| *** zhill has quit IRC | 00:00 | |
| *** noye has quit IRC | 00:02 | |
| *** wshao has quit IRC | 00:13 | |
| *** petertr7 has quit IRC | 00:21 | |
| clayg | torgomatic: I'm curious what you might think about adding some TBD sections the symlink spec? Like "we acknowledge this should be discussed in the spec - but rational minds may hold differring opinions - so we explictly do not attempt to resolve them in this revision of the spec in order to find a minimal set of concensus that we can merge" - or something like that? | 00:21 |
|---|---|---|
| *** jkugel has joined #openstack-swift | 00:21 | |
| clayg | acoles_away: oh crap - i guess we should get on reviewing fast-POST?! | 00:22 |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Better storage server eventlet logging https://review.openstack.org/173622 | 00:25 |
| ho | good morning! | 00:31 |
| *** dmorita has joined #openstack-swift | 00:35 | |
| mattoliverau | ho: morning | 00:40 |
| ho | mattoliverau: morning! | 00:42 |
| mattoliverau | clayg: specs are suppose to be alive, have many iterations and exist so they can be discussed.. So in my opinion TBD's are fine.. So long as there are none in the final (completed) version. In fact TBD can be used as good points to discuss at summit (my 2 cents) | 00:44 |
| *** thumpba has quit IRC | 00:47 | |
| *** peluse has quit IRC | 00:53 | |
| *** kota_ has joined #openstack-swift | 00:57 | |
| kota_ | morning | 00:58 |
| ho | kota_: good morning! | 00:59 |
| *** fanyaohong has quit IRC | 00:59 | |
| kota_ | ho: good morning :) | 00:59 |
| mattoliverau | kota_: morning | 01:01 |
| kota_ | mattoliverau: morning :) | 01:02 |
| kota_ | yeah, I'm happy to see at commit log that the EC staff has been merged to the master branch :D | 01:06 |
| kota_ | great work! | 01:06 |
| openstackgerrit | Merged openstack/swift: Fix broken test_policy_IO_override test https://review.openstack.org/171593 | 01:18 |
| openstackgerrit | Merged openstack/swift: Add lowest option to swift-recon disk usage output https://review.openstack.org/167236 | 01:18 |
| openstackgerrit | Merged openstack/swift: Add test for swift_recon.disk_usage https://review.openstack.org/172752 | 01:22 |
| openstackgerrit | Merged openstack/swift: Check if REST API version is valid https://review.openstack.org/168509 | 01:22 |
| openstackgerrit | Merged openstack/swift: 2.3.0 authors and changelog updates https://review.openstack.org/172573 | 01:22 |
| openstackgerrit | Merged openstack/swift: fixup random test failure https://review.openstack.org/173579 | 01:22 |
| *** esker has joined #openstack-swift | 01:24 | |
| *** MVenesio has quit IRC | 01:31 | |
| notmyname | everything has merged! | 01:31 |
| notmyname | SHA for the RC, and will be the 2.3.0 release unless anything major is found, is 5ef2e9ec00b1498fb411fc918b71ae9c50b1308f (ie current HEAD of master) | 01:33 |
| notmyname | master is now unfrozen | 01:33 |
| *** ChanServ changes topic to "Review Dashboard: http://goo.gl/vysJqI | Summary Dashboard: http://goo.gl/2By1qv | Summit scheduling: https://etherpad.openstack.org/p/liberty-swift-summit-topics | Logs: http://eavesdrop.openstack.org/irclogs/%23openstack-swift/" | 01:33 | |
| notmyname | great work, everyone! | 01:33 |
| notmyname | thanks | 01:33 |
| * notmyname logs off for the night | 01:37 | |
| *** gyee has quit IRC | 02:02 | |
| *** occupant has quit IRC | 02:06 | |
| *** Guest__ has joined #openstack-swift | 02:12 | |
| *** km has joined #openstack-swift | 02:21 | |
| *** Guest__ has quit IRC | 02:24 | |
| *** Guest__ has joined #openstack-swift | 02:24 | |
| *** esker has quit IRC | 02:57 | |
| *** esker has joined #openstack-swift | 02:57 | |
| *** Guest__ has quit IRC | 02:59 | |
| zaitcev | What a day | 03:00 |
| zaitcev | Picking up pieces from the EC merge, and just got PyECLib building now, barely. | 03:00 |
| zaitcev | OSX and Ubuntu people forgot to support /usr/lib64 | 03:01 |
| *** jrichli has quit IRC | 03:09 | |
| mattoliverau | zaitcev: I guess I need to get liberasurecode compiling and working on OSX now if I want to do any work on it, did you have any trouble doing that? (I guess I shold just go try) | 03:11 |
| zaitcev | mattoliverau: I'm on Fedora, so no, I didn't. Only made an RPM for Fedora review - https://bugzilla.redhat.com/show_bug.cgi?id=1208695 | 03:15 |
| openstack | bugzilla.redhat.com bug 1208695 in Package Review "Review Request: liberasurecode - Erasure Code API library written in C with pluggable backends" [Medium,Assigned] - Assigned to ppisar | 03:16 |
| mattoliverau | zaitcev: fair enough, nice work! Maybe I just need to convince work to give me dev laptop I can just run linux on, then I don't have to worry about OSX nuances ;) | 03:19 |
| zaitcev | I gave up on work laptops. We have Lenovos now and I'm not friendly with the left Ctrl inside the Fn. My pinkie hurts on that. | 03:23 |
| mattoliverau | lol | 03:28 |
| *** km has quit IRC | 03:40 | |
| zaitcev | ERROR 0001: file '/usr/lib64/python2.7/site-packages/pyeclib_c.so' contains a standard rpath '/usr/lib' in [/usr/lib/python2.7/site-packages:/usr/lib:/lib:/usr/local/lib:src/c/liberasurecode-1.0.6/src/.libs] | 03:54 |
| zaitcev | Kevin, Kevin | 03:54 |
| *** kota_ has quit IRC | 04:04 | |
| *** krtaylor has quit IRC | 04:07 | |
| *** krtaylor has joined #openstack-swift | 04:10 | |
| *** peluse has joined #openstack-swift | 04:13 | |
| *** ChanServ sets mode: +v peluse | 04:13 | |
| *** zaitcev has quit IRC | 04:24 | |
| *** SkyRocknRoll has joined #openstack-swift | 05:23 | |
| *** resker has joined #openstack-swift | 05:32 | |
| *** esker has quit IRC | 05:35 | |
| *** kota_ has joined #openstack-swift | 05:48 | |
| *** ppai has joined #openstack-swift | 05:51 | |
| *** bkopilov has quit IRC | 05:53 | |
| *** bkopilov has joined #openstack-swift | 05:59 | |
| *** silor has joined #openstack-swift | 06:07 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift: Imported Translations from Transifex https://review.openstack.org/173687 | 06:07 |
| *** ppai has quit IRC | 06:07 | |
| *** ppai has joined #openstack-swift | 06:08 | |
| *** SkyRocknRoll has quit IRC | 06:18 | |
| *** SkyRocknRoll has joined #openstack-swift | 06:21 | |
| *** bkopilov has quit IRC | 06:24 | |
| openstackgerrit | Kota Tsuyuzaki proposed openstack/swift: Small minor refactor on ec diskfile https://review.openstack.org/173691 | 06:25 |
| openstackgerrit | Samuel Merritt proposed openstack/swift-specs: Spec for symlinks in Swift https://review.openstack.org/173609 | 06:28 |
| cschwede | Good Morning! | 06:31 |
| openstackgerrit | Christian Schwede proposed openstack/swift: More tests for swift recon https://review.openstack.org/173266 | 06:37 |
| kota_ | cschwede: good morning, Christian! | 06:37 |
| mattoliverau | cschwede: morning | 06:38 |
| cschwede | Hello Kota, Hello Matthew! | 06:38 |
| *** SkyRocknRoll has quit IRC | 06:39 | |
| *** bkopilov has joined #openstack-swift | 06:50 | |
| *** km has joined #openstack-swift | 06:50 | |
| *** bkopilov has quit IRC | 06:56 | |
| *** peluse has quit IRC | 06:56 | |
| *** joeljwright has joined #openstack-swift | 06:56 | |
| *** mmcardle has joined #openstack-swift | 06:58 | |
| *** ho has quit IRC | 07:00 | |
| *** nshaikh has joined #openstack-swift | 07:00 | |
| *** jeblair has quit IRC | 07:05 | |
| *** peluse has joined #openstack-swift | 07:06 | |
| *** peluse has quit IRC | 07:06 | |
| *** jeblair has joined #openstack-swift | 07:07 | |
| *** bkopilov has joined #openstack-swift | 07:11 | |
| *** jordanP has joined #openstack-swift | 07:13 | |
| *** jistr has joined #openstack-swift | 07:24 | |
| *** SkyRocknRoll has joined #openstack-swift | 07:26 | |
| *** krykowski has joined #openstack-swift | 07:33 | |
| *** chlong has quit IRC | 07:35 | |
| *** ppai has quit IRC | 07:35 | |
| *** bkopilov has quit IRC | 07:45 | |
| *** ppai has joined #openstack-swift | 07:49 | |
| *** bkopilov has joined #openstack-swift | 07:49 | |
| openstackgerrit | Andreas Jaeger proposed openstack/swift: Fix untranslatable string https://review.openstack.org/173713 | 07:52 |
| *** bkopilov has quit IRC | 07:54 | |
| *** bkopilov has joined #openstack-swift | 07:54 | |
| *** geaaru has joined #openstack-swift | 07:55 | |
