| *** chlong has quit IRC | 00:01 | |
| *** chlong has joined #openstack-swift | 00:02 | |
| *** chlong has quit IRC | 00:06 | |
| ho | good morning! | 00:10 |
|---|---|---|
| ho | clayg: thanks for the review. I will check your comments on it. | 00:11 |
| clayg | ho: yeah... at somepoint I realized I don't entirely understand the use case - so it may be better to start with some evangelizing | 00:11 |
| clayg | it's much easier to get some reviews on a feature if we can get some people excited about the feature! | 00:11 |
| ho | clayg: thanks! in BP there is a link to a mail-list thread. In the thread i disccussed a use-case with notmyname. http://lists.openstack.org/pipermail/openstack-dev/2014-July/040948.html | 00:14 |
| ho | clayg: i see. i need some evangelizing first :) | 00:16 |
| EmilienM | I have a stupid question and I don't find the answer in the doc. With the API, can I get an object MD5 result? | 00:16 |
| clayg | ho: oh nice, very good then - maybe you just needed to smack me in the head for not reading carefully enough! | 00:17 |
| EmilienM | maybe with ETag in fact. | 00:18 |
| *** Masahiro has joined #openstack-swift | 00:20 | |
| notmyname | EmilienM: the etag (md5 of the object contents) is returned with every read request | 00:20 |
| EmilienM | notmyname: excellent, it helps me to do what I wanted, thanks for confirmation ! | 00:22 |
| clayg | ho: ahh - this one is great! include this link in the commit: http://lists.openstack.org/pipermail/openstack-dev/2014-July/041076.html, the following like 4 or 5 emails sum up the issues pretty well - although there was no obvious concensus in the thread :\ | 00:22 |
| clayg | I think maybe now that we have some code up it might be a good time to dust off the old thread and see if we can get anyone to bite - I think i have some questions and wouldn't mind some smart swift devs or other swift operaters with different experience to think about and chime in on | 00:23 |
| *** Masahiro has quit IRC | 00:24 | |
| EmilienM | notmyname: I go for another stupid question. Does it work also for write requests? | 00:27 |
| EmilienM | If-match -> upload | 00:27 |
| ho | clayg: thanks. i will summarize the thread as a lure for getting interest :) | 00:28 |
| *** chlong has joined #openstack-swift | 00:29 | |
| ho | clayg: i talked with operators for maintenance of swift nodes. i will write their concerns on it. | 00:38 |
| *** km__ has joined #openstack-swift | 00:43 | |
| *** km has quit IRC | 00:43 | |
| *** addnull has joined #openstack-swift | 00:49 | |
| openstackgerrit | Samuel Merritt proposed openstack/swift: Make ThreadPools deallocatable. https://review.openstack.org/145647 | 00:55 |
| torgomatic | otherjon: swifterdarrell: ^^ | 00:56 |
| *** dank has quit IRC | 01:04 | |
| openstackgerrit | Samuel Merritt proposed openstack/swift: Make ThreadPools deallocatable. https://review.openstack.org/145647 | 01:16 |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Fix large out of sync out of date containers https://review.openstack.org/141019 | 01:23 |
| *** gyee has quit IRC | 01:36 | |
| *** nosnos has joined #openstack-swift | 01:54 | |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Remove the X-Newest pre-flight request on X-Timestamp https://review.openstack.org/103778 | 02:01 |
| *** haomaiwang has joined #openstack-swift | 02:08 | |
| *** Masahiro has joined #openstack-swift | 02:09 | |
| openstackgerrit | Merged openstack/swift: Substituted object storage paragraph with simple definition https://review.openstack.org/145294 | 02:11 |
| *** chlong has quit IRC | 02:11 | |
| *** SkyRocknRoll has joined #openstack-swift | 02:12 | |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Remove the X-Newest pre-flight request on X-Timestamp https://review.openstack.org/103778 | 02:16 |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Add support for multiple container-reconciler https://review.openstack.org/103779 | 02:22 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift: Updated from global requirements https://review.openstack.org/88736 | 02:22 |
| *** fandi has quit IRC | 02:42 | |
| *** lcurtis has joined #openstack-swift | 03:03 | |
| *** bill_az has quit IRC | 03:22 | |
| *** annegent_ has joined #openstack-swift | 03:24 | |
| *** reed has quit IRC | 03:38 | |
| *** nosnos has quit IRC | 03:43 | |
| *** mahatic has quit IRC | 03:44 | |
| *** lihkin has joined #openstack-swift | 03:45 | |
| *** chlong has joined #openstack-swift | 03:57 | |
| *** k_ has joined #openstack-swift | 03:59 | |
| *** bkopilov has quit IRC | 04:01 | |
| *** chlong has quit IRC | 04:02 | |
