21:04:32 <mattoliver> #startmeeting swift
21:04:32 <opendevmeet> Meeting started Wed Dec 13 21:04:32 2023 UTC and is due to finish in 60 minutes.  The chair is mattoliver. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:04:32 <opendevmeet> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:04:32 <opendevmeet> The meeting name has been set to 'swift'
21:04:46 <mattoliver> Whose here for the swift meeting?
21:04:55 <mattoliver> o/
21:04:57 <jianjian> maybe it's eventlet issue, does eventlet support 3.12 yet?
21:05:11 <mattoliver> jianjian: welcome!
21:05:24 <jianjian> thanks :-)
21:05:32 <mattoliver> acoles: wont be here unfortuantely
21:06:07 <mattoliver> I've only just updated the agenda (last minute based on what I could see in email), so as you can see I'm super ready for this :P
21:06:22 <mattoliver> #link https://wiki.openstack.org/wiki/Meetings/Swift
21:07:06 <mattoliver> Looks like it's just you and me today jianjian, not sure where everyone else is 🤷
21:07:20 <mattoliver> but thats ok. Could be a nice quick one :)
21:07:39 <mattoliver> #topic Tim on jury duty
21:08:19 <mattoliver> So tim isn't around and he may be off until mid to late Jan he says.
21:08:55 <jianjian> yeah, Jan 19 I remember
21:09:23 <mattoliver> As we all know timburke does an amazing amount of work for the project, so we may need to pitch in and help while he's absent.
21:09:28 <mattoliver> oh nice thanks jianjian
21:10:16 <mattoliver> I'm happy to chair the meetings while he's gone and attempt to do other things, but I'm defintely no timburke .
21:10:28 <mattoliver> Next topic
21:10:44 <mattoliver> #topic  Christmas break
21:10:58 <jianjian> thanks Matt for carrying the baton
21:11:12 <mattoliver> nps
21:12:01 <jianjian> and a lot of swift developers will be on christmas break, plus company holiday starting from Dec.21
21:12:01 <mattoliver> For those of us in countries that celebrate this time of year the next few weeks will be a little spotty with availablity. So we need to decide what to do about meetings and such
21:12:50 <mattoliver> yeah!
21:14:17 <jianjian> I am available for next week, but not the one after
21:14:25 <mattoliver> kk, I think meeting on the 27th should defintely be off. But I'll also be on vacation on 3rd Jan. So I wonder if we should skip those 2.
21:15:09 <mattoliver> OK, so lets have a meeting next week and that can be the last. Maybe the first meeting of the new year will be on 10th of Jan.
21:15:56 <mattoliver> I'll make sure I make a not of that the swift meeting page and maybe figure out how to add it to the channel header here
21:16:19 <mattoliver> Sound good to you jianjian seeing as we seem to get to decide :P
21:16:47 <jianjian> 10th of Jan works for me.
21:17:02 <jianjian> 😉
21:17:05 <mattoliver> I'll remind and bring it up again at next weeks meeting. If someone else wants to chair on the 3rd Jan then they until next week to let me know ;)
21:17:08 <mattoliver> nice!
21:17:33 <mattoliver> #topic Unit test failure under Python 3.12
21:17:42 <mattoliver> #link https://bugs.launchpad.net/swift/+bug/2046352
21:17:54 <jianjian> I just saw this https://review.opendev.org/c/openstack/governance/+/902585/12/goals/proposed/modernize-openstack-async-model.rst#14
21:18:30 <jianjian> "Indeed this library currently do not support Python 3.12 and face many issues with Python 3.11 (those are described below)."
21:18:46 <mattoliver> oh interesting
21:18:55 <jianjian> so eventlet needs work to support 3.11
21:18:58 <mattoliver> #link https://review.opendev.org/c/openstack/governance/+/902585/12/goals/proposed/modernize-openstack-async-model.rst#14
21:19:41 <jianjian> this is the link from @zaitcev yesterday
21:19:59 <mattoliver> ahh ok
21:20:21 <mattoliver> this also doesn't help the eventlet 3.11 problems:
21:20:27 <mattoliver> #link https://github.com/eventlet/eventlet/issues/824
21:20:37 <mattoliver> Basically they need to dev support.
21:20:51 <jianjian> unfortunately
21:20:59 <jianjian> yeah
21:21:42 <mattoliver> yeah. I think this is a dicussion that probably needs the whole team to work through.
