| *** gyee has quit IRC | 00:04 | |
| not_me | Thank you for your answers clayg! | 00:17 |
|---|---|---|
| clayg | not_me: oh - did you figure something out? | 00:18 |
| clayg | Well... keep at it. Let us know what you figure out! | 00:21 |
| not_me | I reduced the replication pass time by 10 by increasing the object-replicator concurrency from 2 to 16 | 00:23 |
| not_me | from 1.4 partition/sec to 11.27 partition/sec | 00:24 |
| not_me | I'm still trying to understand this better though. I'm still unclear on why it's still slower than my other rings. | 00:26 |
| *** vint_bra has joined #openstack-swift | 00:38 | |
| *** tovin07_ has joined #openstack-swift | 00:48 | |
| *** mingyu has quit IRC | 01:02 | |
| *** mingyu has joined #openstack-swift | 01:03 | |
| *** vint_bra has quit IRC | 01:05 | |
| *** cshastri has joined #openstack-swift | 01:05 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 01:12 |
| openstackgerrit | Merged openstack/swift feature/deep: unit tests for _misplaced_objects https://review.openstack.org/509256 | 01:13 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 01:13 |
| *** wes_dillingham has quit IRC | 01:24 | |
| *** hoonetorg has quit IRC | 02:02 | |
| *** afazekas has quit IRC | 02:02 | |
| *** afazekas has joined #openstack-swift | 02:02 | |
| *** hoonetorg has joined #openstack-swift | 02:03 | |
| *** stewie925 has joined #openstack-swift | 02:11 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 02:44 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 02:45 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 03:15 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 03:17 |
| *** abhitechie has joined #openstack-swift | 03:26 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 03:36 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 03:37 |
| *** gkadam has joined #openstack-swift | 03:37 | |
| *** tonyb has quit IRC | 03:42 | |
| *** two_tired has joined #openstack-swift | 03:46 | |
| *** links has joined #openstack-swift | 03:48 | |
| *** tonyb has joined #openstack-swift | 03:56 | |
| *** klrmn has quit IRC | 03:59 | |
| *** SkyRocknRoll has joined #openstack-swift | 03:59 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 04:05 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 04:07 |
| *** abhitechie has quit IRC | 04:12 | |
| *** jamielennox has quit IRC | 04:14 | |
| *** jamielennox has joined #openstack-swift | 04:18 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 04:22 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 04:23 |
| *** two_tired has quit IRC | 04:29 | |
| *** abhitechie has joined #openstack-swift | 04:31 | |
| *** gyee has joined #openstack-swift | 04:33 | |
| *** mingyu has quit IRC | 04:34 | |
| *** mingyu has joined #openstack-swift | 04:35 | |
| *** mingyu has quit IRC | 04:39 | |
| *** mingyu has joined #openstack-swift | 04:44 | |
| *** chlong_ has joined #openstack-swift | 04:50 | |
| *** psachin has joined #openstack-swift | 05:01 | |
| *** gyee has quit IRC | 05:05 | |
| *** abhinavtechie has joined #openstack-swift | 05:12 | |
| *** abhitechie has quit IRC | 05:12 | |
| *** SkyRocknRoll has quit IRC | 05:30 | |
| openstackgerrit | Tim Burke proposed openstack/swift feature/deep: Include deleted objects when moving misplaced objects https://review.openstack.org/509496 | 05:35 |
| openstackgerrit | Tim Burke proposed openstack/swift feature/deep: Eliminate _generate_object_list method https://review.openstack.org/509508 | 05:35 |
| openstackgerrit | Tim Burke proposed openstack/swift feature/deep: Don't lose timestamp info when moving misplaced objects https://review.openstack.org/509589 | 05:35 |