| *** ppai has joined #openstack-swift | 04:06 | |
| *** chlong has joined #openstack-swift | 04:18 | |
| *** fandi has joined #openstack-swift | 04:22 | |
| *** fandi has quit IRC | 04:23 | |
| *** fandi has joined #openstack-swift | 04:24 | |
| *** fandi has quit IRC | 04:24 | |
| *** fandi has joined #openstack-swift | 04:26 | |
| *** annegent_ has quit IRC | 04:32 | |
| *** nosnos has joined #openstack-swift | 04:36 | |
| *** lcurtis has quit IRC | 04:39 | |
| *** david-lyle has quit IRC | 04:50 | |
| notmyname | good evening | 04:53 |
| ho | notmyname: hi! | 04:56 |
| mattoliverau | notmyname: good evening | 05:05 |
| *** annegent_ has joined #openstack-swift | 05:06 | |
| notmyname | mattoliverau: good news! I have a the rough draft of my title slide done for my lca talk ;-) (proof! http://d.not.mn/lca_prep_screenshot.png) | 05:09 |
| notmyname | mattoliverau: so you could say things are getting serious ;-) | 05:10 |
| mattoliverau | notmyname: nice, good to see things are coming along and not last minute :P | 05:10 |
| notmyname | lol @ ~40 hours before my plane takes off being "not last minute" | 05:11 |
| mattoliverau | True, doing it during the openstack mini-conf is last minute :P | 05:12 |
| zaitcev | make sure to pack earplugs | 05:12 |
| mattoliverau | zaitcev: what for the conference? us Aussies aren't that noisy.. ok maybe we are but still :P | 05:13 |
| mattoliverau | notmyname: I don't know if I want to hear the answer to this but.. with cats in the title of a talk, how many lol cat references will there be :P | 05:14 |
| notmyname | mattoliverau: all of them | 05:16 |
| notmyname | obviously | 05:16 |
| mattoliverau | lol | 05:18 |
| *** bill_az has joined #openstack-swift | 05:30 | |
| notmyname | hmmm...organizing this talk isn't as easy as it has been for other talks | 05:35 |
| *** nshaikh has joined #openstack-swift | 05:36 | |
| notmyname | mattoliverau: we may need to visit http://mexico.net.nz/menu/tequila next week | 05:44 |
| mattoliverau | notmyname: I'm sure we can find time for that ;) | 05:45 |
| *** annegent_ has quit IRC | 05:52 | |
| ho | the following url is the backgroud of the change https://review.openstack.org/#/c/133155/ (FQDN in Rings). i think this is a good enhancement for realization of swift system's DevOps | 05:52 |
| ho | http://paste.openstack.org/show/155943/ | 05:52 |
| ho | I would like to have your feedback! | 05:53 |
| notmyname | that's a really big patch | 05:55 |
| *** lihkin has quit IRC | 05:57 | |
| *** ppai has quit IRC | 05:58 | |
| *** addnull has quit IRC | 05:58 | |
| notmyname | ho: just curious, but emc was interested in this functionality a while back (identifying a node in a ring by hostname instead of IP). have you had any conversations with them about this? | 05:58 |
| ho | notmyname: I talked with previous owner of this idea, he is an engineer of emc. | 06:00 |
| notmyname | ah cool | 06:00 |
| ho | notmyname: his name is "Co-Authored-By: Alex Pecoraro <alex.pecoraro@emc.com>" in the commit message | 06:02 |
| ho | notmyname: fyi. | 06:02 |
| notmyname | oh, yeah. missed that :-) | 06:03 |
| notmyname | ho: is this something you've got interest in a fujitsu or ...? IOW, how did you get involved in writing this functionality? :-) | 06:09 |
| *** ppai has joined #openstack-swift | 06:12 | |
| ho | notmyname: not decided yet. but i think i can offer this func for our users after i get approval :) | 06:12 |
| notmyname | :-) | 06:12 |
| notmyname | I actually didn't know that you were using swift at fujitsu. is it a public cloud? part of a storage product sold to companies? used internally for stuff? | 06:13 |
| mattoliverau | ho: before I was at Rackspace I worked at Fujitsu (in Australia).. wish I knew I could have worked on swift back then! | 06:15 |
| ho | notmyname: as for public cloud, we decided to use Openstack so we will use swift also. | 06:16 |
| notmyname | cool! | 06:16 |
| notmyname | ho: are you in prod yet or are you still building out the product? | 06:16 |
| ho | notmyname: really? you worked at FAST. sorry i dont understand prod mean... | 06:19 |
| notmyname | prod == production | 06:19 |
| *** annegentle has joined #openstack-swift | 06:19 | |
| notmyname | and that was mattoliverau who worked there | 06:19 |
| mattoliverau | ho: yeah I worked there :) notmyname has never worked in Australia, but he would be very welcome too ;) | 06:21 |
| ho | notmyname: under developing for public cloud not product yet | 06:21 |
| notmyname | ho: ah ok. is it part of http://welcome.globalcloud.global.fujitsu.com/index.html ? | 06:21 |
