| *** gyee has quit IRC | 00:22 | |
| openstackgerrit | Merged openstack/swift master: Skip test when object versioning is not enabled https://review.opendev.org/682668 | 00:23 |
|---|---|---|
| *** NM has joined #openstack-swift | 00:31 | |
| openstackgerrit | Merged openstack/swift master: Close leaking opened requests https://review.opendev.org/682698 | 00:36 |
| *** NM has quit IRC | 00:39 | |
| *** BjoernT has joined #openstack-swift | 02:50 | |
| *** BjoernT_ has joined #openstack-swift | 02:55 | |
| *** BjoernT has quit IRC | 02:56 | |
| *** psachin has joined #openstack-swift | 03:34 | |
| *** pcaruana has joined #openstack-swift | 04:42 | |
| *** rcernin_ has joined #openstack-swift | 04:56 | |
| *** rcernin has quit IRC | 04:59 | |
| *** BjoernT_ has quit IRC | 05:45 | |
| *** early has quit IRC | 06:15 | |
| *** early has joined #openstack-swift | 06:18 | |
| *** rcernin_ has quit IRC | 06:48 | |
| *** takamatsu has quit IRC | 06:50 | |
| *** psachin has quit IRC | 06:53 | |
| *** e0ne has joined #openstack-swift | 07:17 | |
| *** tesseract has joined #openstack-swift | 07:26 | |
| *** takamatsu has joined #openstack-swift | 07:44 | |
| *** zaitcev__ has joined #openstack-swift | 08:01 | |
| *** ChanServ sets mode: +v zaitcev__ | 08:01 | |
| *** zaitcev_ has quit IRC | 08:05 | |
| *** tkajinam has quit IRC | 08:19 | |
| *** takamatsu_ has joined #openstack-swift | 08:46 | |
| *** takamatsu has quit IRC | 08:46 | |
| *** takamatsu_ has quit IRC | 09:19 | |
| *** takamatsu has joined #openstack-swift | 09:36 | |
| *** takamatsu has quit IRC | 10:00 | |
| *** takamatsu has joined #openstack-swift | 10:06 | |
| *** rcernin_ has joined #openstack-swift | 10:16 | |
| *** rcernin_ has quit IRC | 10:36 | |
| *** e0ne has quit IRC | 10:42 | |
| *** pcaruana has quit IRC | 10:54 | |
| *** takamatsu has quit IRC | 11:00 | |
| *** takamatsu has joined #openstack-swift | 11:06 | |
| *** pcaruana has joined #openstack-swift | 11:17 | |
| *** e0ne has joined #openstack-swift | 11:21 | |
| *** takamatsu has quit IRC | 11:22 | |
| *** takamatsu has joined #openstack-swift | 11:28 | |
| *** takamatsu has quit IRC | 12:21 | |
| *** takamatsu has joined #openstack-swift | 12:27 | |
| *** takamatsu has quit IRC | 12:38 | |
| *** takamatsu has joined #openstack-swift | 13:16 | |
| *** pcaruana has quit IRC | 13:28 | |
| *** NM has joined #openstack-swift | 13:34 | |
| *** pcaruana has joined #openstack-swift | 13:54 | |
| *** BjoernT has joined #openstack-swift | 13:59 | |
| *** tesseract has quit IRC | 14:00 | |
| *** BjoernT_ has joined #openstack-swift | 14:02 | |
| *** BjoernT has quit IRC | 14:04 | |
| *** takamatsu has quit IRC | 14:16 | |
| *** e0ne has quit IRC | 14:28 | |
| *** pcaruana has quit IRC | 14:33 | |
| NM | clayg: thanks for the answer. Besides authentication, should I worry about other types of cache ? Let's say memcache-region-a and memcache-region-b have the same content about an specific container. Then proxy from region A updates the container and then updates memcache-region-a. Will the proxy from region B keep getting the outdated content from memcache-region-b? | 14:48 |
| DHE | yes, but that's just an inherent risk in how swift works with its eventually-consistent behaviour. I believe those memcache timeouts are somewhat short anyway | 14:49 |
| clayg | right cache invalidation is a little weak - so if you have two clients form different regions who both want to coordinate on consistent changes from different regions there's going to be slightly more "eventual consistency" that they may have otherwise come to expect | 14:50 |
| clayg | DHE: agree memcache timeouts are pretty short, ~30s range | 14:50 |
| clayg | but you don't need to "worry about it" - it's the job of the consistency engine to "worry about it" | 14:51 |
| NM | That was really helpfull clayg and DHE! Thank you! | 14:52 |