| *** peluse has joined #openstack-swift | 07:55 | |
| *** ktsuyuzaki has joined #openstack-swift | 07:58 | |
| *** ktsuyuzaki has quit IRC | 07:58 | |
| *** kota_ has quit IRC | 07:59 | |
| *** peluse has quit IRC | 08:01 | |
| *** bkopilov has quit IRC | 08:10 | |
| *** bkopilov has joined #openstack-swift | 08:17 | |
| *** ujjain has joined #openstack-swift | 08:20 | |
| *** km has quit IRC | 08:35 | |
| openstackgerrit | Kamil Rykowski proposed openstack/swift: More user-friendly output for object metadata https://review.openstack.org/164019 | 08:37 |
| *** bkopilov has quit IRC | 08:39 | |
| *** bkopilov has joined #openstack-swift | 08:45 | |
| *** AliRezaTaleghani has joined #openstack-swift | 08:46 | |
| AliRezaTaleghani | I would like to add some specific feature on my swift cluster, the case is as follow: | 08:47 |
| AliRezaTaleghani | I will push some large acale pic on swift after some while one other will try to download that pic name | 08:48 |
| AliRezaTaleghani | but will smaller size | 08:48 |
| AliRezaTaleghani | ex: original file name is "swift-obj-image_1280x1024.png" | 08:49 |
| *** acoles_away is now known as acoles | 08:49 | |
| AliRezaTaleghani | requested file name is "swift-obj-image_640x480.png" | 08:49 |
| AliRezaTaleghani | is such a case I would like to have a middleware to convert the object on the fly and deliver it | 08:50 |
| AliRezaTaleghani | I'm new to swift and don't have more exprience with python but I'm looking for point to start work on | 08:51 |
| *** bkopilov has quit IRC | 08:51 | |
| AliRezaTaleghani | please let me know if I am going wrong with middleware for this porpose | 08:51 |
| AliRezaTaleghani | plz | 08:51 |
| *** bkopilov has joined #openstack-swift | 08:52 | |
| *** ppai has quit IRC | 08:53 | |
| ahale | yeah sounds like you need to write some middleware, I did something similar a long time ago for a very old version of swift, https://github.com/ahale/swascii/blob/master/swascii/middleware.py | 08:53 |
| ahale | thats kind amessy and probably needs updated for newer swift, but might help maybe, i dunno | 08:54 |
| AliRezaTaleghani | ahale: tnx i will chk it r-now | 08:57 |
| *** krtaylor has quit IRC | 08:57 | |
| *** bkopilov has quit IRC | 08:57 | |
| ahale | nps, I think I started by copying a real swift middleware and hacking at it until it did what I wanted, they're pretty straightforward though | 08:59 |
| AliRezaTaleghani | ahale: :-) 1+ so will start alike | 09:01 |
| ahale | :) | 09:02 |
| *** krtaylor has joined #openstack-swift | 09:04 | |
| *** ppai has joined #openstack-swift | 09:05 | |
| *** dmorita has quit IRC | 09:07 | |
| *** nshaikh has left #openstack-swift | 09:31 | |
| *** okdas has joined #openstack-swift | 09:44 | |
| *** ppai has quit IRC | 09:48 | |
| *** jamielennox is now known as jamielennox|away | 09:52 | |
| *** nshaikh has joined #openstack-swift | 09:59 | |
| *** ppai has joined #openstack-swift | 10:01 | |
| *** theanalyst has quit IRC | 10:06 | |
| *** theanalyst has joined #openstack-swift | 10:09 | |
| *** kota_ has joined #openstack-swift | 10:18 | |
| openstackgerrit | Alistair Coles proposed openstack/swift: Fix untranslatable string https://review.openstack.org/173713 | 10:32 |
| *** AliRezaTaleghani has left #openstack-swift | 10:51 | |
| openstackgerrit | Lorcan Browne proposed openstack/swift: Update swift-recon doc with more options https://review.openstack.org/173767 | 10:55 |
| *** ujjain has quit IRC | 10:59 | |
| *** bkopilov has joined #openstack-swift | 11:07 | |
| *** ppai has quit IRC | 11:10 | |
| *** acoles is now known as acoles_away | 11:13 | |
| *** ppai has joined #openstack-swift | 11:22 | |
| *** ppai has quit IRC | 11:45 | |
| *** ppai has joined #openstack-swift | 11:46 | |
| *** aix has quit IRC | 11:50 | |
| *** dencaval has joined #openstack-swift | 11:51 | |
| *** mmcardle has quit IRC | 11:56 | |
| *** AJaeger has joined #openstack-swift | 12:00 | |
| AJaeger | Thanks a lot swift team for a really speed review and approval of https://review.openstack.org/173713 ! | 12:00 |
| *** MVenesio has joined #openstack-swift | 12:13 | |
| *** ppai has quit IRC | 12:21 | |
| *** aix has joined #openstack-swift | 12:28 | |
| *** jkugel has quit IRC | 12:30 | |
| *** SkyRocknRoll has quit IRC | 12:33 | |
| *** mmcardle has joined #openstack-swift | 12:36 | |
| *** nshaikh has left #openstack-swift | 12:40 | |
| *** acoles_away is now known as acoles | 12:40 | |
| openstackgerrit | Merged openstack/swift: More tests for swift recon https://review.openstack.org/173266 | 12:46 |
| *** SkyRocknRoll has joined #openstack-swift | 12:47 | |
| openstackgerrit | Merged openstack/swift: Fix untranslatable string https://review.openstack.org/173713 | 12:47 |
| *** kota_ has quit IRC | 12:49 | |
| *** annegentle has joined #openstack-swift | 12:50 | |
| openstackgerrit | Lorcan Browne proposed openstack/swift: Update swift-recon doc with more options https://review.openstack.org/173767 | 12:53 |
| *** acoles is now known as acoles_away | 12:56 | |
| *** AJaeger has left #openstack-swift | 12:56 | |
| *** resker has quit IRC | 12:58 | |
| *** peluse_ has joined #openstack-swift | 13:01 | |
| *** proteusguy has quit IRC | 13:03 | |
| *** peluse_ is now known as peluse | 13:06 | |
| *** ChanServ sets mode: +v peluse | 13:06 | |
| *** acoles_away is now known as acoles | 13:12 | |
| *** MVenesio has quit IRC | 13:23 | |
| *** david-lyle has quit IRC | 13:24 | |
| *** jkugel has joined #openstack-swift | 13:24 | |
| *** SkyRocknRoll has quit IRC | 13:35 | |
| *** ozialien has joined #openstack-swift | 13:39 | |
| *** proteusguy has joined #openstack-swift | 13:41 | |
| *** Gue______ has joined #openstack-swift | 13:50 | |
| *** SkyRocknRoll has joined #openstack-swift | 13:51 | |
| *** noye has joined #openstack-swift | 13:53 | |
| peluse | acoles, thanks for the trello updates. | 14:01 |
| acoles | peluse: np | 14:02 |
| peluse | all, trello still has misc EC todo items that weren't considered required for beta and is now up to date at https://trello.com/b/LlvIFIQs/swift-erasure-codes | 14:02 |
| peluse | kota, thanks for confirming that pyeclib item also! | 14:03 |
| *** esker has joined #openstack-swift | 14:06 | |
| *** zaitcev has joined #openstack-swift | 14:07 | |
| *** ChanServ sets mode: +v zaitcev | 14:07 | |
| *** ujjain has joined #openstack-swift | 14:08 | |
| *** SkyRocknRoll has quit IRC | 14:15 | |
| *** ozialien has quit IRC | 14:17 | |
| *** ozialien has joined #openstack-swift | 14:19 | |
| *** ozialien has quit IRC | 14:26 | |
| *** SkyRocknRoll has joined #openstack-swift | 14:31 | |
| *** Fin1te has joined #openstack-swift | 14:31 | |
| *** Gue______ has quit IRC | 14:36 | |
| *** ahale has left #openstack-swift | 14:42 | |
| *** jrichli has joined #openstack-swift | 14:50 | |
| *** Fin1te has quit IRC | 14:53 | |
| *** annegentle has quit IRC | 14:54 | |
| *** jordanP has quit IRC | 14:55 | |
| *** Gu_______ has joined #openstack-swift | 15:01 | |
| notmyname | good morning | 15:01 |
| *** jdaggett has left #openstack-swift | 15:02 | |
| acoles | notmyname: good morning. enjoyed your podcast. | 15:02 |
| notmyname | acoles: thanks! (I'm glad I didn't do something terrible and forget your name!) | 15:03 |
| acoles | notmyname: heh. | 15:03 |
| notmyname | I was walking around the world in my head, and was like "I'm missing someone in EU..." | 15:03 |
| acoles | notmyname: well you *nearly* did :D | 15:03 |
| notmyname | lol | 15:03 |
| notmyname | I know. I feel bad ;-) | 15:03 |
| acoles | don't! | 15:03 |
| notmyname | :-) | 15:04 |
| acoles | you were very gracious and complimentary about the whole community which was great | 15:04 |
| notmyname | a fraction of what is deserved | 15:04 |
| peluse | morning | 15:05 |
| peluse | hey, where's the pdcast link? | 15:05 |
| notmyname | cool. looks like we have an upstream 2.3.0rc1 tag | 15:05 |
| notmyname | peluse: https://www.youtube.com/watch?v=1ORJKia1pQs | 15:05 |
| notmyname | so today I'll be taking it easy. team meeting today will be "good work on EC. let's talk summit scheduling" | 15:08 |
| notmyname | acoles: cschwede: I wouldn't expect you to stay up late for it today :-) | 15:09 |
| *** Gu_______ has quit IRC | 15:12 | |