| notmyname | mattoliverau: ya, but I hear your internet connections are pretty terrible. but I do like most things about australia that I've seen.... :-) | 06:22 |
| *** lihkin has joined #openstack-swift | 06:22 | |
| ho | notmyname: i'm searching public info. wait a minutes | 06:23 |
| *** addnull has joined #openstack-swift | 06:23 | |
| mattoliverau | notmyname: yeah, damn slow internet.. downside of being such a large country with so little people largely dispurst :( | 06:23 |
| mattoliverau | some places you can get decent speed.. but still rather evensive and limited. At least hwne compared to Japan and probably San Fran. | 06:25 |
| mattoliverau | Well I'm going to go make a start on dinner before the wife arrives home, still around if I'm needed. :) | 06:26 |
| *** lihkin has quit IRC | 06:29 | |
| *** annegentle has quit IRC | 06:29 | |
| *** bkopilov has joined #openstack-swift | 06:33 | |
| *** nshaikh has quit IRC | 06:41 | |
| ho | notmyname: sorry. i'm not allowed to say plans for swift usage at this time. i only found the info. http://www.businesscloudnews.com/2014/07/14/fujitsu-announces-2bn-cloud-offensive/ | 06:43 |
| notmyname | lol, ok :-) | 06:44 |
| notmyname | thanks | 06:44 |
| ho | notmyname: sorry for that. | 06:47 |
| notmyname | ho: no, I understand. and "Fujitsu has committed $2bn to build a cloud on openstack and is contributing to swift" is enough of a good story for now :-) | 06:48 |
| ho | notmyname: thanks! | 06:50 |
| *** silor has joined #openstack-swift | 06:57 | |
| notmyname | ok, ideas and rough outline transferred from notes to presentation. still a ton of work to do on this talk, but I'm done for today | 07:02 |
| notmyname | good night | 07:02 |
| notmyname | well, actually... | 07:02 |
| ho | notmyname: good night! | 07:03 |
| notmyname | ho: I'm really glad that you and, by proxy, fujitsu, are contributing to swift. that's really cool. | 07:03 |
| notmyname | ho: I fully understand not being able to share details, so I'm not worried by that at all | 07:03 |
| *** zaitcev has quit IRC | 07:03 | |
| notmyname | but please let us know how we can help and what you are seeing in you experience | 07:03 |
| notmyname | because even if I've seen different stuff and someone else has a few different experiences, yours will be unique and the whole community is better when we can understand your use case too | 07:04 |
| notmyname | and also, it's just really cool and motivating to hear about and see how others are using swift. so don't be shy :-) | 07:05 |
| notmyname | the positive good stories are just as important (and maybe more so) than the bugs | 07:05 |
| notmyname | ...and now I'm going to bed :-) | 07:07 |
| ho | notmyname: thanks a lot! and good night! | 07:07 |
| *** bill_az has quit IRC | 07:13 | |
| *** ppai has quit IRC | 07:23 | |
| *** annegent_ has joined #openstack-swift | 07:25 | |
| *** annegent_ has quit IRC | 07:31 | |
| *** addnull has quit IRC | 07:31 | |
| ho | notmyname: i try to dump my experience (background of my thought) thanks for advice:) | 07:31 |
| *** ppai has joined #openstack-swift | 07:35 | |
| *** k4n0 has joined #openstack-swift | 07:38 | |
| *** k4n0 has quit IRC | 07:42 | |
| *** addnull has joined #openstack-swift | 07:43 | |
| *** chlong has quit IRC | 07:47 | |
| *** nshaikh has joined #openstack-swift | 07:51 | |
| *** rledisez has joined #openstack-swift | 08:14 | |
| *** delattec has joined #openstack-swift | 08:21 | |
| *** geaaru has joined #openstack-swift | 08:22 | |
| *** cdelatte has quit IRC | 08:23 | |
| *** ppai has quit IRC | 08:25 | |
| *** annegentle has joined #openstack-swift | 08:25 | |
| *** annegentle has quit IRC | 08:30 | |
| *** silor has quit IRC | 08:33 | |
| *** ppai has joined #openstack-swift | 08:37 | |
| openstackgerrit | Prashanth Pai proposed openstack/swift: fsync() on directories https://review.openstack.org/126923 | 08:47 |
| *** addnull has quit IRC | 09:06 | |
| *** jistr has joined #openstack-swift | 09:10 | |
| *** acoles_away is now known as acoles | 09:11 | |
| *** jordanP has joined #openstack-swift | 09:12 | |
| *** addnull has joined #openstack-swift | 09:14 | |
| *** addnull has quit IRC | 09:20 | |
| *** annegentle has joined #openstack-swift | 09:25 | |
| *** ahonda has quit IRC | 09:28 | |
| *** annegentle has quit IRC | 09:30 | |
| *** nellysmitt has joined #openstack-swift | 09:45 | |
| *** jordanP has quit IRC | 09:45 | |
| *** jordanP has joined #openstack-swift | 09:45 | |
| *** silor has joined #openstack-swift | 10:04 | |
| *** silor has quit IRC | 10:09 | |