| timburke | acoles: haven't actually looked at p 509589 yet; will take a look tomorrow though | 05:36 |
| patchbot | https://review.openstack.org/#/c/509589/ - swift (feature/deep) - Don't lose timestamp info when moving misplaced ob... | 05:36 |
| *** ChubYann has quit IRC | 05:41 | |
| *** SkyRocknRoll has joined #openstack-swift | 05:46 | |
| *** dhellmann has quit IRC | 05:56 | |
| *** dhellmann has joined #openstack-swift | 05:58 | |
| *** alecuyer has quit IRC | 06:06 | |
| *** rledisez has quit IRC | 06:06 | |
| *** alecuyer has joined #openstack-swift | 06:07 | |
| *** rledisez has joined #openstack-swift | 06:07 | |
| *** chlong_ has quit IRC | 06:13 | |
| *** rcernin has joined #openstack-swift | 06:24 | |
| *** spectr has quit IRC | 06:34 | |
| *** spectr has joined #openstack-swift | 06:34 | |
| *** zaitcev has quit IRC | 06:48 | |
| *** zaitcev has joined #openstack-swift | 06:55 | |
| *** ChanServ sets mode: +v zaitcev | 06:55 | |
| *** silor has joined #openstack-swift | 06:57 | |
| *** abhinavtechie has quit IRC | 07:04 | |
| *** abhitechie has joined #openstack-swift | 07:04 | |
| *** cbartz has joined #openstack-swift | 07:06 | |
| *** tesseract has joined #openstack-swift | 07:07 | |
| *** pcaruana has joined #openstack-swift | 07:08 | |
| *** silor has quit IRC | 07:21 | |
| *** abhitechie has quit IRC | 07:47 | |
| *** geaaru has joined #openstack-swift | 07:49 | |
| *** abhitechie has joined #openstack-swift | 07:51 | |
| *** abhitechie has quit IRC | 08:08 | |
| *** spectr has quit IRC | 08:09 | |
| *** spectr has joined #openstack-swift | 08:10 | |
| *** abhitechie has joined #openstack-swift | 08:16 | |
| *** spectr has quit IRC | 08:17 | |
| *** spectr has joined #openstack-swift | 08:18 | |
| openstackgerrit | Merged openstack/swift feature/deep: Include deleted objects when moving misplaced objects https://review.openstack.org/509496 | 08:24 |
| openstackgerrit | Merged openstack/swift feature/deep: Eliminate _generate_object_list method https://review.openstack.org/509508 | 08:24 |
| *** abhitechie has quit IRC | 08:25 | |
| *** oshritf has joined #openstack-swift | 08:44 | |
| acoles | mattoliverau: thanks for this comment https://review.openstack.org/#/c/509508/1/swift/container/sharder.py@1297, makes sense now! | 09:09 |
| patchbot | patch 509508 - swift (feature/deep) - Eliminate _generate_object_list method (MERGED) | 09:09 |
| acoles | I copied it to a trello card (low priority) for us to consider the business of the PUTs lying in pending file for a period of time | 09:10 |
| acoles | https://trello.com/c/P3ZgtFBV/108-check-shard-range-put-path-to-other-containers | 09:11 |
| acoles | timburke: NP thanks for getting all the others in that chain merged! | 09:12 |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: remove unused method https://review.openstack.org/509723 | 09:17 |
| acoles | timburke: I'm feeling more courageous today ^^ :P | 09:18 |
| *** mvk has quit IRC | 09:28 | |
| openstackgerrit | Kota Tsuyuzaki proposed openstack/swift master: Follow up for symlink implementaion https://review.openstack.org/509149 | 09:28 |
| *** mvk has joined #openstack-swift | 09:47 | |
| *** janonymous has quit IRC | 09:49 | |
| acoles | I have created a 'low hanging fruit' label on https://trello.com/b/z6oKKI4Q/container-sharding and marked a few cards | 09:50 |
| acoles | notmyname: ^^ | 09:50 |
| acoles | go to Show Menu->Filter Cards | 09:51 |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: Don't lose timestamp info when moving misplaced objects https://review.openstack.org/509589 | 09:58 |