| acoles | notmyname: will most likely be there | 15:13 |
| notmyname | ok :-) | 15:13 |
| *** SkyRocknRoll has quit IRC | 15:15 | |
| *** Gu_______ has joined #openstack-swift | 15:18 | |
| cschwede | notmyname: good morning! i’ll attend the meeting too | 15:21 |
| cschwede | notmyname: acoles: yeah, i liked the praise for the community in the podcast too. well done! | 15:21 |
| notmyname | well ok then ;-) | 15:21 |
| *** ozialien has joined #openstack-swift | 15:26 | |
| peluse | yup, great video! oddly enough watching at 2x speed (for my ADD), John, you don't sound that much different :) | 15:36 |
| acoles | lol | 15:37 |
| notmyname | heh | 15:37 |
| clayg | peluse: how do you watch videos at 2x? I hate video because it's so slow! | 15:41 |
| peluse | the little cog wheel icon in the lower right | 15:41 |
| peluse | I watch everything at 2x :) | 15:41 |
| clayg | peluse: this changes *everything* | 15:42 |
| peluse | yeah, its like... almost efficient! | 15:42 |
| peluse | i wouldn't suggest watching your porn at 2x though :) | 15:42 |
| clayg | noted | 15:43 |
| *** fanyaohong has joined #openstack-swift | 15:43 | |
| acoles | try 0.5 speed, notmyname sounds totally spaced :D | 15:44 |
| acoles | notmyname: you *have* to do a summit presentation in the style of 0.5 speed! | 15:44 |
| acoles | i can't stop laughing | 15:45 |
| cschwede | haha, 0.5 is really funny! | 15:45 |
| acoles | notmyname: thats how you will be speaking in 50 years time | 15:45 |
| notmyname | lol | 15:46 |
| cschwede | acoles: if he’s talking with the same proficiency on storage technologies in 50 years everything is alright :) | 15:47 |
| acoles | its where he's talking about swift birds at 3.35 thats cracking me up | 15:47 |
| openstackgerrit | Merged openstack/swift: Imported Translations from Transifex https://review.openstack.org/173687 | 15:48 |
| cschwede | is this youtube 2x/0.5x thingy the modern version of „you have to play this record backwards to get the hidden message“? | 15:48 |
| acoles | cschwede: :) | 15:49 |
| peluse | OMG, yeah the .5 is classic. Sorry I brought all this up notmyname but you gotta love it! | 15:49 |
| * peluse thinks it must be tought to be PTL sometimes.... | 15:49 | |
| *** annegentle has joined #openstack-swift | 15:54 | |
| portante | peluse: I am sure notmyname could teach us how to be a good PTL | 15:54 |
| *** jistr has quit IRC | 16:03 | |
| *** jistr has joined #openstack-swift | 16:04 | |
| *** krykowski has quit IRC | 16:06 | |
| peluse | portante! What's happenin man? | 16:08 |
| peluse | notmyname, chances are good I can't get out of my noon conflict today so wrt scheduling sign me up for whatever... | 16:09 |
| *** jistr has quit IRC | 16:09 | |
| notmyname | peluse: heh, it's fine. I think i want to go over some summit stuff. what's been proposed, some other ideas. just talk about it in general. not a "hand out tasks" thing | 16:10 |
| peluse | cool | 16:10 |
| peluse | good call whoever is typing right now on the etherbad wrt EC topic for summit :) | 16:11 |
| peluse | note that IBM has open sources storlet code now, the link is in the session proposal on the etherpad | 16:13 |
| *** dosaboy has quit IRC | 16:13 | |
| peluse | https://github.com/Open-I-Beam/swift-storlets | 16:14 |
| notmyname | thanks | 16:14 |
| *** dosaboy has joined #openstack-swift | 16:14 | |
| openstackgerrit | Merged openstack/swift: Update swift-recon doc with more options https://review.openstack.org/173767 | 16:15 |
| *** jistr has joined #openstack-swift | 16:15 | |
| *** annegentle has quit IRC | 16:16 | |
| *** jistr has quit IRC | 16:17 | |
| acoles | peluse: could be me typing | 16:23 |
| peluse | yeah, that was my guess | 16:25 |
| *** welldannit has quit IRC | 16:25 | |
| openstackgerrit | Alistair Coles proposed openstack/swift: Filter etag key from ssync replication-headers https://review.openstack.org/173973 | 16:34 |
| acoles | peluse: clayg: ^^ pls tell me i'm wrong about the reconstructor scenario | 16:35 |
| *** EmilienM is now known as EmilienM|afk | 16:38 | |
| peluse | acoles, on my way out the door but will check it out in a few hrs... thanks! | 16:39 |
| *** Gu_______ has quit IRC | 16:45 | |
| *** peluse has quit IRC | 16:46 | |
| notmyname | who proposed the "test framework" session on the etherpad? | 16:47 |
| zaitcev | "the etherpad"? | 16:48 |
| *** Gu_______ has joined #openstack-swift | 16:48 | |
| notmyname | the summit talks | 16:49 |
| acoles | notmyname: i believe jrichli added it and she proposed me to lead it but i made it a tbd cos i'm a chicken | 16:49 |
| notmyname | zaitcev: sorry. WHY CAN"T YOU READ MY MIND? | 16:49 |
| notmyname | lol | 16:49 |
| zaitcev | No, it makes complete sense. I'm just out of context. | 16:49 |
| acoles | notmyname: not sure about a fishbowl on that topic, maybe working session??? | 16:49 |
| notmyname | depends on what you want to get out of it. as written, it might be ok as a fishbowl. but if it's more of "here's what we need to change and the next steps" then a working session might be better | 16:51 |
| acoles | not sure if jrichli is about? | 16:51 |
| notmyname | jrichli: jrichli: jrichli: | 16:51 |
| *** ozialien has quit IRC | 17:00 | |
| cschwede | notmyname: i put myself onto the „test framework“ topic | 17:01 |
| notmyname | thanks | 17:02 |
| *** aix has quit IRC | 17:22 | |
| *** bkopilov has quit IRC | 17:26 | |
| *** bkopilov has joined #openstack-swift | 17:30 | |
| *** geaaru has quit IRC | 17:39 | |
| *** bkopilov has quit IRC | 17:39 | |
| *** bkopilov has joined #openstack-swift | 17:39 | |
| *** bkopilov has quit IRC | 17:44 | |
| *** erlon has joined #openstack-swift | 17:50 | |
| *** openstackstatus has quit IRC | 17:58 | |
| *** fanyaohong has quit IRC | 17:59 | |
| *** openstackstatus has joined #openstack-swift | 18:00 | |
| *** ChanServ sets mode: +v openstackstatus | 18:00 | |
| *** ozialien has joined #openstack-swift | 18:02 | |
| *** annegentle has joined #openstack-swift | 18:04 | |
| -openstackstatus- NOTICE: Gerrit has stopped emitting events so Zuul is not alerted to changes. We will restart Gerrit shortly to correct the problem. | 18:04 | |
| *** ChanServ changes topic to "Gerrit has stopped emitting events so Zuul is not alerted to changes. We will restart Gerrit shortly to correct the problem." | 18:04 | |
| *** MVenesio has joined #openstack-swift | 18:13 | |
| *** dosaboy has quit IRC | 18:13 | |
| *** dosaboy has joined #openstack-swift | 18:14 | |
| *** cutforth has joined #openstack-swift | 18:18 | |
| *** dosaboy has quit IRC | 18:20 | |
| *** Gu_______ has quit IRC | 18:20 | |
| *** dosaboy has joined #openstack-swift | 18:21 | |
| *** acoles is now known as acoles_away | 18:21 | |
| *** EmilienM|afk is now known as EmilienM | 18:22 | |
| *** dosaboy has quit IRC | 18:22 | |
| *** dosaboy has joined #openstack-swift | 18:24 | |
| *** dosaboy has quit IRC | 18:24 | |
| *** dosaboy has joined #openstack-swift | 18:24 | |
| *** ChanServ changes topic to "Review Dashboard: http://goo.gl/vysJqI | Summary Dashboard: http://goo.gl/2By1qv | Summit scheduling: https://etherpad.openstack.org/p/liberty-swift-summit-topics | Logs: http://eavesdrop.openstack.org/irclogs/%23openstack-swift/" | 18:27 | |
| -openstackstatus- NOTICE: Gerrit has been restarted. New patches, approvals, and rechecks between 17:30 and 18:20 UTC may have been missed by Zuul and will need rechecks or new approvals added. | 18:27 | |
| *** annegentle has quit IRC | 18:35 | |
| *** annegentle has joined #openstack-swift | 18:35 | |
| notmyname | FYI, feature/ec and feature/ec_review have been deleted from upstream repos | 18:40 |
| notmyname | the old feature/ec has been tagged as erasure_code_dev_history | 18:40 |
| *** Vinsh has quit IRC | 18:47 | |
| *** thumpba has joined #openstack-swift | 18:49 | |
| *** thumpba_ has joined #openstack-swift | 18:50 | |
| *** ho has joined #openstack-swift | 18:53 | |
| ho | good morning guys! | 18:53 |
| *** thumpba has quit IRC | 18:53 | |
| notmyname | ho: hello! | 18:54 |
| ho | notmyname: hi! | 18:55 |
| *** kota_ has joined #openstack-swift | 18:55 | |
| notmyname | thanks for waking up early | 18:55 |
| kota_ | morning :) | 18:56 |
| clayg | are we doing this? | 18:56 |
