Chat Log - Sessions 17-29¶
*** Vjarjadian has quit IRC 00:07
*** saabylaptop has joined #ceph-summit 00:07
*** saabylaptop has left #ceph-summit 00:08
*** dalgaaf has joined #ceph-summit 00:44
*** Vincent_Valentine has joined #ceph-summit 00:46
*** dalgaaf has quit IRC 01:08
*** dalgaaf-tmp has joined #ceph-summit 01:08
*** jaydee has joined #ceph-summit 01:38
*** silversurfer has quit IRC 01:43
*** Vincent_Valentine has quit IRC 02:19
*** nwat has quit IRC 02:24
*** silversurfer has joined #ceph-summit 02:33
*** jaydee has quit IRC 02:37
*** mozg has quit IRC 02:41
*** nwat has joined #ceph-summit 02:46
*** dalgaaf has joined #ceph-summit 02:55
*** dalgaaf-tmp has quit IRC 02:57
*** nwat has quit IRC 02:58
*** fireD has joined #ceph-summit 03:08
* loicd listening to playback of tonight feed 03:30
loicd ccourtaut: I'm breaking up sometimes ( not hearing what people say ) around 0:38 inhttp://www.youtube.com/watch?v=j0JQvE5uGgs&feature=youtu.be . Do you experience the same problem ? If you listen from 0:38 to 0:39 without breaking up you don't have the same problem and it's just on my side ;-) 03:47
ccourtaut loicd, oh you mean you don't hear sage but his lips are moving? 03:50
ccourtaut he told yesterday he was sorry about that, we he types, it mutes, and it happens that he types while talking 03:51
ccourtaut loicd, iirc 03:51
joelio yea 03:51
loicd ccourtaut: yes 03:51
loicd oh, ok 03:51
loicd ccourtaut: thanks 03:52
loicd it's better now 03:52
*** nhm has quit IRC 04:05
*** dalgaaf has quit IRC 04:33
*** dalgaaf has joined #ceph-summit 04:35
loicd ccourtaut: I prepared this http://pad.ceph.com/p/osd-erasure-coding 04:39
loicd does it make sense to you ? 04:40
* ccourtaut looking 05:00
loicd I've added a few lines at the beginning of http://wiki.ceph.com/01Planning/02Blueprints/Emperor/Create_and_Maintain_S3_feature_list_for_compatibility_tracking feel free to edit if you don't like how it's done 05:03
loicd leseb: \o 05:03
leseb loicd: o/ 05:03
loicd leseb: did you attend yestereday ? 05:06
leseb loicd: barely and you? it was too late 05:07
leseb loicd: and you? 05:07
loicd same :-) 05:07
leseb just read some BPs and pads today 05:07
loicd which ones ? 05:07
* loicd starting with http://pad.ceph.com/p/continuous-integration 05:08
leseb loicd: rbd: shared read cache and rbd: namespace support 05:09
leseb loicd: bbl 05:11
*** zhu has quit IRC 05:58
*** zhu has joined #ceph-summit 05:58
*** Vincent_Valentine has joined #ceph-summit 06:04
*** zhu_ has joined #ceph-summit 06:05
*** zhu has quit IRC 06:07
*** yanzheng has joined #ceph-summit 06:09
loicd ccourtaut: http://pad.ceph.com/p/continuous-integration if you have a few minutes, it would be great if you could review / patch this :-) 06:11
loicd yehudasa: http://pad.ceph.com/p/continuous-integration I suggest that teuthology is made more friendly for non-inktank users as part of a better continuous integration 06:13
*** nwat has joined #ceph-summit 06:14
joelio should the tenants be driven by a standard API (EC2, OCCI etc.?) 06:15
loicd joelio: good idea :-) using euca2ools would make teuthology somewhat agnostic 06:21
joelio :) 06:22
joelio I use OpenNebula here, so anything that is cross-platform is win for me 06:22
ccourtaut loicd, looks good to me 06:24
loicd :-) 06:25
* loicd takes a break before the summit starts 06:25
joelio loicd: added that to pad btw 06:25
*** scuttlemonkey has joined #ceph-summit 06:31
*** yanzheng has quit IRC 06:31
sagewk good morning! 06:46
joelio sagewk: morning 06:49
joelio (afternoon!) 06:49