| openstackgerrit | Merged openstack/swift: Add tests for unavailability of `tee` and `splice` in `libc` https://review.openstack.org/142836 | 10:19 |
| *** silor has joined #openstack-swift | 10:24 | |
| *** annegent_ has joined #openstack-swift | 10:25 | |
| *** Masahiro has quit IRC | 10:27 | |
| *** annegent_ has quit IRC | 10:30 | |
| *** haomaiwang has quit IRC | 10:56 | |
| *** annegent_ has joined #openstack-swift | 11:25 | |
| *** NM has joined #openstack-swift | 11:30 | |
| *** annegent_ has quit IRC | 11:30 | |
| *** mahatic has joined #openstack-swift | 11:31 | |
| *** mahatic has quit IRC | 11:41 | |
| *** chlong has joined #openstack-swift | 11:43 | |
| *** mahatic has joined #openstack-swift | 12:03 | |
| ho | good night! | 12:08 |
| *** ho has quit IRC | 12:08 | |
| *** silor has quit IRC | 12:16 | |
| *** Masahiro has joined #openstack-swift | 12:16 | |
| *** Masahiro has quit IRC | 12:20 | |
| *** ppai has quit IRC | 12:22 | |
| *** lucasagomes has joined #openstack-swift | 12:23 | |
| *** sandywalsh has quit IRC | 12:24 | |
| *** km__ has quit IRC | 12:25 | |
| *** k_ has quit IRC | 12:25 | |
| *** kei_yama has quit IRC | 12:27 | |
| *** nosnos has quit IRC | 12:30 | |
| *** ppai has joined #openstack-swift | 12:33 | |
| EmilienM | cschwede: why PUT does not support If-None-Match ? | 12:38 |
| EmilienM | cschwede: because it's not implemented or it's a limitation of HTTP ? | 12:39 |
| *** silor has joined #openstack-swift | 12:42 | |
| cschwede | EmilienM: it’s not fully implemented yet, there were some difficulties with it: https://review.openstack.org/#/c/81646/ | 12:55 |
| *** silor has quit IRC | 12:57 | |
| EmilienM | cschwede: so what is missing here? | 12:57 |
| cschwede | EmilienM: the part in the proxy. i think one of the problems is the eventual consistency, making this a little bit unreliable. for example, if the primary nodes are down and there is an older copy on a fallback node | 13:09 |
| EmilienM | cschwede: fair enough - thanks | 13:11 |
| *** addnull has joined #openstack-swift | 13:22 | |
| *** annegentle has joined #openstack-swift | 13:25 | |
| *** bill_az has joined #openstack-swift | 13:29 | |
| *** addnull has quit IRC | 13:36 | |
| *** tdasilva has joined #openstack-swift | 13:36 | |
| *** Masahiro has joined #openstack-swift | 14:05 | |
| *** NM has quit IRC | 14:07 | |
| *** Masahiro has quit IRC | 14:09 | |
| *** annegentle has quit IRC | 14:25 | |
| *** annegentle has joined #openstack-swift | 14:25 | |
| *** ppai has quit IRC | 14:32 | |
| *** nshaikh has quit IRC | 14:33 | |
| *** lihkin has joined #openstack-swift | 14:39 | |
| *** SkyRocknRoll has quit IRC | 14:41 | |
| *** miqui_ has joined #openstack-swift | 14:41 | |
| *** lcurtis has joined #openstack-swift | 14:53 | |
| *** SkyRocknRoll has joined #openstack-swift | 14:54 | |
| openstackgerrit | Romain LE DISEZ proposed openstack/swift: Support of the Linux socket option SO_REUSEPORT https://review.openstack.org/137659 | 14:55 |
| *** NM has joined #openstack-swift | 15:06 | |
| *** lpabon has joined #openstack-swift | 15:14 | |
| *** annegent_ has joined #openstack-swift | 15:14 | |
| *** annegentle has quit IRC | 15:14 | |
| *** lihkin has quit IRC | 15:32 | |
| openstackgerrit | paul luse proposed openstack/swift: New hashes.pkl format and other reconstructor changes https://review.openstack.org/131872 | 15:42 |
| peluse | clayg, ^ removes the 'reconstruct before send' deal... all on the fly now :) | 15:43 |
| *** annegent_ has quit IRC | 15:44 | |
| *** annegentle has joined #openstack-swift | 15:45 | |
| *** elambert has joined #openstack-swift | 15:50 | |
| tdasilva | peluse: hi, I remember that before storage policies was finalized there was an idea to have 'types' defined in swift.conf, which was then removed. Is that being brought back with EC? | 15:53 |
| *** Masahiro has joined #openstack-swift | 15:53 | |
| *** annegent_ has joined #openstack-swift | 15:55 | |
| *** Masahiro has quit IRC | 15:58 | |
| *** annegentle has quit IRC | 16:00 | |
| tdasilva | peluse: got it, it's already there in the feature/ec branch, cool! | 16:11 |
| *** nshaikh has joined #openstack-swift | 16:16 | |
| *** lpabon_ has joined #openstack-swift | 16:20 | |
| *** chlong has quit IRC | 16:34 | |
| thurloat | notmyname: where is the best place to run the object-expirer daemon? | 16:34 |
| thurloat | separate service, or with the proxy server? | 16:35 |
| peluse | tdasilva, sorry, stepped away for a minute but you got the answer right :) | 16:36 |
| *** gyee has joined #openstack-swift | 16:39 | |