| clayg | it'd be more like client in region 1 is writing objects into a versioned container and region 2 turns off versioning - a couple more writes might get versioned before region 1 realizes it can start doing normal overwrites | 14:52 |
| clayg | but a lot of that kind of metadata (which on update normally invalidates the cache) is set once and forget it | 14:52 |
| clayg | if you have clients turning stuff on and off while other actors are trying to do things based off that - and the clients are trying to use swift to coordinate atomic state transistions - they're sort of already in for a bad time | 14:53 |
| clayg | but maybe they've designed for that and "mostly works" is good enough... | 14:53 |
| *** gyee has joined #openstack-swift | 14:56 | |
| NM | Never thought about that. Do we have any doc about memcache timeouts and how swift uses memcache? I couldn't find any. | 14:57 |
| *** e0ne has joined #openstack-swift | 15:02 | |
| tdasilva | I'm not sure I understand this statement from s3: "Only Amazon S3 can create a delete marker, and it does so whenever you send a DELETE Object request on an object in a versioning-enabled or suspended bucket" Does that mean they keep "versioning" if the "overwrite" is actually a DELETE request when versioning is suspended ?? | 15:11 |
| clayg | so is p 681970 dependent on the continue on empty shard range patch? | 15:15 |
| patchbot | https://review.opendev.org/#/c/681970/ - swift - Sharding: Clean up old CleaveConext's during audit - 2 patch sets | 15:15 |
| clayg | tdasilva: where'd you read that? i wouldn't expect delete markers to be created in versioning suspended mode? | 15:16 |
| clayg | ... but if that's what s3 does it would be good to know that even if it's surprising | 15:17 |
| tdasilva | clayg: ah so this is interesting, i just verified the behavior. the delete markerj just overwrites whatever is latest. | 15:20 |
| clayg | what?! really? thats... weird as hell | 15:20 |
| tdasilva | if current is something that is "versioned", then of-course it won't overwrite that | 15:20 |
| clayg | rly? i guess that makes sense | 15:21 |
| tdasilva | but if the current was written down while it was in suspend mode, then it will write and delete marker that overwrites the current | 15:21 |
| tdasilva | s/write and delete/write a delete | 15:21 |
| tdasilva | make sense? | 15:22 |
| tdasilva | read it here: https://docs.aws.amazon.com/AmazonS3/latest/dev/DeleteMarker.html | 15:22 |
| tdasilva | clayg: my concern if we want to mirror that behavior is how it would mess up object count on the container, because we would need to always write down a delete marker | 15:23 |
| tdasilva | and just cleaning up the container in general...shoot! | 15:24 |
| tdasilva | i guess we could write down delete marker but not write a symlink to it?? | 15:24 |
| clayg | yeah i guess what happens in suspended mode looks basically the same as what happens on delete during enabled? | 15:28 |
| clayg | right? we write the delete marker, delete the symlink - but we don't issue a delete for the versioned data | 15:29 |
| tdasilva | i'm not really sure I understand WHY they implemented this way, seems weird | 15:38 |
| clayg | well the alternative is to delete the versioned object... I'm guessing DELETE while suspended treats "objects that were written while versioning was enabled" different from "objects that were written while versioning was suspended" | 15:40 |
| *** NM has quit IRC | 15:47 | |
| tdasilva | clayg: correct, which is actually easy for us to do too. because ""objects that were written while versioning was enabled" are placed in versions container, while "objects that were written while versioning was suspended" are written to actual uses container, so while we are in suspended mode, we will always issue a DELETE to users container, whatever is in there gets blow-away | 15:52 |
| tdasilva | the only part that sucks, is that now on a "overwrite" while is suspended mode, we will still need a look into the versions container to see if latest is a "null" versioned delete marker. | 15:53 |