*** rturk-away is now known as rturk 06:51
sagewk good afternoon, EMEA! :) 06:52
rturk Hello all! 06:52
*** erice_ has joined #ceph-summit 06:52
*** mikedawson has joined #ceph-summit 06:52
*** rturk changes topic to "summit schedule: http://ceph.com/cds || live video stream: http://youtu.be/-K8bSHx7zJ0 || ping rturk to join the hangout" 06:54
rturk ** Session "Welcome, Introduction, and Summit Process" starting in 5 minutes! || pad: http://bit.ly/1b936Xb 06:55
*** topro has joined #ceph-summit 06:56
*** wido has joined #ceph-summit 06:57
rturk ** Session "Welcome, Introduction, and Summit Process" starting now! || pad: http://bit.ly/1b936Xb 07:00
rturk live video stream: http://youtu.be/-K8bSHx7zJ0 07:00
*** davidzlap has joined #ceph-summit 07:02
*** ircolle has joined #ceph-summit 07:03
loicd joelio: cool :-) 07:05
loicd ccourtaut: would you agree to take notes during the erasure coding session ? 07:06
*** sjustlaptop has joined #ceph-summit 07:08
*** joao has joined #ceph-summit 07:09
loicd http://pad.ceph.com/p/osd-erasure-coding sjust would you agree to walkthru your erasure code design document ? I outlined the agenda at http://pad.ceph.com/p/osd-erasure-coding 07:11
sjustlaptop sure 07:13
ccourtaut loicd, sure 07:13
loicd ccourtaut: sjustlaptop cool ;-) 07:16
*** alfredodeza has joined #ceph-summit 07:17
*** ircolle1 has joined #ceph-summit 07:18
*** Vincent_Valentine has quit IRC 07:22
*** Vincent_Valentine has joined #ceph-summit 07:22
*** jyluke has joined #ceph-summit 07:22
*** ircolle has quit IRC 07:24
rturk ** Session "osd: Erasure coded storage backend" starting in 5 minutes! || blueprint: http://bit.ly/1epNHS5 || pad: http://bit.ly/1b936Xb 07:25
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 07:25
rturk I know sjustlaptop and loicd were listed as interested parties for the erasure coded storage backend session 07:26
rturk anyone else? 07:26
scuttlemonkey ccourtaut was going to take notes, so he should probably get an invite 07:27
rturk ok 07:27
scuttlemonkey so he's live and not delayed 07:27
rturk ok - ccourtaut, invite sent via pm 07:28
rturk ** Session "osd: Erasure coded storage backend" starting now! || blueprint: http://bit.ly/1epNHS5 || pad: http://bit.ly/1b936Xb 07:31
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 07:31
rturk hey just kidding about that pad link. It's actually here: http://pad.ceph.com/p/osd-erasure-coding 07:34
mikedawson loicd: My team will be able to be an alpha tester 07:40
loicd mikedawson: excellent news :-) 07:40
sagewk awesome! 07:40
joelio loicd: me too, inherited a few more test nodes today 07:42
scuttlemonkey joelio: hey, did you ever do that internal technote? 07:42
loicd coool added to http://pad.ceph.com/p/osd-erasure-coding 07:42
joelio scuttlemonkey: oh, much more than that - built the full platform :) 07:43
joelio I guess the asnwer is yes, I can do a community post :) 07:43
scuttlemonkey haha 07:43
scuttlemonkey only if you have time/interest 07:43
rturk :) 07:43
scuttlemonkey I just remember you saying the technote would be an easy drop-in post 07:43
scuttlemonkey but anything works :) 07:44
joelio scuttlemonkey: sure have, talking about it at OpenNebulaConf in September, along with out use cases 07:44
scuttlemonkey yeah, that's what made me think of it actually 07:44
scuttlemonkey Jaime mentioned you were speaking 07:44
joelio s/out/our 07:44
joelio yea, wanted to drive some actual usage by users.. otherwise we'd be talking about a fairly sparse system. They are coming thick and fast now, users love it 07:45
scuttlemonkey also didn't realize you were doing OpenNebula until recently 07:45