| lcurtis | question for swift experts. Have decided to go with swift on zfs, where each 'dev' will be a zfs dataset/volume. Should I treat each zfs dev as a single disk even though it will be ~ 100T in determine my part power? | 16:41 |
| *** nellysmitt has quit IRC | 16:44 | |
| *** dmsimard_away is now known as dmsimard | 16:45 | |
| *** lpabon_ has quit IRC | 16:51 | |
| *** reed has joined #openstack-swift | 16:55 | |
| *** EmilienM is now known as EmilienM|afk | 17:15 | |
| *** nshaikh has quit IRC | 17:18 | |
| ahale_ | I havent used zfs with swift lcurtis but have been thinking about it recently, I was tending towards lots of zpools each of single devices.. otherwise with any kind of raid it gets expensive or complicated by storage policy.. i think.. | 17:28 |
| notmyname | good morning | 17:28 |
| openstackgerrit | Romain LE DISEZ proposed openstack/swift: Support of the Linux socket option SO_REUSEPORT https://review.openstack.org/137659 | 17:29 |
| *** rledisez has quit IRC | 17:29 | |
| *** pberis has quit IRC | 17:30 | |
| *** pberis has joined #openstack-swift | 17:30 | |
| *** SkyRocknRoll has quit IRC | 17:34 | |
| lcurtis | ahale...interesting thought | 17:34 |
| lcurtis | why do you say it gets expensive | 17:34 |
| notmyname | raid rebuild (or zpool repairs) really kill performance | 17:35 |
| lcurtis | okay...understood | 17:36 |
| lcurtis | yeah that makes sense | 17:36 |
| *** jistr has quit IRC | 17:43 | |
| *** Masahiro has joined #openstack-swift | 17:43 | |
| ahale_ | yeah that and as well if you're making a big raid volume you're using more disks for the same replicas, or drop replica count but then you lose some availability when stuffs down | 17:47 |
| *** Masahiro has quit IRC | 17:48 | |
| ahale_ | we were discussing zfs for its "not xfs" features rather than any of the cool stuff | 17:48 |
| notmyname | ahale_: xfs: the worst filesystem for swift except for all the others? | 17:49 |
| ahale_ | :) | 17:49 |
| notmyname | ahale_: have you played with running on object server per drive? | 17:50 |
| *** fandi has quit IRC | 17:50 | |
| ahale_ | nope - thought about it a bit but its a lot of processes | 17:50 |
| *** annegent_ has quit IRC | 17:51 | |
| ahale_ | daniele was playing with reproducing some of the nasty hangs we get when umounting horribly broken filesystems, we ended up in https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/tree/fs/xfs/xfs_trans_ail.c?id=refs/tags/v3.18.1#n592 | 17:53 |
| swifterdarrell | ahale_: notmyname: seems like you can choose between A) blocking I/O syscalls interfering with eventlet's hub; or B) lots of OS threads or processes to at least isolate the stalls to requests depending on the same source of blocking (physical device) | 17:53 |
| ahale_ | interestingly zfs was able to umount without sending everything uninteruptable | 17:54 |
| ahale_ | yeah - what a choice :) | 17:54 |
| lcurtis | on that note....if you made big volume and treated it as a drive, would swift therefore NOT place 2 replicas on same drive? | 17:54 |
| notmyname | lcurtis: ya, swift actually maps objects to a particular mount point. but we often say "drives" cause that's the strongly recommended way to do it in almost every situation | 17:55 |
| notmyname | lcurtis: but technically, it's "storage volume at a mount point" | 17:55 |
| swifterdarrell | ahale_: notmyname: i haven't been able to put my money where my mouth is, but I think: normal deployment << one-obj-server-per-disk << thread-pool-per-disk | 17:56 |
| swifterdarrell | ahale_: notmyname: IOW, extra processes involved would still be better than the standard deployment | 17:56 |
| lcurtis | notmyname: thank you | 17:57 |
| swifterdarrell | ahale_: notmyname: I know the thread-pool-per-disk has overhead that doesn't really make it worthwhile :-/ | 17:59 |
| ahale_ | can't be sure, but I have a funny recollection we already looked at the thread pool and it didn't work out well or something .. | 17:59 |
| swifterdarrell | ahale_: notmyname: wouldn't surprise me ;) | 18:00 |
| notmyname | ahale_: probably not with you 90 drive servers ;-) | 18:00 |
| ahale_ | yeah :) | 18:00 |
| swifterdarrell | ahale_: notmyname: there, you get obj-serv-count * disk-count * threads-per-disk number of OS threads... to keep that sane, it drives down the other two factors with dense servers :( | 18:00 |
| swifterdarrell | ahale_: notmyname: and even if you don't have >1k OS threads, the per-I/O-request overhead is higher going through Queues. | 18:01 |
| *** jordanP has quit IRC | 18:01 | |
| *** geaaru has quit IRC | 18:01 | |
| *** fandi has joined #openstack-swift | 18:02 | |
| *** gyee has quit IRC | 18:08 | |