| *** tovin07_ has quit IRC | 10:09 | |
| *** HCLTech-SSW has joined #openstack-swift | 10:35 | |
| HCLTech-SSW | Hi | 10:36 |
| rledisez | hi HCLTech-SSW | 10:36 |
| HCLTech-SSW | we have posted few observations over the gerrit | 10:37 |
| rledisez | yeah, i got the notification, i'll have a look this afternoon | 10:40 |
| rledisez | HCLTech-SSW: btw, what timezone are you from? | 10:41 |
| *** abhitechie has joined #openstack-swift | 10:41 | |
| HCLTech-SSW | rledisez: We are from IST timezone | 10:43 |
| *** spectr has quit IRC | 10:54 | |
| *** HCLTech-SSW has quit IRC | 11:03 | |
| *** kei_yama has quit IRC | 11:24 | |
| *** cshastri has quit IRC | 11:40 | |
| *** ianychoi has quit IRC | 11:46 | |
| *** f0o has quit IRC | 12:02 | |
| *** f0o has joined #openstack-swift | 12:19 | |
| *** f0o has quit IRC | 12:33 | |
| kota_ | acoles: still? | 12:33 |
| *** f0o1 has joined #openstack-swift | 12:34 | |
| *** f0o1 has quit IRC | 12:35 | |
| kota_ | it seems acoles not here now. | 12:36 |
| kota_ | it's ok, will ping tommorow again | 12:36 |
| *** f0o1 has joined #openstack-swift | 12:37 | |
| *** f0o has joined #openstack-swift | 12:37 | |
| openstackgerrit | Merged openstack/swift feature/deep: remove unused method https://review.openstack.org/509723 | 12:38 |
| *** f0o1 has quit IRC | 12:44 | |
| *** links has quit IRC | 12:44 | |
| acoles | kota_: sorry I was away for lunch | 12:49 |
| kota_ | acoles! no worries | 12:50 |
| *** catintheroof has joined #openstack-swift | 12:50 | |
| *** catintheroof has quit IRC | 12:51 | |
| *** catintheroof has joined #openstack-swift | 12:51 | |
| kota_ | acoles: quick question. could we use content-type field for specific middleware use case? | 12:52 |
| kota_ | e.g. i'm trying to think of using content-type for symlink likely Content-Type: application/symlink; target=/acc/cont/obj | 12:53 |
| acoles | kota_: do you mean overwrite a client content-type or append to the end of the client content-type? | 12:53 |
| kota_ | yup | 12:53 |
| kota_ | ah | 12:53 |
| kota_ | overwrite a client content-type | 12:53 |
| kota_ | only when it's symlink request | 12:54 |
| kota_ | symlink=put | 12:54 |
| *** gkadam has quit IRC | 12:54 | |
| kota_ | and add the note for docs absolutely. | 12:54 |
| acoles | kota_: I just looked to see if SLO does similar with the manifest, but I think not | 12:55 |
| kota_ | any reasons? | 12:56 |
| *** gkadam has joined #openstack-swift | 12:57 | |
| acoles | kota_: content-type can be updated on a POST | 12:58 |
| acoles | I think | 12:58 |
| kota_ | oh | 12:58 |
| kota_ | exactly | 12:58 |
| kota_ | hmm... | 13:00 |
| acoles | so anything added to content-type on PUT must be also added on POST | 13:00 |
| acoles | we have special case in object server to handle swift_bytes that is appended by slo to content-type but I would not recommend repeating that pattern | 13:01 |
| acoles | but, symlink POST is weird anyway, so maybe? | 13:01 |
| *** gkadam has quit IRC | 13:02 | |
| *** spectr has joined #openstack-swift | 13:02 | |
| acoles | kota_: what is your goal? | 13:02 |
| kota_ | acoles: lol but i think we cannot find if the POST is against to symlink or not. It's not guranteed. | 13:03 |
| kota_ | acoles: the reason I thought it, is... | 13:03 |
| kota_ | https://review.openstack.org/#/c/232162/59/swift/common/middleware/symlink.py@297 | 13:04 |
| patchbot | patch 232162 - swift - Symlink implementation. | 13:04 |
| *** abhitechie has quit IRC | 13:05 | |
| kota_ | I'd like to avoid to use etag space to keep symlink path info. | 13:05 |
| kota_ | hmm... m_kazuhiro commented something, not yet checked. | 13:05 |
