turn off my IRC bouncer at some point, please use email.14:55
TheJuliaoh wow, have a great time off14:59
* TheJulia needs to take some PTO off soon15:02
NobodyCamGood Morning OpenStack Folks... And of course.....15:19
NobodyCamTGIF!15:19
TheJuliagood morning NobodyCam 15:21
TheJulia\o/ no more wedding this weekend /o\15:21
NobodyCamO/ morning TheJulia 15:21
NobodyCamoh that's good (because of the heat) and ....15:21
TheJuliagroom crashed his motorcycle15:22
TheJulia:(15:22
TheJuliaso he needs to heal first15:22
NobodyCamoh no :( 15:22
TheJuliaand... well.. 119F was the updated forecast as of last night :(15:23
TheJuliaThis weekend is going to be "not fun"15:23
NobodyCamHope he'll recover quickly15:23
TheJuliaI suspect he will15:23
NobodyCamyea, we're planing on heading out early (for us) on Saturday15:23
NobodyCam\o/ for recovery15:24
TheJuliamakes sense15:25
TheJuliasooo many things need to be touched when you start to monkey with the hash ring15:42
opendevreviewJulia Kreger proposed openstack/ironic master: Don't actually heartbeat with sqlite  https://review.opendev.org/c/openstack/ironic/+/88850616:22
TheJuliaalmost have a meta retry decorator17:31
TheJuliajust have to remember how to twist tenacity's arm to my will17:32
*** tzumainn__ is now known as tzumainn17:47
*** janders6 is now known as janders20:34
opendevreviewMerged openstack/bifrost master: Reduce the libvirt/qemu packages list  https://review.opendev.org/c/openstack/bifrost/+/88844623:48

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