scuttlemonkey I'll need to stand that up and become more familiar with it now that there are real live people using ceph+opennebula together 07:46
joelio yea, I had to wait for about a month to get approval.. BBC beaurocracy.. <rolls eyes? 07:46
scuttlemonkey haha 07:46
Vincent_Valentine why are we writing multiple OSDs for a single write? 07:46
scuttlemonkey big orgs same the world over 07:46
joelio yup" 07:46
jyluke I will be interested in opennebula+ceph as well, will be starting to do within the a few weeks time 07:46
Vincent_Valentine If we write single OSD, the problem of logs factored over multiple OSDs will not be there. 07:46
joshd Vincent_Valentine: replication is synchronous, writes go to the primary osd, which forwards them to other osds in the acting set and waits for all of them to apply the write before telling the client it's complete 07:47
Vincent_Valentine Who is going to do erasure coding? Client or Another OSD? 07:49
Vincent_Valentine If its another OSD, I can think of just replicating logs over 2/3 OSDs 07:52
rturk ** Session "osd: cache pool overlay" starting in 5 minutes! || blueprint: http://bit.ly/154Op1e || pad: http://bit.ly/1cGbUW4 07:55
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 07:55
Vincent_Valentine but it would simplify recovery ... am i wrong ? 07:55
rturk sjust: ^ 07:56
sjustlaptop I'm not sure, can you expand on that? 07:56
*** jyluke has quit IRC 07:58
Vincent_Valentine Im thinking on rolling forward for erasure coded tier. If logs are replicated, instead of fragmented by coding, recovery would be similar to replication PG 07:58
scuttlemonkey sjustlaptop: ^^ 07:59
Vincent_Valentine 1 more question ... How efficient would erasure coding would be for small object? And how does this translate to erasure coded RDB? 07:59
sjustlaptop oh, logs will be replicated 08:00
sjustlaptop rather than coded 08:00
sjustlaptop the trouble is rolling forward the actual objects 08:00
sjustlaptop the pg logs do not contain the full append data, for example 08:00
sagewk Vincent_Valentine: can only roll forward if you have the M/N shards 08:00
sjustlaptop (though perhaps they could) 08:00
sagewk for replication, any 1 replica is enough 08:00
Vincent_Valentine Understand that .. trying to simplify log consistency problem by detaching logs from erasure coding. 08:02
*** rongze_ has quit IRC 08:02
Vincent_Valentine [Repost] 1 more question ... How efficient would erasure coding would be for small object? And how does this translate to erasure coded RDB? 08:02
*** fooz has joined #ceph-summit 08:03
sagewk Vincent_Valentine: see bottom of pad 08:04
Vincent_Valentine sagewk .. what abt erasure coded RDB 08:05
*** ksp has joined #ceph-summit 08:05
Vincent_Valentine so would there be tiers .. ? Is it planned in this release? 08:06
sjustlaptop stay tuned! 08:06
rturk ** Session "osd: cache pool overlay" starting now! || blueprint: http://bit.ly/154Op1e || pad: http://bit.ly/1cGbUW4 08:06
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 08:06
Vincent_Valentine :) .. i think its not same as caching ... 08:07
rturk Vincent_Valentine: would you like to join the hangout? 08:07
*** dragonfly has quit IRC 08:07
rturk might be easier than dealing with the live video feed delay 08:07
sjustlaptop there are discusions for both tiering and caching 08:07
Vincent_Valentine Heh ... m fine ... would like to participate in this though .. whom to contact ? 08:08
scuttlemonkey loicd is the one driving erasure coding 08:08
scuttlemonkey but for anything I would add yourself to the blueprint and start chatting with folks in irc and on the lists 08:09