| *** elambert has quit IRC | 18:12 | |
| *** EmilienM|afk is now known as EmilienM | 18:14 | |
| *** elambert has joined #openstack-swift | 18:15 | |
| lcurtis | as far as partition power goes...say if I do 2^18...that is 262144 partitions...then say 100 partitions per 'volume', would be enough for 2600 servers | 18:22 |
| lcurtis | ? | 18:22 |
| lcurtis | 2612 | 18:22 |
| *** elambert has quit IRC | 18:24 | |
| *** acoles is now known as acoles_away | 18:26 | |
| *** elambert has joined #openstack-swift | 18:26 | |
| notmyname | lcurtis: servers or drives? | 18:28 |
| notmyname | drive count is what matters | 18:28 |
| lcurtis | sorry...drives | 18:30 |
| lcurtis | aka the zfs conundrum | 18:30 |
| mahatic | notmyname, good morning | 18:33 |
| mahatic | notmyname, where do i find what replication server's allowed headers are? | 18:35 |
| mahatic | is there any code that says so? | 18:35 |
| notmyname | lcurtis: part power of 16 or 17 would be fine with 2600 drives | 18:35 |
| lcurtis | do replicas play into the equation? | 18:36 |
| notmyname | mahatic: look in the __call__ method to see when HTTPMethodNotAllowed is returned | 18:36 |
| lcurtis | is it literally 2^x /100 / # drives | 18:36 |
| notmyname | lcurtis: ya. 2**<part power> * <replica count> / <drives> | 18:37 |
| lcurtis | okay...thanks so much...reading various docs and still struggling to understand it all | 18:37 |
| *** lpabon has quit IRC | 18:39 | |
| mahatic | notmyname, ah okay, thanks! | 18:40 |
| lcurtis | or 2^16 * 3 replicas / 100 partitions = 1966 drives? | 18:42 |
| notmyname | ya. but the point is to have the ability to spread out partitions so that a partition is a small percentage of the total drive space. so 100 partitions on a drive means each partition is about 1% of the space. part power of 16 with 2600 drives is about 75 parts per drive | 18:44 |
| *** nellysmitt has joined #openstack-swift | 18:45 | |
| lcurtis | okay..that makes sense | 18:46 |
| notmyname | lcurtis: the risk is that if you have a too high part power, especially if you have a smaller cluster (eg initially), you'll end up with 10k or more directories on a drive. and that's really bad for performance and efficient use of the space | 18:49 |
| lcurtis | yes exactly | 18:49 |
| lcurtis | i get that | 18:49 |
| *** nellysmitt has quit IRC | 18:50 | |
| lcurtis | but for some reason i dont get if you have fewer partitions 'your data distribution will be more uneven than you want. You'll have to keep more free space on your disks in order to avoid their getting full.' | 18:50 |
| *** elambert has quit IRC | 18:51 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift: Updated from global requirements https://review.openstack.org/88736 | 18:51 |
| notmyname | lcurtis: objects are placed based on the hash of their name | 18:55 |
| notmyname | which means that for a lot of objects and with a good hash function (which swift uses) you'll get roughly the same number of objects in each partition | 18:55 |
| notmyname | and the placement isn't to a drive. objects are mapped to a partition and the partitions are assigned to drives | 18:56 |
| notmyname | note that the size of the object doesn't factor into the placement at all | 18:56 |
| notmyname | so if you are terribly unlucky, you could end up with all of your 2-byte objects mapping to the partitions assigned to drive A, but all of your 5GB objects mapping to the partitions assigned to drive B | 18:57 |
| notmyname | which would result in uneven filling | 18:57 |
| notmyname | of the drive space (ie as reported by df) | 18:57 |
| notmyname | now, it's probably more likely that you'll win ever state's lottery every week for the rest of your life than that happening, but you know, technically it's possible | 18:58 |
| ahale_ | i would love if you could double the total partition count | 18:58 |
| notmyname | ahale_: yes, that would be nice | 18:58 |
| notmyname | lcurtis: so if you have a few number of partitions on a drive, it means that each partitions size (ie what's reported by du for that directory) is more impactful for the drive's fullness | 18:59 |
| notmyname | and if you have fewer partitions overall, they will be more uneven in capacity used, based on how homogenous the size objects in your clusters are | 19:00 |
| notmyname | so all that works out to: less partitions on a drive can make the relative drive fullness more uneven. | 19:01 |
| lcurtis | sigh of relief... | 19:01 |
| lcurtis | yep...okay...got it | 19:01 |
| notmyname | it also works out to: swift is better the bigger it gets | 19:01 |