21:21:54 <jianjian> I agree
21:21:59 <zaitcev> Primarily Tim I think
21:22:10 <zaitcev> He's been our interface to Eventlet for years.
21:22:26 <mattoliver> yeah
21:23:18 <mattoliver> I'll take a poke at the bug at least and see where Tim got with py312 support in python.. so attempt some kind of progress or at least familarization.
21:23:40 <mattoliver> I'll contact timburke and get some of his thoughts too.
21:23:47 <jianjian> great!
21:24:02 <mattoliver> Not sure what else to discuss on it now though.
21:25:03 <mattoliver> The next few topics are gleaned from what I saw in recent work in Swift via my emails. I'll do better a curating a proper list for next week.
21:25:56 <mattoliver> The next 2 topics I was hoping al or ash would be here to talk about.. so I'll bring them up for the minutes but there probably wont be anything to say or update about them.
21:26:07 <mattoliver> #topic s3acl and fakeswift
21:26:47 <mattoliver> I've seen a bunch of work on s3acl and fakeswift from Al and clay, so was curious about all this work.
21:27:05 <mattoliver> I assume its fall out from the s3api PartNum stuff, although maybe not
21:27:30 <mattoliver> chain seems to start
21:27:35 <mattoliver> #link https://review.opendev.org/c/openstack/swift/+/902564
21:27:52 <jianjian> ack
21:28:18 <mattoliver> When Al and Clay are going back a forwards on something I assume there is a reason ;)
21:28:26 <jianjian> and both patches are work in progress
21:28:31 <mattoliver> But I have no context, so moving on :P
21:28:33 <mattoliver> yeah
21:28:45 <mattoliver> #topic Partnum support
21:29:09 <mattoliver> I was hoping to get an update from ash on this
21:29:33 <mattoliver> But he doesn't seem to be around. So I'll drop a link and move on I guess
21:30:16 <mattoliver> chain starts:
21:30:25 <mattoliver> #link https://review.opendev.org/c/openstack/swift/+/894570
21:30:49 <jianjian> I heard from clay, seems this patch is getting closer to done.
21:30:49 <mattoliver> There are some patches at the end of the chain that are meant for squashing in, so things are still happening
21:30:57 <mattoliver> nice!
21:31:24 <mattoliver> well maybe next week we wont have to talk about it at all. and with the base at patchset 76, I bet Ash would be happy about that!
21:31:39 <mattoliver> #topic get_namespaces api
21:31:50 <mattoliver> This is the one I'm more interesting in :)
21:32:07 <jianjian> yess
21:32:26 <mattoliver> I've been stuck working on a Nvidia downstream project (related to swift) and just starting to get back to speed on this one
21:32:56 <jianjian> saw you started again to review them, thanks
21:33:19 <mattoliver> nps
21:33:21 <jianjian> we just got two prerequisite patches from al merged
21:33:32 <mattoliver> this feels like a good first patch in the chain:
21:33:37 <mattoliver> #link https://review.opendev.org/c/openstack/swift/+/902646
21:34:01 <mattoliver> yeah, looks like first 3 (in the link I just posted) have now been merged.
21:34:33 <mattoliver> jianjian: your container server patch and my proxy side patch seem to be pretty good too
21:35:01 <jianjian> I feel so, they are in good shape :-)
21:35:24 <mattoliver> I'm excited to see 900740 to finally get marker/end_marker and includes support
21:35:32 <jianjian> so next step will be, squash the container follow-up into the container patch, and squash proxy follow-up into your proxy patch
21:35:41 <mattoliver> kk
21:35:50 <mattoliver> or at least land them at the same time I guess
21:36:50 <mattoliver> I've almost finished reviewing 900740, just a wanted to run some vsaio testing of my own and run the probetests myself.
21:36:52 <jianjian> yes. maybe carry and deploy them internally at nvidia after each follow-up get +1
21:37:11 <jianjian> since we are already running the first container/proxy patches.
21:37:30 <mattoliver> But the really interesting one in the chain (for me) is
21:37:37 <mattoliver> #link https://review.opendev.org/c/openstack/swift/+/900350
21:38:01 <mattoliver> yeah, true.. I still feel a bit weird that we use nvidia prod as a preprod for upstream patches :P
21:38:26 <jianjian> thanks for testing 900740
21:38:28 <mattoliver> it's like burn in testing of patches :P
21:39:11 <jianjian> for 900350, I have tested in my vsaio, and ran those probe tests. all good
21:39:34 <mattoliver> 900350 reworks and refactors a bunch of the when and how we use caching of shardranges/namespaces
21:39:39 <mattoliver> oh great to know!