*** davidzlap has quit IRC 08:09
Vincent_Valentine Sure .. will do.. thanks 08:10
*** mikedawson_ has joined #ceph-summit 08:12
rturk ** Session "osd: object redirects" starting soon! || blueprint: http://bit.ly/1b9dy0H || pad: http://bit.ly/12XE2Cf 08:13
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 08:13
Vincent_Valentine Is cache pool replicated? 08:13
*** davidzlap has joined #ceph-summit 08:13
*** ksp has quit IRC 08:14
Vincent_Valentine isnt that expensive for NAND ? 08:14
sjustlaptop Vincent_Valentine: well, you could set replication to 1 (no replication) 08:15
sjustlaptop but that would have an effect on durability 08:15
Vincent_Valentine But then I loose all durability requirement ... cause we are modifying in cache pool 08:15
sjustlaptop yes 08:16
sjustlaptop could do write-through, I suppose 08:16
Vincent_Valentine I was assuming write-through and no replication 08:16
sjustlaptop would only accelerate reads 08:16
sjustlaptop might be worthwhile though 08:16
Vincent_Valentine To simplify this .. I would suggest two types of pools ... replicated for write-backs and unreplicated for unmodified data ...Question is who is managing write-back flushing? What policies? Who maintain write ordering? 08:18
sjustlaptop Vincent_Valentine: as I understand it, the cache pool primary manages write ordering 08:19
*** mikedawson has quit IRC 08:19
joshd there are no ordering guarantees with operations on different objects 08:19
Vincent_Valentine aah ... then how does this work on RBD? 4MB objects with no ordering garuntee? 08:20
sjustlaptop Vincent_Valentine: most disks don't give much in the way of ordering guarrantees either 08:20
joshd all you can tell with most disks is 1 sector will be atomic (with rbd it's anything to a single object, and object boundaries will align with sectors in the virtual disk) 08:21
Vincent_Valentine It was regarding eviction ordering for write-back cache. (especially RBD) 08:22
rturk ** Session "osd: object redirects" starting now! || blueprint: http://bit.ly/1b9dy0H || pad: http://bit.ly/12XE2Cf 08:29
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 08:29
Vincent_Valentine Just to be sure, we will promote whole object on caching pool? .... There would be case for doing small block(4KB) caching than object caching ... 08:29
*** sjustlaptop has quit IRC 08:34
scuttlemonkey sagewk: just fyi, about 10m until objecter is slated, although we can probably eat into the break a bit 08:36
rturk ** Session "librados/objecter: smarter localized reads" starting soon! || blueprint: http://bit.ly/12XHlJI || pad: http://bit.ly/1b9h4bi 08:40
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 08:40
davidzlap Need new code for a "delete foo, create foo" seeing with DELETING state? 08:42
*** sjustlaptop has joined #ceph-summit 08:46
rturk ** Session "librados/objecter: smarter localized reads" starting now! || blueprint: http://bit.ly/12XHlJI || pad: http://bit.ly/1b9h4bi 08:49
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 08:49
*** jyluke has joined #ceph-summit 08:53
loicd what parts of ceph use Objecter ? 08:56
sjustlaptop loicd: clients 08:57
sjustlaptop it's the ball of logic responsible for determining where to send client io 08:57
loicd ohhhh 08:57
joshd mainly librados, ceph-fuse, and the mds use it directly - other clients use librados 08:57
loicd :-D 08:57
loicd now I get it ! 08:57
* loicd relieved 08:58
*** nwat has quit IRC 08:58
sagewk :) 09:00
sagewk the goofy Objecter name strikes again 09:00