| lcurtis | you seriously rock...have not had much sleep in trying to set this all up...lol..so having hard time grasping the why | 19:04 |
| mahatic | notmyname, I'm not sure if you are following the mail thread (OPW meeting), but tomorrow 5pm UTC there is a meeting update scheduled. You can drop by if you are free/possible. | 19:06 |
| mahatic | in #openstack-opw | 19:07 |
| notmyname | mahatic: ah, no I hadn't seen that | 19:08 |
| notmyname | mahatic: I'm not sure if I'll make it. hard to say at this point. I'm flying out tomorrow night, so I'll be doing a lot of last minute prep tomorrow | 19:09 |
| mahatic | notmyname, ah okay. linuxcon I believe? No problem | 19:09 |
| notmyname | ya. LCA in new zealand | 19:10 |
| mahatic | okay no, linuxcon is at the end of the month. My bad | 19:10 |
| mahatic | oh okay. great | 19:10 |
| *** elambert has joined #openstack-swift | 19:15 | |
| notmyname | ah. 5pm UTC | 19:16 |
| mahatic | yup | 19:17 |
| *** elambert has quit IRC | 19:18 | |
| *** elambert has joined #openstack-swift | 19:19 | |
| lcurtis | and min_part_hours can be changed? | 19:19 |
| *** Masahiro has joined #openstack-swift | 19:31 | |
| torgomatic | lcurtis: yes | 19:32 |
| lcurtis | thanks! | 19:33 |
| lcurtis | okay...pushing <enter> | 19:33 |
| lcurtis | or at least having one of my guys do the puppet run | 19:34 |
| lcurtis | ;) | 19:34 |
| *** Masahiro has quit IRC | 19:35 | |
| *** mahatic has quit IRC | 19:42 | |
| *** fandi has quit IRC | 20:13 | |
| *** zaitcev has joined #openstack-swift | 20:19 | |
| *** ChanServ sets mode: +v zaitcev | 20:19 | |
| *** fifieldt_ has quit IRC | 20:32 | |
| *** fifieldt_ has joined #openstack-swift | 20:33 | |
| *** nellysmitt has joined #openstack-swift | 20:46 | |
| *** nellysmitt has quit IRC | 20:51 | |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Add dispersion command to swift-ring-builder https://review.openstack.org/144432 | 20:52 |
| *** david-lyle has joined #openstack-swift | 20:54 | |
| *** lpabon has joined #openstack-swift | 20:58 | |
| *** rdaly2 has joined #openstack-swift | 20:59 | |
| *** NM has quit IRC | 21:01 | |
| *** mariusv__ is now known as mariusv | 21:12 | |
| *** mariusv is now known as Guest33665 | 21:13 | |
| *** Masahiro has joined #openstack-swift | 21:20 | |
| *** Masahiro has quit IRC | 21:25 | |
| mattoliverau | Morning | 21:25 |
| *** angelastreeter has joined #openstack-swift | 21:27 | |
| notmyname | hi | 21:27 |
| notmyname | mattoliverau: any chance that AU and NZ currency are the same? | 21:28 |
| * notmyname has some AU$ from the last trip... | 21:28 | |
| *** gyee has joined #openstack-swift | 21:28 | |
| mattoliverau | notmyname: nope, unfortunately not.. Its come up many times over the decades but NZ would never do it. You'll just have to come back to Oz at a later time ;) | 21:31 |
| notmyname | heh ok | 21:32 |
| *** lpabon has quit IRC | 21:32 | |
| notmyname | also, why is Oz a nickname/abbreviation for Australia? I've always wondered that | 21:32 |
| notmyname | is it from aussie? shortened to oz? | 21:33 |
| clayg | notmyname: it's because it's so fucking weird and trippy down there | 21:33 |
| clayg | notmyname: you know they have the only mammels that lay eggs right? | 21:33 |
| mattoliverau | clay one lays eggs and has venom | 21:37 |
| mattoliverau | notmyname: short of Aussie... We're just lazy, if we could shortening even more we would :p | 21:38 |
| clayg | lol | 21:39 |
| clayg | cschwede: idk, i keep trying to make set_overload take a % but I'm not really liking how it's turning out | 21:40 |
| clayg | cschwede: but since the dispersion graph talks about dispersion as a % it seems like allowing the operator to set overload as a percent makes sense | 21:41 |
| *** takotuesday has joined #openstack-swift | 22:05 | |
| takotuesday | hey guys, is anyone familiar with swiftclient api for python? | 22:05 |
| notmyname | takotuesday: ya, what's up? | 22:05 |
| takotuesday | notmyname: awesome, Im looking for some examples of usage. Im wanting to have my app save some generated files to a swift container | 22:07 |
| notmyname | ok, cool. what have you tried so far? | 22:07 |
| takotuesday | just reading through the api | 22:07 |
| clayg | from swiftclient.client import get_auth, put_object; token, url = get_auth(**creds); put_object(token, url, container_name, open_file_like_object_string_or_generator) | 22:08 |
| clayg | hrmm... no there's probably a object_name param in there somewhere | 22:09 |
| clayg | but that's the general idea | 22:09 |
| clayg | if you use a Connection object there's automatic retries and stuff | 22:09 |
