07:01:32 <mattoliver> #startmeeting swift
07:01:32 <opendevmeet> Meeting started Wed Jul 16 07:01:32 2025 UTC and is due to finish in 60 minutes.  The chair is mattoliver. Information about MeetBot at http://wiki.debian.org/MeetBot.
07:01:32 <opendevmeet> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
07:01:32 <opendevmeet> The meeting name has been set to 'swift'
07:01:41 <mattoliver> whose here for the swift meeting
07:01:55 <mattoliver> As always the agenda is at..
07:02:02 <mattoliver> #link https://wiki.openstack.org/wiki/Meetings/Swift
07:02:42 <mattoliver> I was away last week on vacation so also haven't done much Swift stuff, so mostly info and a checkin with cschwede :)
07:02:50 <mattoliver> #topic  Bug triage
07:03:09 <mattoliver> I've seen a bunch of bug emails
07:03:18 <mattoliver> So you've been busy cschwede  :)
07:03:40 <cschwede> Jupp, got back to some triaging earlier today. And rabbit holes! Was checking some bugs and code and this took way more time than expected...
07:04:01 <cschwede> Not sure if this strategy is sustainable.
07:04:48 <cschwede> During the triage I stumbled upon https://bugs.launchpad.net/swift/+bug/2058945 - this looks annoying, and looking at the linked issue it seems still not resolved. Anybody else seen this?
07:05:01 <mattoliver> yeah, dont push it too hard. Maybe we should all be taking a peice of the work.
07:06:19 <mattoliver> oh thats an interesting one. We have been having trouble in the past with high accep queues which starts to give timeouts to clients.
07:06:29 <mattoliver> maybe somewhat related.
07:07:40 <mattoliver> a bad client if they're doing pipelined requests slowly can also hold up a max_clients slot
07:08:30 <mattoliver> I'll repond to the bug with what we ran into in the hopes its similar and we can lock something down.
07:09:03 <cschwede> Thanks, that would be great!
07:09:46 <mattoliver> Checking the proxy accept queues is good practice when proxies or client requests start timing out.. I really should get the SO_TIMESPAMPING patch finished/landed :)
07:10:36 <mattoliver> Keep up the good work cschwede and feel free to take a break from it if required. We should all probably be working through this backlog
07:10:53 <mattoliver> #topic eventlet removal POC update
07:11:07 <mattoliver> cschwede: I assume you've been stuck down other rabbit holes :P
07:11:53 <mattoliver> Eventlet removal is a topic that comes up a bit here at Nvidia.
07:12:08 <cschwede> Yes, but still making slow progress. Though I also started to look into the object-server - we chatted about starting with the object-server first at the last PTG as well
07:12:26 <mattoliver> kk
07:13:03 <mattoliver> well if there is anything we can help with then let us know
07:13:23 <mattoliver> I might need to speed up, getting looks from the family :P
07:13:36 <cschwede> sure, will do!
07:13:43 <mattoliver> #topic October vPTG
07:14:12 <mattoliver> Just a reminder that another virtual PTG is coming up in Oct
07:14:29 <mattoliver> Oct 27-31
07:14:38 <mattoliver> #link https://ptg.openinfra.org/
07:15:02 <mattoliver> timburke or I will make sure there is an etherpad to start gathering topics.
07:15:37 <mattoliver> Already another one coming means I need to get my butt into gear and work on some of my projects :)
07:15:51 <mattoliver> #topic aws-chunked
07:16:15 <mattoliver> We need to make sure more of this chain is landed BEFORE the next upstream release
07:16:39 <mattoliver> Because we don't want to release a half baked (even though it works) solution.
07:17:04 <mattoliver> We should at least get the crc64-nvme stuff landed
07:17:20 <mattoliver> #link https://review.opendev.org/c/openstack/swift/+/946141
07:17:38 <mattoliver> jianjian has been reviewing and looking at this chain, so thanks!
07:17:52 <mattoliver> I need to dig back in too. Esp now that I'm back from vacation.
07:18:10 <mattoliver> #topic 2026.1 codename: Gazpacho
07:18:23 <mattoliver> Speaking on the next upstream release, we have a name for it
07:18:38 <mattoliver> #link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/message/QVZK5PFN233WNDX764B2NZ4W65U27FF4/
07:19:05 <mattoliver> Not much else to add there..
07:19:18 <mattoliver> #topic New ISA-L backend for liberasurecode
07:19:25 <mattoliver> This one is interesting
07:19:34 <mattoliver> #link https://review.opendev.org/c/openstack/liberasurecode/+/954285
07:20:00 <mattoliver> Similar to isa_l_rs_vand / isa_l_rs_cauchy split, the only difference is the encoding matrix (which is then used to compute the wider encoding tables)
07:20:00 <mattoliver> Related to Vandermonde matrices, but should avoid non-invertiblity
07:20:18 <cschwede> Interesting!
07:20:33 <mattoliver> The bug seems to indicate the isa_l_rs_vand can have issues with parties > 5
07:21:12 <mattoliver> So I think once this lands in liberasurecode we probably should add some kind of warning in swift if people use the current isa_l_rs_vand with > 5 parity
07:21:50 <mattoliver> Let me paste more of Tims awesome discriptions from the agenda
07:21:58 <cschwede> Isnt' that already the case, mentioned in https://bugs.launchpad.net/liberasurecode/+bug/1639691
07:21:59 <mattoliver> Related to Vandermonde matrices, but should avoid non-invertiblity problem from https://launchpad.net/bugs/1639691
07:22:11 <mattoliver> #link https://launchpad.net/bugs/1639691
07:22:12 <cschwede> https://review.opendev.org/c/openstack/swift/+/468105
07:22:28 * cschwede should use #link more often.
07:22:41 <mattoliver> oh nice
07:23:00 <mattoliver> so already done.. and I+2ed it.. I fail at memory :P
07:23:23 <mattoliver> Tim says: Seems to match encoding matrix used elsewhere, eg https://github.com/tahoe-lafs/zfec/blob/zfec-1.6.0.0/zfec/fec.c#L430-L479
07:23:33 <mattoliver> #link https://github.com/tahoe-lafs/zfec/blob/zfec-1.6.0.0/zfec/fec.c#L430-L479
07:24:08 <mattoliver> But the pressing question, other then testing and landing it is what name we should give it?
07:24:21 <mattoliver> isa_l_rs_vand_inv? isa_l_rs_vand2?
07:24:57 <mattoliver> if its backward compat with < 5 parity maybe replace.. but that might not be possible.
07:25:14 <mattoliver> _inv might not mean much to people. maybe 2 is better.
07:25:33 <mattoliver> Anything feel free to weigh in in the patch :)
07:25:38 <mattoliver> That's all I got
07:25:49 <mattoliver> #topic open discussion
07:26:33 <mattoliver> I've still got a bunch of patches in the air.. but need to revisit them to see where they're at and I'm in a rush tonight.. so I got nothing else :P
07:28:12 <cschwede> Nothing else from my side as well today. Thanks Matt for running this!
07:28:30 <mattoliver> Cool, nps.
07:28:46 <mattoliver> I'll call it then. Thanks for coming and thanks for working on Swift
07:29:00 <mattoliver> cschwede: so glad you catch up with you more regularly now!
07:29:04 <mattoliver> #endmeeting