| kota_ | and IF we want to know if the object is symlink or not in object listing API (i.e. GET container), content_type seems to be better place I thought. | 13:08 |
| kota_ | not sure if we REALLY want the symlink info at object listing though. | 13:09 |
| kota_ | acoles: and i thought you know more about content-type handling than me :) | 13:09 |
| kota_ | that's why I asked you, acoles | 13:09 |
| *** SkyRocknRoll has quit IRC | 13:10 | |
| acoles | kota_: the goal makes sense | 13:10 |
| acoles | we did append to etag for encryption, need to remind myself what we did there | 13:10 |
| acoles | etag has the advantage of only being set on PUT | 13:10 |
| kota_ | oic | 13:11 |
| kota_ | it cannot be updated by POST | 13:11 |
| kota_ | however, it's also risky for symlink because if the middleware pulled out (removing from pipeline), no middleware translate the hash appendix (symlink info) anymore? | 13:13 |
| kota_ | i think, the reason why encryption could do it, is encryption should not be removed once it was deployed, yeah? | 13:13 |
| acoles | correct | 13:14 |
| kota_ | or, if no special use case for now, removing the staff for now to land it, and then, thinking of how to implement when it is needed later <- it seems good approach to me. | 13:16 |
| kota_ | though we need to ask to primary users, m_kazuhiro, tdasilva, and maybe timburke? | 13:17 |
| openstackgerrit | Merged openstack/swift master: Add account_autocreate=true to internal-client.conf-sample https://review.openstack.org/507447 | 13:17 |
| kota_ | and from the comment, m_kazuhiro doesn't have strong requirement for now... hmmm | 13:18 |
| acoles | kota_: but if the symlink info is not in listing in first version then a client can never assume it is in listing so client will always need to check the object :( | 13:19 |
| * kota_ have to think about the use case and risk if we don't set the info for object listing. | 13:19 | |
| kota_ | acoles: sure, that could be painful | 13:19 |
| acoles | I guess if symlink is added to the listing then a client can be optimised when it finds info in listing | 13:19 |
| kota_ | and should it be into hash space in your opinion? | 13:20 |
| acoles | if the concern is symlink middleware being removed from pipeline, is there a case to have the gatekeeper for example always clean up etag response headers (truncate to length of etag)...hmmm..that scares me | 13:21 |
| acoles | split on ; would be safer | 13:21 |
| acoles | kota_: I need to understand better symlink POST - if it is not possible to POST to a symlink without it ceasing to be a symlink, then maybe content-type is ok to override | 13:22 |
| acoles | ...but swift_bytes was a pain, because it mixes data-state with content-type and the two can change at different times | 13:24 |
| kota_ | acoles: oops. true. | 13:24 |
| kota_ | sigh, I wonder if i could make object-server could handle the POST, like if it's symlink (diskfile has X-Symlink-Target), drop content-type change for container-update. | 13:26 |
| kota_ | probably, I need to dig out and learn swift_bytes and their logics. | 13:26 |
| kota_ | thx acoles, that was helpful! | 13:27 |
| acoles | problem is object server may not have up to date .data file so may not know it should drop the content-type from the update | 13:27 |
| acoles | kota_: swift_bytes is a bad pattern | 13:28 |
| kota_ | ok | 13:28 |
| acoles | X-Object-Sysmeta-Container-Update-Override-Size=swift_bytes would be better if we could rewrite history :) | 13:29 |
| kota_ | acoles: sometimes, we wonder if we had rewrite history ;) | 13:29 |
| kota_ | like PUT+POST for EC :/ | 13:30 |
| kota_ | had be able to | 13:30 |