*** Vincent_Valentine has quit IRC 09:02
*** Vincent_Valentine has joined #ceph-summit 09:02
*** sjustlaptop has quit IRC 09:07
*** rturk changes topic to "summit schedule: http://ceph.com/cds || live video stream: http://youtu.be/cGosx5zD4FM || ping rturk to join the hangout" 09:09
*** mikedawson_ has quit IRC 09:10
rturk ** Session "rgw: Multi-region / Disaster Recovery" starting in 5 minutes || blueprint: http://bit.ly/1eq1J5X || pad: http://bit.ly/12XJWmU 09:10
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 09:10
*** hskinner has joined #ceph-summit 09:11
*** fireD has left #ceph-summit 09:11
loicd I'd like to join in ~1h :-) 09:12
loicd rturk: ^ 09:12
rturk We are about to relaunch our live stream: http://youtu.be/cGosx5zD4FM 09:12
rturk loic: ok 09:12
*** mikedawson has joined #ceph-summit 09:15
sagewk ccourtaut: are you joining as well? 09:16
rturk sagewk: he's already in the hangout 09:16
ccourtaut sagewk, already in 09:16
ccourtaut :) 09:16
rturk ** Session "rgw: Multi-region / Disaster Recovery" starting now! || blueprint: http://bit.ly/1eq1J5X || pad: http://bit.ly/12XJWmU 09:17
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 09:17
*** mikedawson_ has joined #ceph-summit 09:20
*** [1]jyluke has joined #ceph-summit 09:21
*** jyluke has quit IRC 09:21
*** [1]jyluke is now known as jyluke 09:21
*** mikedawson has quit IRC 09:26
*** mikedawson_ is now known as mikedawson 09:26
sagewk we lost yehudasa! 09:29
joelio heading home, will be about for the rbd: cloud management platform features 09:35
joelio rturk: could I register my interest in hangouts now please pal? 09:36
rturk Yep, I'll write you down for that session 09:36
joelio ace, thanks 09:37
*** erice has quit IRC 09:41
rturk ** Session "rgw: bucket level quota" scheduled to start in 5 minutes || blueprint: http://bit.ly/16s1rsC || pad: http://bit.ly/11K0Vq2 09:42
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 09:42
rturk ** Session "rgw: bucket level quota" starting now! || blueprint: blueprint: http://bit.ly/16s1rsC || pad: http://bit.ly/11K0Vq2 09:47
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 09:47
*** erice_ has quit IRC 09:54
*** VincentValentine has joined #ceph-summit 09:59
wido so I missed the question :) 10:01
wido sagewk: ^^ 10:02
*** Vincent_Valentine has quit IRC 10:02
*** nwl has joined #ceph-summit 10:04
*** VincentValentine has quit IRC 10:04
*** Vincent_Valentine has joined #ceph-summit 10:04
scuttlemonkey hey wido 10:05
wido hey 10:05
nwl hi 10:05
scuttlemonkey trying to figure out the use case for bucket-level quotas 10:05
* wido was already listening in 10:05
wido just missed the questions about the quota enforcement 10:05
wido there is some serious lag in the video I think :) 10:06
scuttlemonkey yes, the youtube link is maybe 2-3 mins behind the hangout 10:07
nwl so it can be censored for bad language 10:07
scuttlemonkey or wardrobe malfunctions :P 10:07
wido Anyway, I don't think quota restriction has to be very tight 10:08
wido a near perfect quota is fine for 99% of the situations 10:08
wido maybe even 99.99% 10:08
*** lyncos has joined #ceph-summit 10:08
rturk ** Session "rgw: Create and Maintain S3 feature list" scheduled to start in 5 minutes || blueprint: http://bit.ly/16s4bWV || pad: http://bit.ly/11K2Gnl 10:10
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 10:10
rturk ** Session "rgw: Create and Maintain S3 feature list" starting now! || blueprint: http://bit.ly/16s4bWV || pad: http://bit.ly/11K2Gnl 10:13
loicd https://github.com/kri5/ceph/blob/wip-s3-compliance-doc/doc/radosgw/s3_compliance.rst 10:17