| ho | notmyname: I can enjoy the current situation :-) | 18:57 |
| ho | kota_: good morning! | 18:57 |
| *** acoles_away is now known as acoles | 18:57 | |
| kota_ | ho: Good :) | 18:58 |
| *** tsg_ has joined #openstack-swift | 18:58 | |
| mattoliverau | Morning | 19:00 |
| *** jrichli_ has joined #openstack-swift | 19:00 | |
| kota_ | mattoliverau: morning | 19:00 |
| notmyname | hi | 19:00 |
| notmyname | ok, meeting time | 19:00 |
| notmyname | in #openstack-meeting | 19:01 |
| *** Vinsh has joined #openstack-swift | 19:01 | |
| clayg | acoles: i'm doing some reconstruction and not seeing the Etag (only ETag) | 19:01 |
| *** silor1 has joined #openstack-swift | 19:04 | |
| clayg | i'm just reading this and remember trying it one time a long time ago - i could be wrong -> http://docs.openstack.org/developer/swift/misc.html#acls | 19:05 |
| acoles | clayg: the ssync_sender test does blow up if you revert my fix in server.py | 19:06 |
| *** silor has quit IRC | 19:06 | |
| clayg | acoles: i'm sure it's worth fixing - i'm trying to understand if we're hozed (or if not, why not - so far we don't look hozed) | 19:06 |
| acoles | clayg: yeah i hope we're not hozed! as my commit msg said it was a hypothetical risk of harm, but a real risk of just being confusing | 19:10 |
| *** jrichli_ has quit IRC | 19:13 | |
| *** jrichli_ has joined #openstack-swift | 19:13 | |
| *** jrichli_ has quit IRC | 19:16 | |
| *** silor1 has quit IRC | 19:23 | |
| *** mmcardle has quit IRC | 19:27 | |
| *** david-lyle has joined #openstack-swift | 19:30 | |
| *** peluse has joined #openstack-swift | 19:47 | |
| *** ozialien has quit IRC | 19:50 | |
| *** bkopilov has joined #openstack-swift | 19:54 | |
| clayg | acoles: i'm missing something | 19:59 |
| acoles | clayg: ? | 20:00 |
| clayg | i *think* replication rsync is adding the "sent" etag as Etag next to the *computed* ETag | 20:00 |
| clayg | acoles: but I don't think ec every *sends* the etag (as Etag or otherwise) and the *only* think that gets stored is the computed ETag | 20:00 |
| acoles | clayg: looking... | 20:02 |
| mattoliverau | wow too many uses of the work etag this early in the morning.. I'm going back to bed for a while, be back later :P | 20:02 |
| clayg | acoles: yeah replication (having not *excluded* the etag header) is then adding the 'etag'.title() (in addtiton to the computed value stored as ETag). | 20:03 |
| kota_ | mattoliverau: yes, of course :P | 20:03 |
| *** kota_ has quit IRC | 20:03 | |
| acoles | clayg: hmmm ssync_sender send_put just reads the fd metadata and sends all apart from content-length so how would ec not send an etag header? | 20:04 |
| clayg | so that's why we're not screwed - ec fixed this ssync bug on accident by removing the data integrity check? I'm not acctually sure if replication ssync will validate the computed etag matching the sent value. | 20:04 |
| clayg | acoles: so you're thinking in the revert case - not where we have a StreamingRebuildingDiskFile and we've manipulated the diskfile | 20:05 |
| acoles | clayg: i'm thinking first you have an FA that has been reverted via ssync so its has ETag and Etag in its meta, then you use that FA as a source of meta for a rebuild | 20:06 |
| acoles | the rebuild disk stream thingy strips out ETag but not Etag, so Etag gets sent | 20:06 |
| clayg | i haven't been able to get a fragment archive with ETag and Etag - i tried rebuild and revert - will keep trying | 20:06 |
| clayg | fwiw it was easy to get both on replication | 20:07 |
| acoles | clayg: hm, thats what confuses me cos my ssync_sender test does show them popping up, so have i seeded the test wrong | 20:07 |
| acoles | ? | 20:07 |
| clayg | acoles: not sure - unanticipated interaction between some other component? maybe *I'm* testing wrong? I want to understand which. | 20:08 |
| acoles | clayg: you mean replication+ssync?? | 20:08 |
| clayg | acoles: yes i tested that - obviously demonstrates the issue you described - but it's not a risk because the etag is the same | 20:09 |
| clayg | acoles: i'm *trying* to get the ETag Etag with reconstructor+ssync because it would be really damning if kilo did in fact behave in the way you described | 20:10 |
| acoles | clayg: yup just double checking you, and yup no risk of harm with replication | 20:10 |
| clayg | acoles: I *am* saying that *at the moment* I haven't been able to reproduce the potential issue with reconstructor+ssync on master and I'd like to understand why | 20:10 |
| acoles | me too | 20:10 |
| acoles | as in understand! | 20:10 |
| clayg | ok - i'll keep testing ;) | 20:10 |
| peluse | OK, I'm lost | 20:11 |
| clayg | acoles: I think it has to be down in diskfile somewhere - i'm just not seeing the header come across to the object server in the ssync message on revert | 20:11 |
| clayg | so you're right it must never get to ssync | 20:11 |
| clayg | idk, i'll add prints there first | 20:12 |
| acoles | clayg: just to check - you have any .meta files in dir? | 20:12 |
| clayg | acoles: nope | 20:12 |
| acoles | not that i can think what difference | 20:12 |
| acoles | ok | 20:12 |
| clayg | acoles: i'll test that next!!! | 20:12 |
| clayg | :D | 20:12 |
| clayg | i'm trying to get kota his rc2! | 20:12 |
| acoles | note to all for liberty - land a bug just in case you need to invoke a reason for RC2 ;) | 20:13 |
| clayg | nice! | 20:13 |
| clayg | it has to be this wierd thing like some combination of config options that no one would ever use | 20:14 |
| acoles | i did for kilo but forgot what it was, darn! | 20:14 |
| peluse | so not to disrupt the case sensitive etag discussion to much, but... | 20:14 |
| clayg | but you can totally be like - and *see* 500 - can't have that. | 20:14 |
| clayg | acoles: LOL | 20:14 |
| clayg | acoles: ok, some "name" also comes out of metadata - but I don't see *that* in the replication headers - how does that work? | 20:15 |
| acoles | oh i think that gets filtered somewhere, hang on | 20:15 |
| peluse | ahhh never mind | 20:16 |
| acoles | clayg: yes, sorry, ssync_sender filters out name and content-length | 20:16 |
| clayg | oh... right pthththth | 20:16 |
| clayg | i dont' get it - they *have* an ETag metdata | 20:17 |
| clayg | but send_put get_metadata never sees it | 20:17 |
| acoles | oh man, where'd it go ??? | 20:19 |
| peluse | where iexactly is the lower case etag being created? | 20:19 |
| acoles | peluse: in the obj server PUT path | 20:20 |
| acoles | it gets an Etag header from ssync_rx (well, i thought it did) and copies it into PUT metadata | 20:20 |
| acoles | based on it being in replication-headers | 20:20 |
| clayg | acoles: i just popped a ECDiskFileManager in the console and it's *totally* in df.read_metadata | 20:21 |
| acoles | peluse: but also computes the hash and adds that to metadata as ETag - with a big T | 20:21 |
| clayg | so confused | 20:21 |
| acoles | clayg: right, we didnt futz with any of that code | 20:21 |
| acoles | did we? | 20:21 |
| acoles | peluse: this bit me when i wrote some ssync tests and the rx metadata != sender metadata and the diff was Etag key, and its so confusing that the 'correct' metadata key on disk is ETag but of course headers are always titlecaseed :/ | 20:24 |
| peluse | yeah, I'm almost caught up :) | 20:25 |
| torgomatic | Is there a reason not to run metadata through a HeaderKeyDict? | 20:26 |
| torgomatic | Just asking; I'm not really up to speed here. | 20:26 |
| *** cutforth has quit IRC | 20:26 | |
| acoles | torgomatic: good question | 20:27 |
| acoles | torgomatic: there's lots of files out there with ETag in meta but i guess nothing would break if we started to titlecase it??? like, it happens anyway | 20:28 |
| clayg | ^ this is a *seperate* question | 20:28 |
| acoles | torgomatic: we'd still need to fix this replication header thing tho | 20:28 |
| acoles | clayg: yes ^^ | 20:29 |
| torgomatic | acoles: ok. Like i said, not up to speed. | 20:29 |
| clayg | we know the current situation is confusing and should be fixed - how to fix it is a good question - if we need an rc2 is another, more pressing question | 20:29 |
| acoles | torgomatic: no, its a good thought | 20:29 |
| clayg | ok - i have a fragment with Etag and ETag | 20:30 |