| kota_ | ah, we wish? no sense for my English. | 13:31 |
| acoles | probably 'wish' but both are funny :) | 13:31 |
| kota_ | i wish i were a bird <- it's the easiest example, students learn at schools. | 13:32 |
| kota_ | acoles: thanks teacher ;) | 13:32 |
| acoles | i wish I knew the answer ;) | 13:33 |
| kota_ | k, anyway, thanks and being offline to prepare to go my bed ;) | 13:34 |
| acoles | good night kota_ | 13:34 |
| kota_ | good night. | 13:34 |
| *** spectr has quit IRC | 13:40 | |
| *** chlong_ has joined #openstack-swift | 13:46 | |
| *** spectr has joined #openstack-swift | 13:57 | |
| *** spectr has quit IRC | 14:07 | |
| *** spectr has joined #openstack-swift | 14:08 | |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: include deleted objects when cleaving https://review.openstack.org/509793 | 14:23 |
| *** vint_bra has joined #openstack-swift | 14:37 | |
| *** oshritf has quit IRC | 14:45 | |
| *** gyee has joined #openstack-swift | 14:51 | |
| *** wes_dillingham has joined #openstack-swift | 14:52 | |
| *** oshritf has joined #openstack-swift | 15:00 | |
| *** cbartz has quit IRC | 15:02 | |
| *** mat128 has joined #openstack-swift | 15:05 | |
| notmyname | good morning | 15:20 |
| *** MVenesio has joined #openstack-swift | 15:23 | |
| *** oshritf has quit IRC | 15:28 | |
| *** chinkara has joined #openstack-swift | 15:28 | |
| *** MVenesio has quit IRC | 15:34 | |
| *** cbartz has joined #openstack-swift | 15:35 | |
| *** MVenesio has joined #openstack-swift | 15:36 | |
| *** chsc has joined #openstack-swift | 15:36 | |
| *** chsc has joined #openstack-swift | 15:36 | |
| *** chinkara has quit IRC | 15:38 | |
| *** pcaruana has quit IRC | 15:39 | |
| timburke | good morning | 15:45 |
| *** spectr has quit IRC | 15:51 | |
| timburke | acoles: kota_: X-Object-Sysmeta-Container-Update-Override-Size=swift_bytes would mess up our account/container bytes-used bookkeeping -- there just isn't a good answer on SLO sizes in container listings :-( | 15:52 |
| acoles | timburke: right. don't worry, it wasn't a serious suggestion. | 15:53 |
| timburke | hey, rewriting history's easy -- just a `git push -f` away ;-) | 15:54 |
| *** chsc has quit IRC | 16:11 | |
| *** cbartz has quit IRC | 16:12 | |
| *** rcernin has quit IRC | 16:13 | |
| *** rcernin has joined #openstack-swift | 16:13 | |
| *** psachin has quit IRC | 16:14 | |
| *** rcernin has quit IRC | 16:14 | |
| *** klrmn has joined #openstack-swift | 16:14 | |
| *** janonymous has joined #openstack-swift | 16:25 | |
| *** tesseract has quit IRC | 16:29 | |
| *** geaaru has quit IRC | 16:29 | |
| *** mvk has quit IRC | 16:33 | |
| *** abhitechie has joined #openstack-swift | 16:33 | |
| *** abhitechie has quit IRC | 16:35 | |
| *** abhitechie has joined #openstack-swift | 16:35 | |
| *** abhitechie has quit IRC | 16:36 | |
| *** abhitechie has joined #openstack-swift | 16:37 | |
| *** abhitechie has quit IRC | 16:38 | |
| openstackgerrit | Tim Burke proposed openstack/swift feature/deep: Add Container Sharder to source docs https://review.openstack.org/509830 | 16:38 |
| *** abhitechie has joined #openstack-swift | 16:39 | |
| *** SkyRocknRoll has joined #openstack-swift | 16:48 | |
| *** SkyRocknRoll_ has joined #openstack-swift | 16:49 | |
| *** tesseract has joined #openstack-swift | 16:50 | |
| *** abhitechie has quit IRC | 16:51 | |
| *** abhitechie has joined #openstack-swift | 16:51 | |
| *** abhitechie has quit IRC | 16:54 | |
| *** abhitechie has joined #openstack-swift | 16:54 | |
| *** chsc has joined #openstack-swift | 17:01 | |
| *** chsc has joined #openstack-swift | 17:01 | |