| *** openstackgerrit has quit IRC | 16:06 | |
| *** takamatsu has joined #openstack-swift | 16:06 | |
| *** e0ne has quit IRC | 16:14 | |
| timburke | or we cook up a fake entry when you go to do the listing and see a bunch of versions for the object but no link in the primary container | 16:30 |
| timburke | ...or do we need to track a last-modified for that delete? | 16:31 |
| *** itlinux has joined #openstack-swift | 16:35 | |
| timburke | clayg, p 681970 and p 675820 are mostly independent. i suppose there's a slight preference that we address the context-reaping before making handoffs go faster -- i think moving through handoffs quickly would exacerbate the metadata growth | 16:36 |
| patchbot | https://review.opendev.org/#/c/681970/ - swift - Sharding: Clean up old CleaveConext's during audit - 2 patch sets | 16:36 |
| patchbot | https://review.opendev.org/#/c/675820/ - swift - sharder: Keep cleaving on empty shard ranges - 4 patch sets | 16:36 |
| *** e0ne has joined #openstack-swift | 16:45 | |
| *** openstackgerrit has joined #openstack-swift | 16:54 | |
| openstackgerrit | Tim Burke proposed openstack/swift stable/stein: bufferedhttp: ensure query params are properly quoted https://review.opendev.org/682112 | 16:54 |
| openstackgerrit | Tim Burke proposed openstack/swift stable/stein: py2/3: Stop using stdlib's putrequest(); it only does ASCII https://review.opendev.org/683204 | 16:54 |
| openstackgerrit | Tim Burke proposed openstack/swift stable/pike: bufferedhttp: ensure query params are properly quoted https://review.opendev.org/681879 | 17:00 |
| openstackgerrit | Tim Burke proposed openstack/swift stable/pike: py2/3: Stop using stdlib's putrequest(); it only does ASCII https://review.opendev.org/683206 | 17:00 |
| *** pcaruana has joined #openstack-swift | 17:01 | |
| *** pcaruana has quit IRC | 17:14 | |
| *** gmann_afk is now known as gmann | 17:21 | |
| *** pcaruana has joined #openstack-swift | 17:25 | |
| *** NM has joined #openstack-swift | 17:29 | |
| openstackgerrit | Clay Gerrard proposed openstack/swift master: Allow internal clients to use null namespace https://review.opendev.org/682138 | 17:53 |
| *** pcaruana has quit IRC | 17:55 | |
| *** mvkr has quit IRC | 18:48 | |
| *** pcaruana has joined #openstack-swift | 19:51 | |
| *** NM has quit IRC | 20:13 | |
| *** NM has joined #openstack-swift | 20:24 | |
| *** pcaruana has quit IRC | 20:31 | |
| *** e0ne has quit IRC | 21:00 | |
| *** zaitcev__ is now known as zaitcev | 21:08 | |
| *** e0ne has joined #openstack-swift | 21:12 | |
| *** NM has quit IRC | 21:15 | |
| *** benj_ has quit IRC | 21:31 | |
| *** benj has joined #openstack-swift | 21:31 | |
| *** benj is now known as Guest29634 | 21:32 | |
| *** BjoernT_ has quit IRC | 22:00 | |
| *** e0ne has quit IRC | 22:24 | |
| openstackgerrit | Merged openstack/swift master: [train][goal] Run 'tempest-ipv6-only' job in gate https://review.opendev.org/682539 | 22:29 |
| *** notmyname has quit IRC | 22:30 | |
| timburke | wowza! head of the check queue is coming up on *18 hours*! | 22:31 |
| *** notmyname has joined #openstack-swift | 22:31 | |
| *** ChanServ sets mode: +v notmyname | 22:31 | |
| DHE | is that high or low? | 22:48 |
| timburke | high -- and really painful as a dev. means nothing i do today will have CI votes on it until at least tomorrow :-( | 22:49 |
| *** tkajinam has joined #openstack-swift | 23:02 | |
| *** rcernin has joined #openstack-swift | 23:25 | |
| *** e0ne has joined #openstack-swift | 23:30 | |
| *** e0ne has quit IRC | 23:30 | |
| openstackgerrit | Tim Burke proposed openstack/swift master: sharding: Update probe test to verify CleavingContext cleanup https://review.opendev.org/683257 | 23:54 |
| *** e0ne has joined #openstack-swift | 23:55 | |
| *** e0ne has quit IRC | 23:55 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!