| acoles | whoa | 20:30 |
| acoles | how? | 20:31 |
| acoles | or rather what changed? | 20:31 |
| *** Telma has joined #openstack-swift | 20:33 | |
| peluse | so, hmm, I have one too and all I did was run the revert probe test and print the metadata in the obj server put path just before the writer.put(metadata) line | 20:33 |
| peluse | {'Etag': '40cdf2b84f90b5ce4704660efeaa58d6', 'Content-Length': '934208', 'X-Object-Meta-Bar': 'meta-bar', 'X-Object-Sysmeta-Ec-Frag-Index': '1', 'X-Object-Sysmeta-Ec-Content-Length': '3735552', 'X-Object-Sysmeta-Ec-Etag': '6196a6a3747f858b39574c3631244260', 'ETag': '40cdf2b84f90b5ce4704660efeaa58d6', 'X-Timestamp': '1429129717.51206', 'X-Object-Sysmeta-Ec-Scheme': 'isa_l_rs_vand 4+2', 'Content-Type': 'application/octet-stream', 'X-Obj | 20:33 |
| peluse | ect-Sysmeta-Ec-Segment-Size': '1048576'} | 20:33 |
| clayg | i'm not sure - maybe i was missing it before | 20:34 |
| clayg | like i didn't see it - and it was there - it's there now | 20:34 |
| peluse | could be gremlins... | 20:35 |
| acoles | peluse: nice | 20:35 |
| acoles | clayg: peluse : so now imagine you use peluse 's FA as a source for a rebuild metadata, our code discards the E-BIG-T-ag but the Etag will be sent - thats my pathological case | 20:36 |
| peluse | so did we decide if this was on master before the merge (the double case senstive etag) | 20:36 |
| peluse | yeah, I think that's a real case - not so crazy | 20:36 |
| *** Telma has quit IRC | 20:36 | |
| acoles | peluse: yes it was on master we didn't make it so | 20:37 |
| peluse | seems like the 'safe fix' is to discard both, if they are both there, in the EC recon when we use the template metadata and avoid messing with anything about how master used to operate | 20:37 |
| peluse | but probably a better fix would be to just have one Etag | 20:37 |
| peluse | I'm nOt Sure which is better :) | 20:38 |
| peluse | oh come one, case sensitive humor has got to be worth something! | 20:39 |
| acoles | peluse: so that would work but i prefer my fix because it removes confusion in every case not just reconn | 20:39 |
| acoles | peluse: or both, belt and braces | 20:40 |
| peluse | oh yeah, forgot there was a patch associated with that commit message - will go look now! | 20:40 |
| acoles | peluse: HEH | 20:40 |
| clayg | rofl - the object server 422's | 20:41 |
| acoles | hey, so was i right? | 20:41 |
| acoles | clayg: i love that i sit over the ocean dreaming up issues and you actually *test and prove* in like 10% the time it would take me | 20:43 |
| clayg | yeah so i'm not sure why it took me so long to see it - but revert will duplicate etag (same value twice different keys) - then yeah - forcing that object to be used as the base of a rebuild will leave the Etag in the shipped headers - and the remote object server under the ssync reciever will reject the reconstrcted FA because the computed ETag doesn't match the clients requested Etag | 20:43 |
| acoles | voila | 20:43 |
| clayg | i can totally write a probe test for this :P | 20:44 |
| clayg | but I *really* need to go pee | 20:44 |
| acoles | go. now. | 20:44 |
| clayg | acoles: is this my turn to tell YOU to go to bed now? | 20:44 |
| *** gyee has joined #openstack-swift | 20:44 | |
| *** gyee has quit IRC | 20:44 | |
| clayg | acoles: I think peluse and I can take it form here (seeing as how you've already written the fix) | 20:44 |
| peluse | acoles, yeah, I like your change, just did a quick check and it worked for my manual test | 20:45 |
| acoles | so maybe peluse's idea of ALSO scrubbing Etag in the recon rebuilder is good too | 20:45 |
| acoles | peluse: clayg: remember i left a comment in the ec review about this :P ;P | 20:45 |
| peluse | acoles, yeah, we decided it was too confusing to think about at the time :) | 20:46 |
| *** noye has quit IRC | 20:46 | |
| clayg | peluse: yeah and it was a known issue that we convinced ourselves in testing didn't matter in practice | 20:46 |
| clayg | it's a pretty *specific* failure mode | 20:46 |
| clayg | but ... | 20:46 |
| clayg | that being said... | 20:46 |
| peluse | but yeah it can't hurt to drop it in the Ecrecon (if there) also | 20:46 |
| clayg | acoles: YOU WERE RIGHT | 20:46 |
| peluse | indeed! | 20:46 |
| acoles | clayg: did you pee? i'm getting concerned | 20:46 |
| * peluse was also curious about claygs bodily functions | 20:47 | |
| acoles | peluse: clayg: tbh chances are even i would not grok the comment i left in review ;) | 20:48 |
| peluse | so I have another one that I alsmost have a probe test written for... related to the card on trello in the TODO column (the only one) | 20:49 |
| peluse | at one point the GET path used the node index to pass a hint to the storage node about which FI it wanted, that was pulled and made the EC recon issue decribed in that card go away but a simpler test case... | 20:50 |
| peluse | where you just create an extra .data of a different FI in a hash dir and do a GET, its a roll of the dice what you'll get back | 20:50 |
| *** lpabon has joined #openstack-swift | 20:51 | |
| acoles | peluse: so this is plain old proxy GET? | 20:51 |
| peluse | yeah | 20:51 |
| acoles | hum, so you'll get back whatever FI sorts first | 20:52 |
| peluse | well, a #3 doesn't sort before a #0 and I was getting a #3 back (the bad one I seeded) | 20:53 |
| acoles | peluse: is that a problem, as long as proxy gets back ndata distinct frags | 20:53 |
| peluse | well, it came up in the context of the EC recon so I haven't thought that much about the generic GET case | 20:54 |
| peluse | but I suspect if you had a few different participating nodes all holding multiple FI's you may not get enough back of the "right ones" to decode | 20:54 |
| peluse | and maybe that's a bit more pathalogical | 20:54 |
| acoles | peluse: diskfile reverse sorts so would find #3 before #0 | 20:55 |
| peluse | oh yeah, OK that makes sense | 20:55 |
| acoles | peluse: so if #3 is a handoff, gets reverted to node 3 but then the purge of #3 on node 0 fails, then node 0 and 3 will both return #3 to proxy GET?? | 20:56 |
| peluse | I guess the EC recon path is the same as GET now (once upon a time I was only getting the FI's that it needed based on asking pyeclib what those were, now it asks everyone to send theirs) | 20:56 |
| peluse | hmmm | 20:57 |
| peluse | yeah, guess so | 20:58 |
| acoles | umm | 20:58 |
| peluse | feels like we should still have that hint for all GET of "the FI I prefer is..." | 20:59 |
| peluse | man, long pee for clayg | 20:59 |
| peluse | hey, I'm a poet and didn't even know it | 20:59 |
| acoles | peluse: yes, but torgomatic explained to me that the proxy GET will move onto other nodes if a GET fails, so once its not GETting from a primary it doesn't knwo what FI to request? | 21:00 |
| *** gyee has joined #openstack-swift | 21:00 | |
| peluse | yeah, that's the original issue described in the trello card :) | 21:01 |
| peluse | when we had that 'hint' in there | 21:01 |
| acoles | ah | 21:01 |
| peluse | I don't know that there's a "trivial" fix for it but seems like we went througha lot of trouble to suport multiple FI per hash dir and w/o tackling this its sorta for what? | 21:02 |
| peluse | if they alphabetically line up we're good??? :) | 21:03 |
| acoles | peluse: if you always choose a handoff whose node index is < frag index then that node will always return its primary frag | 21:04 |
| clayg | acoles: handoff's don't node indexes | 21:04 |
| peluse | huh? | 21:04 |
| peluse | huh to acoles because of what clayg said | 21:05 |
| acoles | clayg: peluse yeah, scratch that | 21:05 |
| peluse | and even if they did I think there's problems with the shared node_iter being used by the GET path for EC, I have a few rough comments in the bottom of the trello card as to why | 21:05 |
| acoles | so, "if you got lucky it works" | 21:05 |
| peluse | but I could be smoking crack, who knows | 21:05 |
| clayg | anyway - the multi-fi's in the same hashdir is always temporary - and rare - but when it happens it could effect GET - depending on if the doubled fi also exists on another node | 21:06 |
| clayg | you could have all the fi's in the same node - and then get doesn't work at all | 21:06 |