| *** abhitechie has quit IRC | 17:04 | |
| *** abhitechie has joined #openstack-swift | 17:05 | |
| *** silor has joined #openstack-swift | 17:11 | |
| *** tesseract has quit IRC | 17:22 | |
| *** abhitechie has quit IRC | 17:24 | |
| *** SkyRocknRoll_ has quit IRC | 17:30 | |
| *** mvk has joined #openstack-swift | 17:32 | |
| *** mingyu has quit IRC | 17:37 | |
| *** mingyu has joined #openstack-swift | 17:41 | |
| *** SkyRocknRoll_ has joined #openstack-swift | 17:42 | |
| thurloat | what process cleans up deleted objects if an object server were off when the object was deleted initially? | 17:42 |
| thurloat | is that the auditor's job? | 17:43 |
| notmyname | replicator | 17:44 |
| *** mingyu has quit IRC | 17:46 | |
| thurloat | so it comes across an object after scan that it knows it shoudln't have, looks it up and sees it should be toast, then cleans up? | 18:01 |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: split merge_items into merge_objects and merge_shard_ranges https://review.openstack.org/509857 | 18:01 |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: add post-sharding object PUT to probe test https://review.openstack.org/509858 | 18:04 |
| notmyname | thurloat: the tombstone files are replicated, and if the tombstone is newer than the object data, the object data is unlinked | 18:08 |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: remove support for +/- relative object_count in shard updates https://review.openstack.org/509860 | 18:10 |
| timburke | thurloat: when the DELETE request comes in, whichever object servers are available write down "tombstones" and unlink any data they had for the name. those tombstones hang around for some period of time (the reclaim_age; defaults to one week). from there, it's like notmyname said -- the tombstones get replicated and replace any data on nodes that missed the original DELETE | 18:11 |
| timburke | acoles: if you keep sending more patches for me to review, i'll never get around to submitting another address-some-review-comments patch! :P | 18:13 |
| acoles | timburke: sorry :( I *am* reviewing your doc patch right now | 18:13 |
| timburke | thanks. should be pretty quick | 18:14 |
| openstackgerrit | Merged openstack/swift feature/deep: Don't lose timestamp info when moving misplaced objects https://review.openstack.org/509589 | 18:14 |
| timburke | i just need to crank through and set a new high-water mark for myself | 18:14 |
| acoles | oh, that reminds me... mattoliverau can you lease look at patch 509860 - it's removing code that I couldn't see being used by I'd really value your blessing on that | 18:15 |
| patchbot | https://review.openstack.org/#/c/509860/ - swift (feature/deep) - remove support for +/- relative object_count in sh... | 18:15 |
| clayg | thurloat: FWIW SwiftStacks default object reclaim age is more like 1 mo - if you have a node "down" for a long time the the tombstones can get reclaimed while the node holding the data is down and re-introduce the dark data! | 18:15 |
| acoles | timburke: the doc build failed for me with some picky block quote error, I'll try to fix | 18:15 |
| timburke | bah | 18:15 |
| clayg | thurloat: If you have a node out longer than reclaim age you should have already taken it out of rings - and you need to reintroduce it to the cluster and fresh new capacity | 18:15 |
| clayg | *as fresh... | 18:16 |
| acoles | tox -e docs is so opinionated these days | 18:16 |
| acoles | should just be happy we have docs | 18:16 |
| openstackgerrit | Merged openstack/swift feature/deep: include deleted objects when cleaving https://review.openstack.org/509793 | 18:16 |