loicd that's what christophe is describing currently 10:17
yehudasa http://aws.amazon.com/releasenotes/Amazon-S3 10:25
yehudasa copy pasted into my hangout window and the window is stuck now 10:25
rturk ** Session “Continuous Integration” scheduled to start in 5 minutes || pad: http://bit.ly/13Kk34N 10:26
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 10:26
*** cjh_ has left #ceph-summit 10:27
rturk I wonder whether this list should live on the wiki? 10:27
rturk if it's important that the barrier to editing is low, anyway 10:27
*** Andi has joined #ceph-summit 10:28
*** Andi is now known as Guest2516 10:28
*** Guest2516 has left #ceph-summit 10:29
*** abluemle has joined #ceph-summit 10:29
*** abluemle has left #ceph-summit 10:32
rturk ** Session "Continuous Integration" starting now! || pad: http://bit.ly/13Kk34N 10:33
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 10:33
*** buck has joined #ceph-summit 10:37
sagewk question for the room: who loves/hates gerrit? 10:38
scuttlemonkey sagewk: having used it in the past, I really liked it 10:40
scuttlemonkey was very approachable and (if done right) quite scalable 10:40
joshd I strongly dislike gerrit - it only handles one commit at a time (encouraging unreviewable and too-large patches) 10:41
ircolle1 joshd - you can create a tree of patches 10:41
paravoid wikimedia uses gerrit for pretty much everything 10:42
joshd yes, but they're not reviewed or committed atomically 10:42
ircolle1 joshd - right 10:42
paravoid I can provide some insight if you want 10:42
ircolle1 paravoid - want to join the hangout? 10:42
scuttlemonkey paravoid: that would be great 10:42
*** sjustlaptop has joined #ceph-summit 10:42
joshd and you're building and testing and reviewing every singe commit instead of a branch (i.e. a pull request) 10:42
rturk sent link to you, paravoid 10:42
rturk although the hangout is nearly full 10:43
paravoid yeah, full 10:43
rturk and now it's totally full - anyone wanna drop back to the video stream to let paravoid in? 10:44
ircolle1 I'll drop 10:44
nwl "Pull out! You're not doing any good back there!" 10:45
* ircolle1 stuffs nwl back in his cage 10:46
nwl "Good shot, Janson!" 10:47
nwl i am just quoting random shit now 10:47
paravoid http://www.mediawiki.org/wiki/Git/Gerrit_evaluation 10:47
*** nhm has joined #ceph-summit 10:48
*** abluemle has joined #ceph-summit 10:49
rturk I seem to hear "we use garrit but have put some work into it" a lot 10:49
*** lyncos has quit IRC 10:50
*** ksp has joined #ceph-summit 10:53
paravoid wikimedia's CI central stuff is https://integration.wikimedia.org/ 10:54
paravoid central page even 10:54
paravoid all in all, I don't think you can easily use both gerrit & github at the same time 10:55
paravoid and to be frank, I wouldn't switch to gerrit at this point if I were you 10:55
scuttlemonkey ahh 10:56
zackc i'm on IRC at least :) 10:56
paravoid okay, dropping from the hangout to open up a slot 10:56
scuttlemonkey paravoid: thanks for the overview though...that insight is invaluable 10:57
*** sjustlaptop has quit IRC 10:59
*** sjustlaptop has joined #ceph-summit 10:59
*** nhm has quit IRC 11:02
joelio rturk: will just be on IRC later, got people visiting at home - can just about get away with typing.. full conference might be a little rude :) 11:05
rturk ok :) 11:06
joelio hopefully be able to add some ONE insight anyway 11:06
rturk lmk if that changes 11:06
joelio np 11:06
*** alram has joined #ceph-summit 11:07
wido scuttlemonkey: I'm here for the librgw session 11:11