| clayg | you need ec_ndata disks to have a fragment | 21:06 |
| peluse | well, you'd have to hav enough nodes in the primary part carrying other FI's that were of large FI value such that you couldn't decode if all the wrong onces were delivered | 21:06 |
| clayg | peluse: yes it's a thing - but mostly requires the proxy to be able to make extra GET's with more context | 21:07 |
| clayg | then we have to make the object server be able to plumb the context to diskfile | 21:07 |
| peluse | yeah | 21:07 |
| acoles | so i took comment at line 1823 in obj.py to embrace *some* of these cases i.e. 'more work needed' | 21:08 |
| peluse | so what hapens in pyeclib I wonder if we feed him a pile of frags where there's a few dups in there but enough to decode? | 21:08 |
| clayg | peluse: either error, garbage, or something else | 21:09 |
| peluse | likely... | 21:09 |
| peluse | which means even one incorrectly delivered FI could hose us | 21:09 |
| peluse | worth testing | 21:09 |
| clayg | how the... | 21:10 |
| acoles | peluse: does pyeclib add any indexes in FA 'headers'? (i'm out of depth, but it adds its own header to a frag?) so it might at least throw an error | 21:10 |
| peluse | oh yeah, I forgot it does. it has its own metadata that may allow it to sort this out | 21:10 |
| acoles | peluse: cool. thats how i'd have done it :P | 21:10 |
| peluse | opaque to us - the headers are I mean | 21:10 |
| * peluse seems to be speaking like yoda for some reason | 21:11 | |
| peluse | clayg, if you want to wrap the etag stuff up, I'll try feeding pyeclib some sutff like this | 21:11 |
| clayg | i. am. so. confused. | 21:12 |
| *** esker has quit IRC | 21:12 | |
| acoles | peluse: clay: winding back to the etag thing, i should say that until i wrote the patch today i was only aware of the potential for confusion. it was only today that the potential for 422 dawned on me. otherwise i would have jumped up and down more before. | 21:12 |
| peluse | just glad you did | 21:13 |
| *** lpabon has quit IRC | 21:14 | |
| acoles | clayg: you need me, or can you take the etag thing from here on...? | 21:14 |
| *** esker has joined #openstack-swift | 21:15 | |
| clayg | i've somehow gotten myself backinto a world where get_metadata doesn't include the Etag :'( | 21:15 |
| clayg | but something needs to be fixed | 21:16 |
| clayg | acoles: we'll get it - thanks for the clue | 21:16 |
| peluse | clayg, nice trick... I'l be back in just a few... | 21:16 |
| peluse | acoles, later man | 21:16 |
| acoles | clayg: ok, i'm dropping out, leave me a note if there stuff to follow up tomorrow. | 21:17 |
| acoles | clayg: peluse : have a good one | 21:17 |
| *** acoles is now known as acoles_away | 21:18 | |
| clayg | oh wow - ok - I think i was seeing the rebuild/revert race | 21:20 |
| clayg | ok, phew - k - totally need an rc2 | 21:21 |
| clayg | peluse: so you have a probetest to validate a rebuild from a reverted fragment? | 21:21 |
| peluse | clayg, not right now | 21:22 |
| peluse | but I can try and whip one together if you'd like? | 21:22 |
| clayg | peluse: nope - i'm on it | 21:23 |
| peluse | OK, I'm going to dork with feeding pyeclib various crap as a data point in the "GET can return the wrong FI if multiple are available" deal | 21:23 |
| *** jrichli has quit IRC | 21:23 | |
| *** jrichli has joined #openstack-swift | 21:24 | |
| clayg | peluse: wfm | 21:25 |
| *** noye has joined #openstack-swift | 21:35 | |
| peluse | clayg, OK, so pyelib is smart enough not to puke if we give it too many frags including if those 'too many' include duplicates so exoosure to the "GET the wrong FI" is limited to the case where there are enough nodes with extra FI's on them that are returned incorrectly such that the min number of unique fragments isn't given to PyECLib | 21:36 |
| notmyname | clayg: ack | 21:40 |
| clayg | notmyname: my favorite part of this bug is that you can *either* add devices to your ec cluster and rebalance the ring *or* you can have drive failures that require reconstruction - BUT NOT BOTH! | 21:45 |
| clayg | notmyname: someone needs to track down kota and tell him to get cracking on his reconstruct vs. decode/encode config option for the reconstructor ;) | 21:45 |
| *** jamielennox|away is now known as jamielennox | 21:46 | |
| *** annegentle has quit IRC | 21:47 | |
| *** annegentle has joined #openstack-swift | 21:52 | |
| peluse | clayg, what config option is that? | 21:53 |
| clayg | peluse: you know the jerasure hack with the comment about parity fragments in the reconstrctuor - _reconstruct method I think | 21:54 |
| clayg | it works on the isa-l | 21:54 |
| peluse | yeah, but its fixed - why code in a config option? I added a trello TODO item to just upgrade to 1.0.6 or better and then change it to always reconstruct | 21:54 |
| *** jkugel has quit IRC | 21:55 | |
| clayg | because we can't drop the encode/decode hack from kilo as long as people are packaging 1.0.6 (e.g. everyone that has built packages today) | 21:56 |
| peluse | oh man | 21:56 |
| clayg | unfortunately the hack in kilo works for everyone except for kota | 21:58 |
| peluse | oh, because of his library? | 21:58 |
| clayg | something like that | 21:58 |
| peluse | that's a pisser | 21:58 |
| clayg | ... but since we're doing an rc2 - we get the chance to make the hack optional | 21:58 |
| clayg | it'll work better for isa-l anyway | 21:58 |
| peluse | gotcha | 21:59 |
| clayg | it *maybe* doesn't even have to be a config option - it could just "if backend is jerasure" do this dumb thing | 21:59 |
| peluse | well, not sure I understand the comment on working better for isa-l anyway? | 21:59 |
| clayg | and then we make sure to remove the condition as soon as some sha of swift can depend on at *least* 1.0.7 | 21:59 |
| clayg | but there is the kilo case where someone *knows* they have 1.0.7 and jerasure - but wants it to not do the hack | 22:00 |
| peluse | dealing with legacy issues issues before we even release, nice | 22:00 |
| *** esker has quit IRC | 22:02 | |
| *** annegentle has quit IRC | 22:03 | |
| clayg | peluse: well... what are you gunna do - we found the jerasure reconstruct bug too late | 22:05 |
| clayg | notmyname: is there *any* chance we could get a 1.0.7 into the kilo dependencies? Like I *know* swifterdarrell can get it packaged - but what about zaitcev and some debian type person | 22:05 |
| zaitcev | clayg: I don't mind, presuming you mean PyECLib 1.0.7, which does not exist at this time. Now that I got 1.0.6 patched, it should be easier. Poke Kevin or Tushar to cut the release. | 22:06 |
| peluse | tsg just told me in another message that 107 is up and available | 22:07 |
| zaitcev | BUUUUULL | 22:07 |
| peluse | seeing if he can join us... | 22:07 |
| zaitcev | Well, I wanted to have a good word with him about it | 22:07 |
| zaitcev | It's SUPER STUPID | 22:07 |
| zaitcev | He uploads random tarballs to PyPi and calls it a release. That is really effin wrong. He should make a tag at bitbucket and upload those tarballs to PyPi. | 22:08 |
| peluse | clayg, others: separate topic I updated the trello card with more info on the GET FI w/multiple thingy, please take a look when you get a chance. Want to get thoughts on requirement for RC2 or not | 22:08 |
| zaitcev | There's even a commit at Bitbucket that says "tagging this as 1.0.5m because we have 1.0.5 at PyPy". Seriously! | 22:09 |
| peluse | so how do you really feel zaitcev? | 22:09 |
| clayg | zaitcev: us poor developers don't know how to do any of this stuff :) | 22:09 |
| peluse | I thought packaing was done at the FedEx store when I dropped shit off :) | 22:09 |
| clayg | you have to tell us what to type - but with very small words - like teaching a baby | 22:09 |
| *** MVenesio has quit IRC | 22:09 | |
| tsg_ | clayg: thank you ;) | 22:09 |
| tsg_ | zaitcev: why do you think the tarballs are not random and not associated with a tag?? | 22:10 |
| tsg_ | s/not random/random | 22:10 |
| zaitcev | tsg_: well, where is 1.0.7? | 22:10 |
| tsg_ | https://bitbucket.org/tsg-/liberasurecode/commits/all | 22:10 |
| zaitcev | oh | 22:11 |
| zaitcev | url = https://bitbucket.org/kmgreen2/pyeclib.git | 22:11 |
| zaitcev | That... explains it | 22:11 |