| clayg | lol | 18:16 |
| thurloat | clayg: thanks for all the info. looks like it might be in my best interest to bring reclaim_age down a bit | 18:17 |
| thurloat | at least temporarily | 18:17 |
| clayg | *down*?? do you have a lot of old tombstones? | 18:17 |
| clayg | like default is 1 wk, but IMHO 4 wk is a better default (*increate value by factor of 4*) - but maybe IME unless you're a dedicated Swift operations team you don't want to have to be super responsive to hardware outages and tend to not DELETE data once it's in Swift - YMMV | 18:19 |
| thurloat | yeah i have a bunch that I'd like cleaned up in the near term. i'll likely take your advice after the fact to increase the value once all is calm | 18:21 |
| clayg | oic, right on - good luck | 18:24 |
| thurloat | thanks, need it ;) | 18:24 |
| clayg | there was also an old swift bug that caused some hashdirs to get sort of ... stale and they'd just have a bunch of older than reclaim age tombstones in them - that was finally fixed tho - by mahatic acctually IIRC (hi mahatic! HOPE you're sleeping!) | 18:25 |
| thurloat | oh that's interesting | 18:25 |
| thurloat | ez keyword to search the git log for? | 18:25 |
| openstackgerrit | Alistair Coles proposed openstack/swift feature/deep: Add Container Sharder to source docs https://review.openstack.org/509830 | 18:26 |
| timburke | https://github.com/openstack/swift/commit/321bb91 i think? | 18:27 |
| timburke | maybe | 18:28 |
| acoles | timburke: go aheaad and hit +A on 509830 if you're happy with my tweaks - passed locally for me now | 18:28 |
| acoles | good night | 18:28 |
| timburke | good night acoles! | 18:29 |
| thurloat | nice timburke, thanks. definitely applies to me. | 18:29 |
| clayg | *YES* it was mahatic ! | 18:30 |
| clayg | thanks timburke | 18:30 |
| clayg | oh... wait no ... not that one I think... | 18:30 |
| clayg | boom! still mahatic tho! https://github.com/openstack/swift/commit/81d4673674febdbe0bba6e27a6d24739456eb3c4 | 18:31 |
| *** SkyRocknRoll_ has quit IRC | 18:33 | |
| *** janonymous has quit IRC | 18:34 | |
| *** ChubYann has joined #openstack-swift | 18:40 | |
| *** chlong_ has quit IRC | 18:44 | |
| *** chlong_ has joined #openstack-swift | 18:59 | |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-swiftclient master: Updated from global requirements https://review.openstack.org/89250 | 19:01 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/swift master: Updated from global requirements https://review.openstack.org/88736 | 19:02 |
| thurloat | oh great clayg thanks again! | 19:13 |
| openstackgerrit | Tim Burke proposed openstack/swift feature/deep: Address review comments https://review.openstack.org/509884 | 19:36 |
| *** bill_az has joined #openstack-swift | 19:37 | |
| *** chlong_ has quit IRC | 19:44 | |
| *** wes_dillingham has quit IRC | 20:12 | |
| *** stewie925 has quit IRC | 20:18 | |
| *** mat128 has quit IRC | 20:35 | |
| briancline | is there any worrisome/risky reason that the ring builder doesn't let you move devices from one zone to another? | 20:35 |
| *** catintheroof has quit IRC | 20:56 | |
| *** MVenesio has quit IRC | 20:57 | |
| clayg | briancline: not in my opinion | 21:18 |
| clayg | so... obviously doing that can have an effect on dispersion - and as a result some parts might want to move after you do it | 21:18 |
| clayg | and maybe a *lot* of parts depending on the topology and what exactly you're moving where | 21:19 |
| clayg | the pathological case is *every* part-replica on the device needs to go back to the region/zone/server it was on | 21:19 |
| clayg | but... w/e | 21:19 |