*** nwat has joined #ceph-summit 11:11
*** rturk changes topic to "summit schedule: http://ceph.com/cds || live video stream: http://youtu.be/hrQuYzyXvVg || ping rturk to join the hangout" 11:11
wido rturk: ^^ 11:11
rturk wido, sending hangout link via pm 11:12
*** dpippenger has joined #ceph-summit 11:12
rturk ** Session “librgw” starting now! || blueprint: http://bit.ly/15GefyV || pad: http://bit.ly/13Knzfy 11:16
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 11:16
rturk If you're looking for the live video stream, it's here: http://youtu.be/hrQuYzyXvVg 11:21
*** nwat has quit IRC 11:22
*** sjustlaptop has quit IRC 11:23
*** troug has joined #ceph-summit 11:27
*** sjustlaptop has joined #ceph-summit 11:27
*** nhm has joined #ceph-summit 11:28
*** hskinner has quit IRC 11:29
*** nhm has quit IRC 11:29
*** sjustlaptop has quit IRC 11:29
*** sjustlaptop has joined #ceph-summit 11:30
joelio rturk: yea, I should be in after all.. can escape for 1/2 hour 11:30
rturk joelio: ok - link sent via pm 11:31
*** nhm has joined #ceph-summit 11:31
scuttlemonkey joelio: fwiw, that link will work until the end of the summit 11:31
scuttlemonkey so feel free to drop in/out as you need ot 11:31
scuttlemonkey to* 11:31
joelio cheers guys 11:32
ksp I'm new to the chat. Which questions should go to IRC and which ones to the PAD ? 11:38
ircolle1 ksp - either is fine 11:38
scuttlemonkey ksp: feel free to drop them either place 11:38
joelio you'll have to excuse my crappy webcam mic btw, not got my headset with me today 11:40
scuttlemonkey no worries 11:40
rturk ** Session “msgr: infiniband support via rsockets” scheduled to start in 5 minutes || blueprint: http://bit.ly/11Ke7eH || pad: http://bit.ly/191fQyk 11:40
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 11:40
scuttlemonkey I keep putting myself on mute so I don't pick up ambient noise 11:40
joelio yep, right next to a road here too so will have to do that 11:41
*** dstroppa has joined #ceph-summit 11:42
rturk ** Session “msgr: infiniband support via rsockets” starting now! || blueprint: http://bit.ly/11Ke7eH || pad: http://bit.ly/191fQyk 11:46
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 11:46
ircolle1 I dropped to make room 11:46
ccourtaut same for me 11:46
sagewk abluemle: care to join? 11:47
abluemle andreas is here 11:47
ksp Andreas and I are connected 11:47
abluemle but only via chat 11:47
scuttlemonkey no mic? Would love to have you on the broadcast, instead of just the youtube link && irc 11:48
dalgaaf hi dieter and andreas 11:48
ksp Hi danny 11:49
abluemle sorry:Ö no mic, no webcam 11:49
dmick that makes it more difficult :) 11:49
*** sjustlaptop has quit IRC 11:49
*** sjustlaptop has joined #ceph-summit 11:50
abluemle Agree on the internal implementation via r*() 11:50
abluemle IP is only relevant for connection establishment 11:51
sagewk but do you need to know whether the peer is rsockets or does it fallb ack? 11:51
sagewk i.e. can you use r* calls unconditionally and let magic happen? 11:51
abluemle It does not fall back afaik 11:51
ksp libSDP does use the fallback logic, and this fallback behavior can be added to rsockets as well 11:52
abluemle libsdp is different concerning fall back: it does fall back to regular TCP sockets 11:52
abluemle The IP address does not tell you about the rsocket character of the connection 11:54
abluemle sth. like cluster network on rsockets and public network on regular sockets 11:56
ksp The use case would be to run the cluster network with IB + rsockets and the client network still with Eth + IP-socket 11:56