| tsg_ | zaitcev: patience .. kmgreen2 is testing liberasurecode together with latest pyeclib so we can push that tag | 22:11 |
| *** shri1 has joined #openstack-swift | 22:12 | |
| tsg_ | zaitcev: the pyeclib side got complicated because the pypi version needed to be special to keep Jenkins happy | 22:12 |
| zaitcev | tsg_: so, which one is the official upstream, tsg- or kmgreen2? | 22:12 |
| tsg_ | so we maintain 2 branches there: master, 1.0_liberasurecode_integrated | 22:12 |
| clayg | zaitcev: I'd acctually like to know that as well | 22:12 |
| tsg_ | zaitcev: liberasurecode and pyeclib are 2 separate projects :) | 22:12 |
| zaitcev | wait, are we discussing pyeclib or liberasurecode? | 22:12 |
| shri1 | Hey Guys.. I have a question about the object auditor. Does the object-auditor compute the md5 checksum and compare it with the one stored in the extended attributes? | 22:13 |
| clayg | shri1: yeah it does | 22:13 |
| notmyname | I asked in -infra about what it takes to do a dependency version bump. a feature freeze exception to the requirements repo | 22:13 |
| clayg | notmyname: that sounds like "go fuck yourself" | 22:13 |
| peluse | notmyname, and what does *that* take? | 22:13 |
| notmyname | clayg: no, I don't think it is ;-) | 22:14 |
| tsg_ | zaitcev: I was trying to indicate that liberasurecode is already tagged for 1.0.7 - pyeclib should be tagged next | 22:14 |
| notmyname | jeblair: what does a FFE look like? I've never done one. are there docs? | 22:14 |
| * peluse thinks clayg still has some sleep to catch up on :) | 22:14 | |
| shri1 | clayg: Thanks! In the swift/obj/auditor.py file, I don't see any computation of md5(). Where does this happen? | 22:14 |
| clayg | if *I* was deciding to handoff FFE on the *GLOBAL REQUIREMENTS REPO* I would be *way* stingy | 22:14 |
| tsg_ | zaitcev: you should be creating rpms off of the git tags on "master" | 22:14 |
| notmyname | clayg: indeed | 22:14 |
| peluse | shri1, its in diskfile | 22:14 |
| zaitcev | tsg_: I see no problem with liberasurecode. It's fine. | 22:14 |
| clayg | shri1: oh the sneaky - what peluse said | 22:14 |
| tsg_ | zaitcev: for pyeclib tags .. only use the latest tag off of master | 22:15 |
| shri1 | I see.. let me check. | 22:15 |
| peluse | was trying to beat clayg to the punch, I can point you to the LOC if you like :) | 22:15 |
| clayg | waho! | 22:15 |
| peluse | gotta have some sort of accomplishment this week! | 22:15 |
| jeblair | notmyname: i don't know, sorry. maybe this: https://wiki.openstack.org/wiki/FeatureFreeze ? | 22:15 |
| jeblair | notmyname: fungi may know more | 22:16 |
| tsg_ | zaitcev: you will see 2 tags pushed latest today - v1.0.7 and v1.0.7m, please use v1.0.7m (that doesn't have integrated liberasurecode etc) and expects liberasurecode be an external dependency | 22:16 |
| tsg_ | zaitcev: s/latest today/later today | 22:16 |
| * notmyname goes down the yellow-brick road to -infra | 22:16 | |
| zaitcev | tsg_: thanks, I'll mark it in the RPM spec about the -m | 22:17 |
| notmyname | jeblair: thanks | 22:17 |
| peluse | shri1, check out class DiskFileReader() and you'll see it | 22:17 |
| shri1 | oh.. I see. So I guess, DiskFileReader() gets called for auditing as well as when clients issue a GET for the object. Md5 calc has to be done in both cases. | 22:18 |
| tsg_ | zaitcev: thank you! | 22:18 |
| clayg | peluse: debug_wtf got left in! | 22:20 |
| peluse | shri1, yes. the audtor just reads it and that will check the MD5 and quarantine if needed via _handle_close_quarantine() | 22:20 |
| tsg_ | zaitcev: eventually, we'll get rid of the "1.0_liberasurecode_integrated" branch off of pyeclib - the whole reason that branch and related tags were created, was to keep Jenkins using the latest version of pyeclib regardless of where rpm/deb packaging stands | 22:20 |
| peluse | clayg, ooops :) | 22:20 |
| *** annegentle has joined #openstack-swift | 22:20 | |
| zaitcev | tsg_: well, whatever works for you guys | 22:20 |
| peluse | well, we all reviewed it... at least I didn't spell it out! | 22:21 |
| zaitcev | tsg_: I'm going to post a pull request about that madness with /usr/local | 22:21 |
| zaitcev | tsg_: please point out where I'm wrong, including OSX etc. | 22:22 |
| shri1 | peluse: oh.. ok. Looking at the code. | 22:22 |
| notmyname | ok, good news on the dependency front | 22:22 |
| tsg_ | zaitcev: I agree that needs to be cleaned up .. thank you for taking care of that! that stuff stayed around from the original python C backend implementation | 22:22 |
| clayg | notmyname: wow rly!? | 22:22 |
| notmyname | the process appears to be: (1) submit a patch to global requirements (2) send a message to the ML linking the patch and requesting the FFE | 22:23 |
| peluse | wow, just ask huh? | 22:23 |
| peluse | who woulda thunk? | 22:23 |
| notmyname | so since this is a dependency that only affects swift it shoudl be straightforward | 22:23 |
| clayg | wow - that's awesome | 22:23 |
| clayg | WTG OpenStack! | 22:23 |
| tsg_ | awesome! | 22:23 |
| * clayg takes back all the times he was a pessimistic jerk (today) | 22:24 | |
| notmyname | in the message, and in the commit message, we'll have to have justification | 22:24 |
| peluse | so tsg_ when did you say kevin would be done again? :) | 22:24 |
| clayg | peluse: heh | 22:24 |
| *** annegentle has quit IRC | 22:24 | |
| peluse | I did order him a shirt BTW... | 22:24 |
| notmyname | eg "This fixes bugs and eliminates a hack that is causing third-party libraries to not work" | 22:24 |
| mattoliverau | Morning.. Again | 22:24 |
| notmyname | once it lands there, we can do a patch on master and then backport that to proposed/kilo | 22:25 |
| peluse | nice | 22:25 |
| peluse | mattoliverau, morning | 22:25 |
| notmyname | so (1) patch global requirements (2) ML (3) patch swift master (4) backport swift master patch | 22:26 |
| notmyname | #3 happens after #1 lands | 22:26 |
| notmyname | then we also land the other stuff we want in the RC on master and backported to proposed/kilo | 22:26 |
| notmyname | then we tag RC2 | 22:26 |
| tsg_ | notmyname: ack | 22:26 |
| notmyname | so, what part do you want me to do? | 22:27 |
| * notmyname assumes at least #2 | 22:27 | |
| clayg | notmyname: as soon as tsg_ says i can pip install 1.0.7 i'll start working on the swift patch | 22:28 |
| *** keving has joined #openstack-swift | 22:28 | |
| notmyname | hi keving | 22:29 |
| keving | yo | 22:29 |
| tsg_ | notmyname: I will take care of #1 to start with. Your word to the ML will help with #2. | 22:29 |
| peluse | hey, its keving! | 22:29 |
| notmyname | keving: you must have been summoned | 22:29 |
| peluse | or just really good timing :) | 22:29 |
| keving | y, tushar said irc was blowing up | 22:30 |
| notmyname | heh | 22:30 |
| clayg | lol | 22:30 |
| peluse | dust ahs already settled... you're late! | 22:30 |
| notmyname | keving: we want to update the dependency on liberasurecode to 1.0.7 asap | 22:30 |
| notmyname | so on pypi pyeclib and liberasurecode are bundled? | 22:30 |
| *** dencaval has quit IRC | 22:33 | |
| notmyname | ie does pyeclib need to be updated (which is what's in global requirements) or just liberasurecode? | 22:34 |
| *** annegentle has joined #openstack-swift | 22:36 | |
| keving | sorry, had to step out… | 22:39 |
| keving | yeah, i am going to run a few more tests and we can push to pypi | 22:40 |
| keving | we are still pushing the bundled version, so we need to update the global requirements | 22:40 |
| keving | brb | 22:41 |
| tsg_ | notmyname: the pypi version would have to be bundled w/ liberasurecode for now, given 1) liberasurecode rpms for centos65 slave are not available 2) if we fix something in liberasurecode, unless we go through the whole Debian/Red Hat release process, Jenkins won't pull the latest | 22:41 |
| *** annegentle has quit IRC | 22:41 | |
| notmyname | ok | 22:41 |
| notmyname | even though they are in 2 separate repos? are you using submodules or something? | 22:41 |
| tsg_ | notmyname: at the moment, liberasurecode source is included as a "make dist" style "release" tarball | 22:42 |
| notmyname | ah ok | 22:43 |
| tsg_ | notmyname: (that was per discussion on -infra) | 22:43 |