| clayg | oh, this isn't terrible -> http://docs.openstack.org/developer/python-swiftclient/swiftclient.html#module-swiftclient.client | 22:10 |
| *** angelastreeter has quit IRC | 22:11 | |
| takotuesday | so im relatively new to python | 22:13 |
| takotuesday | here let me type up some code of what im thinking | 22:13 |
| torgomatic | takotuesday: in case you're new to IRC as well, let me advise you to use gist.github.com or pastebin.com or any other such thing instead of pasting multiline code in the channel | 22:14 |
| clayg | torgomatic: what are you saying? | 22:15 |
| clayg | :P | 22:15 |
| * clayg is new to irc | 22:16 | |
| * torgomatic whistles innocently ;) | 22:16 | |
| *** EmilienM is now known as EmilienM|afk | 22:17 | |
| clayg | why is so hard to just calculate what overload would need to be to solve the ring?! | 22:18 |
| takotuesday | http://paste.debian.net/hidden/711d8860/ | 22:19 |
| clayg | i hate math, computers are stupid | 22:19 |
| takotuesday | this is where im starting | 22:19 |
| takotuesday | doesnt make sense to store the swiftclient obj in a variable | 22:19 |
| notmyname | takotuesday: here's a small script I wrote recently using python-swiftclient https://github.com/notmyname/directory_uploader/blob/master/uploader.py | 22:20 |
| takotuesday | notmyname: thanks, that really helps | 22:22 |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Add dispersion command to swift-ring-builder https://review.openstack.org/144432 | 22:23 |
| *** lcurtis has quit IRC | 22:26 | |
| *** david-lyle has quit IRC | 22:28 | |
| *** delattec has quit IRC | 22:43 | |
| *** nellysmitt has joined #openstack-swift | 22:47 | |
| takotuesday | can you send a path to put_object | 22:48 |
| takotuesday | for obj, or do you have to send individual files | 22:49 |
| *** nellysmitt has quit IRC | 22:51 | |
| *** rdaly2 has quit IRC | 22:57 | |
| *** Masahiro has joined #openstack-swift | 23:09 | |
| *** Masahiro has quit IRC | 23:13 | |
| *** IRTermite has quit IRC | 23:15 | |
| *** tdasilva has quit IRC | 23:21 | |
| zaitcev | https://review.openstack.org/145809 what madness is this | 23:22 |
| zaitcev | "The implementation is fairly straightforward with the exception of building the ring. I've followed an upstream TripleO model here where we build the actual ring on each node (rather than build once and rsync). This works because Heat will always know all the devices ahead of time." | 23:22 |
| zaitcev | well, as long as it works | 23:22 |
| torgomatic | as long as they set the same random seed on each node, it'll be okay | 23:23 |
| torgomatic | if they've forgotten that, then hilarity will ensue | 23:24 |
| mattoliverau | takotuesday: looking at the code, you pass it a file like object or the contents of the file. | 23:27 |
| *** takotuesday has quit IRC | 23:31 | |
| openstackgerrit | Clay Gerrard proposed openstack/swift: Add some input options for set_overload https://review.openstack.org/145970 | 23:34 |
| *** erlon_ has quit IRC | 23:40 | |
| notmyname | zaitcev: I see you left a comment. I just did too | 23:41 |
| zaitcev | notmyname: thanks | 23:41 |
| notmyname | zaitcev: thanks for bringing it up | 23:41 |
| zaitcev | I'm not convinced it's ideal, out of the general principle where automation is doing something different from what we expect an administrator to do. But I do not have a reasoned objection. | 23:42 |
| zaitcev | notmyname: I appropriated your explanation yesterday here - http://zaitcev.livejournal.com/226242.html - and wasn't sure what to use as your Internet URL identity, so I used Swiftstack's blog | 23:43 |
| notmyname | Piston was the one who first wanted that feature. they were wanting to build rings concurrently on their different servers. (....then they stopped using swift altogether.....but that's a different story) | 23:44 |
| zaitcev | sad_panda.png | 23:44 |
| notmyname | zaitcev: ah you used some author id I have with openstack.org. If you can change it, I'd prefer http://not.mn | 23:45 |
| zaitcev | excellent, thank you | 23:45 |
| *** dmsimard is now known as dmsimard_away | 23:50 | |
| zaitcev | http://programmerthoughts.com/programming/deleted-file-recovery/ - oh good lord, this reminded me about our yesterday conversation | 23:52 |
| zaitcev | Why is it that we want to move those to special containers. Can't we simply write .ts, but do not unlink and let auditor do it with a special API call at a configurable interval? | 23:53 |
| zaitcev | Seems like almost zero effort, except the actual undelete that requires work. But it should be bug-free. | 23:54 |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!