21:39:57 <mattoliver> cool. I'll move onto that one next too. but that gives me great confidence
21:40:10 <mattoliver> there is a lot of code churn in it, but I think it'
21:40:14 <jianjian> I ran probe tests, based the 900350 + your proxy patch on top of master branch, I wanted to check backward compatibility for upgrade purpose
21:40:17 <mattoliver> it's for the the better
21:40:59 <jianjian> only a few probe tests failed, because setup was wrong (checking shard-type to be Namespace)
21:41:14 <jianjian> after fixed, they all passed
21:41:23 <mattoliver> cool
21:41:27 <zaitcev> I wish you named it something other than namespace. But it's in a grand Swift tradition of devices that are not devices, partitions that are not partitions, and of course containers.
21:42:14 <mattoliver> ROFL zaitcev +100
21:42:37 <jianjian> ROFL
21:42:50 <jianjian> any suggestions, zaitcev?
21:43:20 <zaitcev> I thought they were more like segments or regions.
21:43:27 <zaitcev> Looking at what they do.
21:44:02 <zaitcev> But I kept quiet because I know nothing about their design and function.
21:44:04 <jianjian> we already have concept of segment or region, SLO segments, region for swift cluster
21:45:18 <jianjian> thanks for bringing it up though, we will keep working on this most difficult task in programming, naming!
21:45:40 <mattoliver> the proxy shard list building in the container controller is a little nuts and then we threw in caching, which made it even more complicated to follow (although it make things work much better at scale). Al has reworked the logic, found places the old code would do the wrong thing (although we never envoked, luckily these areas) and refactored to make it much more readable and easier to define when things happen.. so it's an amazing
21:45:40 <mattoliver> amount of work.. but needs some close review attention to make sure we haven't missed anything!
21:46:44 <zaitcev> In my mind, a namespace has something to do with a supra-object evolution. When you need two of the same names to exist, but they don't necessarily have an administrative division, that would make them tenants.
21:47:18 <zaitcev> It does not look like anything that happens with sharding.
21:47:30 <jianjian> I can also help on reviewing that shard list patch
21:47:40 <zaitcev> Anyhow, it's too late.
21:48:10 <mattoliver> thanks jianjian
21:48:35 <mattoliver> thats all I had so..
21:48:50 <mattoliver> #topic open discussion
21:48:59 <mattoliver> was there anything else to bring up?
21:49:02 <zaitcev> Matt
21:49:16 <zaitcev> Could you replace Al and just land the watcher patch?
21:49:24 <jianjian> Namespace is a simplified version of ShardRange (a sub-shard of sharded container), mainly for performance reasons. only has attributes name, lower and upper.
21:49:52 <zaitcev> https://review.opendev.org/c/openstack/swift/+/787656
21:49:59 <zaitcev> He'll never get to it
21:50:32 <zaitcev> And I'm afraid it will rot out of the tree. If we commit it, at least you two will remember to change the watcher for the next API change.
21:50:33 <mattoliver> zaitcev: yes! that damn thing is still around, I'll put it on my todo list before Christmas break
21:51:33 <mattoliver> still not landed I mean
21:52:01 <mattoliver> by around.. it should be landed. It makes darkdata better which already exists
21:52:52 <zaitcev> Yeah it's pretty safe for regressions. If you aren't running watchers, you'll never know it's there.
21:52:52 <mattoliver> sorry it's taken so long zaitcev
21:53:20 <zaitcev> Well, Tim invented it.
21:53:33 <zaitcev> I completely forgot about it when I introduced watchers.
21:56:10 <mattoliver> the patch could also grab namespace objects (sorry) instead of shards too. So yeah I good one to maybe try and sneak in before the get_namespace patches land.. although it's ok if it just asks for shardranges, that api isn't going anywhere :)
21:56:41 <mattoliver> might even be able to trick jianjian to take a look too seeing as he's becoming a sharding expert ;)
21:57:02 <mattoliver> anyway, we're almost at time and I have a 5 year old yelling for me
21:57:07 <mattoliver> So maybe I'll call it
21:57:14 <zaitcev> Take care.
21:57:24 <mattoliver> Thanks for coming and thanks for working on swift!
21:57:26 <jianjian> no problem, I will take a look at it
21:57:28 <mattoliver> #endmeeting