| notmyname | proposed ML message: https://etherpad.openstack.org/p/uq5dUh9AJ6 | 22:43 |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Don't apply the wrong Etag validation to rebuilt fragments https://review.openstack.org/174138 | 22:43 |
| notmyname | clayg: that's the one, right, that spells R-C-2 for us? | 22:44 |
| peluse | clayg, just curious, why did you make _get_partners() a module level function in that patch just now? | 22:45 |
| tsg_ | notmyname: the draft ML message looks good | 22:45 |
| peluse | notmyname, there's also a cardon trello I'd like us all to look at and make sure we're OK with RC2 (with or without it) | 22:46 |
| peluse | its the only one in doing right now | 22:46 |
| *** km has joined #openstack-swift | 22:46 | |
| notmyname | I need to step out in about 15 minutes | 22:46 |
| peluse | clayg, and also - do you mean for that to be in addition to acoles patch right? | 22:47 |
| clayg | peluse: i wrote it *three* - fucking *three* times in the probetest before I finally got it right! | 22:48 |
| *** joeljwright has quit IRC | 22:48 | |
| clayg | it shouldn't be that hard - but apparently if you're an idiot - it's easy enough to get wrong - so... module level export helper | 22:48 |
| notmyname | also, it doesn't use anything from the class | 22:49 |
| clayg | peluse: i'm not 100% on acoles_away's approach - I think torgomatic had a good idea for addressing the larger issue | 22:49 |
| clayg | peluse: his works too - but your suggestion to just clear it out in the reconstructor (that was his original suggestion too) - seems like a more surgical obviously correct RC sorta of patch | 22:50 |
| clayg | I added some comments to acoles patch but I'm down either way I think - i am sorta curious if either torgomatic's HeaderKeyDict or fixing the issue in ssync_sender looks better than fixing it in object-server | 22:50 |
| clayg | but even if either of those works as good or better than acoles patch for the larger issue - i'm not sure how much of it belongs in the RC2 backport | 22:51 |
| peluse | clayg, agree for a quick RC fix, thanks | 22:51 |
| clayg | peluse: you worried about the _get_partners extraction - i sorta unilaterally decided it's one of those things you can sorta visually inspect and decide it's obviously correct - and it was helpful to the probe test - I could back off that tho and reduce the change in reconstructor to just the try KeyError (maybe a pop would have been better) | 22:53 |
| clayg | oh or a for all permutations of EtAg metadata.pop() # for realzy don't f'ing send it | 22:54 |
| peluse | nah, I think its fine it just wasn't obvious to me why it moved but makes sense for making the test a bit easier and is a tiny change | 22:54 |
| clayg | yeah i didn't start there - i was really surprized it took me three times | 22:54 |
| peluse | sleep deprived still maybe! | 22:55 |
| clayg | first was onodes[index + 1] % len(onodes) then it was onodes[index + 1 % len(onodes)] then *finally* onodes[(index + 1) % len(onodes)] | 22:55 |
| clayg | so at that point i was like - wow - i should apparently be importing this code from somewhere :P | 22:56 |
| peluse | heh | 22:56 |
| peluse | testing it now | 22:57 |
| *** erlon has quit IRC | 23:01 | |
| *** shri2 has joined #openstack-swift | 23:01 | |
| *** shri1 has quit IRC | 23:02 | |
| notmyname | I've got to step out. I'll be back in an hour or two and take care of whatever's pending for moving RC2 forward | 23:04 |
| *** shri2 has quit IRC | 23:07 | |
| *** chlong has joined #openstack-swift | 23:10 | |
| clayg | does anyone besides notmyname know if i'm supposed to propose patch 174138 to master or some sort of stable/kilo branch | 23:17 |
| patchbot | clayg: https://review.openstack.org/#/c/174138/ | 23:17 |
| *** fanyaohong has joined #openstack-swift | 23:18 | |
| peluse | stuck in mtg for another 40 min or so will fish review then | 23:21 |
| tsg_ | zaitcev: https://bitbucket.org/kmgreen2/pyeclib/commits/tag/v1.0.7m | 23:21 |
| clayg | peluse: perfect - i ended up liking the pop form better - going to post new patch | 23:21 |
| peluse | k | 23:21 |
| zaitcev | tsg_: thanks, got it | 23:21 |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Don't apply the wrong Etag validation to rebuilt fragments https://review.openstack.org/174138 | 23:21 |
| tsg_ | clayg: I have a similar question on the requirements patch .. submit it to master or stable/kilo? :) let me see if folks on -infra would know | 23:28 |
| clayg | tsg_: yeah great - let me know | 23:31 |
| clayg | tsg_: if you get a patch up that includes the 1.0.7 bump - i may re-use that same patch to pull out the encode/decode hack | 23:31 |
| tsg_ | clayg: sure | 23:32 |
| tsg_ | clayg: 1.0.7 is up on pypi btw | 23:32 |
| clayg | yup just pip'd it - working on the patch now | 23:32 |
| tsg_ | cool | 23:33 |
| clayg | tsg_: i was goign to ask - but then was like - oh you know who would know if pypi has 1.0.7???? | 23:33 |
| clayg | PYPI | 23:33 |
| *** distroguy has joined #openstack-swift | 23:33 | |
| *** j_king_ has joined #openstack-swift | 23:34 | |
| *** HenryG_ has joined #openstack-swift | 23:35 | |
| *** torgomatic_ has joined #openstack-swift | 23:37 | |
| *** dmsimard has quit IRC | 23:38 | |
| *** torgomatic has quit IRC | 23:38 | |
| *** j_king has quit IRC | 23:38 | |
| *** bkopilov has quit IRC | 23:38 | |
| *** HenryG has quit IRC | 23:38 | |
| *** torgomatic_ is now known as torgomatic | 23:38 | |
| clayg | tsg_: my ec_crazy_rebuild script is still giving me different md5's from reconstruct vs. encode/decode - worse probetests for rebuild is failing when I revert the hack | 23:38 |
| *** bkopilov has joined #openstack-swift | 23:38 | |
| clayg | pip freeze says 1.0.7 - but I'm not sure liberasure and ldconfig i don't even know | 23:38 |
| *** omame has quit IRC | 23:39 | |
| *** csmart has quit IRC | 23:39 | |
| *** klrmn has quit IRC | 23:39 | |
| *** omame has joined #openstack-swift | 23:39 | |
| *** dmsimard_ has joined #openstack-swift | 23:39 | |
| *** dmsimard_ is now known as dmsimard | 23:39 | |
| *** klrmn1 has joined #openstack-swift | 23:39 | |
| clayg | hrmm.... yeah locate liberasure finds /usr/local/lib/liberasurecode.so.1.0.6 | 23:39 |
| *** klrmn1 is now known as klrmn | 23:39 | |
| tsg_ | clayg: you don't have liberasurecode.so.1.0.7 under /usr/local/lib at all? | 23:42 |
| tsg_ | (and liberasurecode.so.1 relinked to that version) | 23:42 |
| clayg | turns out no - i've clone'd the liberasure repo and going to try to build it | 23:43 |
| clayg | nope it's a symlink to 1.0.6 | 23:43 |
| *** esker has joined #openstack-swift | 23:44 | |
| clayg | tsg_: yeah so i didn't get any error/warning with pyeclib 1.0.7 when liberasure 1.0.6 was installed | 23:45 |
| clayg | tsg_: but everything works when I build and install liberasure 1.0.7 from source | 23:45 |
| clayg | tsg_: which I *guess* is what one has to do? | 23:45 |
| *** distroguy is now known as csmart | 23:45 | |
| tsg_ | clayg: yeah that's the correct process | 23:45 |
| tsg_ | clayg: 1) install latest liberasurecode tag 2) install latest pyeclib tag off of "master" | 23:46 |
| clayg | weird... what good is it having pyeclib on pypi if you can't install it without this other project you have to download and install from source? | 23:46 |
| tsg_ | clayg: there are checks in place for liberasurecode-1.0.7 to be honest | 23:46 |
| tsg_ | (in the pyeclib pypi version) | 23:47 |
| tsg_ | not sure why it chose not to build it .. I am looking and nothing seems to have changed in that area recently | 23:48 |
| clayg | tsg_: so I just uninstalled liberasure and verified you can't create a pyeclib driver - then I pip reinstalled pyeclib and it build/installed liberasure 1.0.7 for me | 23:50 |
| clayg | tsg_: so apparently having 1.0.6 installed made upgrading pyeclib not care to rebuild liberasure code - but not having *any* liberasurecode triggered something that make'd something | 23:50 |
| zaitcev | tsg_: looks like I was able to file an issue just fine, at least for liberasurecode: https://bitbucket.org/tsg-/liberasurecode/issue/11/licensing-in-m4 | 23:50 |
| clayg | tsg_: did you find how to make rc like patches? | 23:53 |
| *** wolsen_ is now known as wolsen | 23:59 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!