| clayg | rebalance is good at that | 21:19 |
| clayg | and there's a better than even chance most of the time it'll be *fine* | 21:19 |
| clayg | rezone it, rebalance, push rings, bob's your uncle | 21:20 |
| clayg | i've done it "by hand" - the data-structures don't give a %^& - there's just no native support from the cli | 21:20 |
| clayg | you should *totally* add it! | 21:20 |
| clayg | ;) | 21:20 |
| *** rcernin has joined #openstack-swift | 21:22 | |
| *** silor has quit IRC | 21:26 | |
| timburke | i feel like it could be particularly handy if you start out with a smallish cluster then decide to add a second region by physically moving existing servers to the new DC | 21:47 |
| clayg | ikr!? | 21:47 |
| clayg | timburke: no worries briancline will save us | 21:47 |
| *** bill_az has quit IRC | 22:00 | |
| *** rcernin has quit IRC | 22:06 | |
| mattoliverau | morning | 22:37 |
| *** wes_dillingham has joined #openstack-swift | 22:51 | |
| briancline | lol | 23:00 |
| briancline | yeah, I can give it a go. I've done it by hand recently for a one-off node and was fairly straightforward, and now need to do it for a few more | 23:01 |
| clayg | yeah if you could just re-imagine the whole swift-ring-builder cli interface that'd be *great* | 23:02 |
| briancline | oh sure! | 23:02 |
| briancline | I'll add that right after my tests ;) | 23:04 |
| clayg | ;) | 23:04 |
| mattoliverau | lol | 23:13 |
| briancline | would also like to add a patch so that the builder's internal version number gets included in the ring.gz as well | 23:20 |
| briancline | makes it a lot easier for a machine to explicitly match a ring back to a specific builder | 23:21 |
| clayg | ^ acoles torgomatic one of these days we're just going to distribute builders | 23:22 |
| clayg | I bet we could serialize them such that you could open then and have a builder (and all it's extra bloat) or just open it and have the replica2part2dev table and the devs list for minimal memory footprint in workers | 23:23 |
| clayg | the problem sure is *not* the on-disk space, and there's no arguing it'd be *convenient* to the extra info | 23:24 |
| *** abhitechie has joined #openstack-swift | 23:24 | |
| briancline | wondered about that. figured memory footprint when you've got beaucoup workers was maybe why | 23:25 |
| *** chsc has quit IRC | 23:29 | |
| clayg | yeah - but one integer isn't going to make a difference | 23:29 |
| clayg | the builders *do* have some largish datastructures that you only need for rebalance that can bloat up the runtime - but it's not really the metadata - we just recently had a conversation about "well... if it's nice to have *this* piece; what about *this* one? when do you stop? should we just distribute the whole builder?" | 23:31 |
| *** kei_yama has joined #openstack-swift | 23:34 | |
| mattoliverau | clayg: +1. it would make life easier and loosing a builder file an obsolete problem.. just need to make sure we can only load what we need as a ring and we should be good :) | 23:44 |
| briancline | oh yeah, def not worried about the version number. probably no more than a 1MB difference in total across all my clusters | 23:44 |
| mattoliverau | i remember discussing it was it PTG or Boston.. can't remember, but good idea | 23:45 |
| briancline | i've never timed it - would loading something like a 2^20 builder take a lot longer than the ring? | 23:46 |
| briancline | like to a point where object workers (in python) would show a brief but noticeable bit of latency, i mean | 23:47 |
| *** vint_bra has quit IRC | 23:55 | |
| clayg | neat! | 23:56 |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!