rturk ** Session “rbd: cloud management platform features” scheduled to start soon || blueprint: http://bit.ly/11Kfxpk || pad: http://bit.ly/191hsbd 11:57
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 11:57
abluemle no kernel so far 11:57
ksp we are considering to port rsocket to the kernel, but not decided yet 11:58
ksp Sean Hefty 11:58
ksp Sean Hefty from Intel own rsockets 11:58
sagewk libdsp is EOL right? 11:58
dstroppa rturk: can I have the link for the hangout for the cloud mgt platform? 11:58
ksp yes, SDP is EOL 11:58
rturk dstroppa: sent via pm, thanks :) 11:59
abluemle For SDP there were userland and kernel parts 11:59
abluemle option support is avaiable in rsocket 12:00
ksp SDP has some license / legal risks ... this might be the reason for EOL 12:00
sagewk what is option support? 12:00
ksp sorry, didn't get the question 12:01
abluemle setsockopt() 12:01
dmick Did Matthew Anderson ever get anywhere, I wonder? http://lists.ceph.com/pipermail/ceph-users-ceph.com/2013-June/002352.html 12:01
dalgaaf Dieter: it's about "option support is avaiable in rsocket" 12:02
abluemle blocking vs. non-blocking: I am not sure about the current support in rsockets. 12:02
rturk ** Session “rbd: cloud management platform features” starting now! || blueprint: http://bit.ly/11Kfxpk || pad: http://bit.ly/191hsbd 12:04
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 12:04
*** dosaboy_ has joined #ceph-summit 12:05
sagewk abluemle: ksp: thanks! 12:06
ksp you are welcome 12:06
loicd leseb: could you please link the blueprint ? 12:13
leseb loicd: which one? 12:14
loicd the one you're pushing 12:15
leseb loicd: https://review.openstack.org/#/c/36042/ 12:15
loicd leseb: you can ignore me for the next 30 minutes and answer later ;-) 12:15
sagewk zackc: what to look at http://pad.ceph.com/p/cds-teuthology ? 12:16
sagewk add any notes for things we should discuss? 12:16
*** sjustlaptop has quit IRC 12:16
*** sjustlaptop has joined #ceph-summit 12:18
*** abluemle has left #ceph-summit 12:19
*** ksp has left #ceph-summit 12:20
zackc sagewk: looking now 12:22
rturk ** Session “Teuthology” scheduled to start soon || pad: http://bit.ly/191jGYa 12:25
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 12:25
zackc rturk: i'd like to join 12:26
joelio I'm out 12:26
scuttlemonkey joelio: thanks for joining :) 12:26
joelio n/p :) 12:26
rturk zackc: link sent via pm 12:26
jyluke just wondering, how's ceph with opennebula? would there be a lot of effort to get them work together? 12:26
rturk thanks, joelio! 12:26
rturk jyluke: someone actually just presented about that at Ceph Day 12:26
scuttlemonkey Jaime has done a pretty decent job of blogging and documenting OpenNebula && Ceph 12:27
scuttlemonkey http://opennebula.org/documentation:archives:rel4.0:ceph_ds 12:27
joelio yea, I'm liking it 12:27
scuttlemonkey http://blog.opennebula.org/?p=4441 12:28
joelio it sounds like the other management stacks have similar challenges, like lack of abstractions to handle types other than file, dev etc.. 12:28
Tamil rturk: pull me in too for teuthology 12:28
jyluke ah.. thanks. will try it out after our holidays here in malaysia. any catchas that i should be aware of? 12:28
joelio jyluke: If you need any help getting it going, ping me on #opennebula 12:29
rturk Tamil: Okay - we're running at capacity now but you should be there 12:29
jyluke thanks joelio 12:29
Tamil rturk: oh ok thanks 12:30
joelio not really, but there are subtle differences depending on your implementation. Most of the gotchas have been cought, the difficult ones to capture are distro and version nuances 12:30
joelio (in terms of gotas I mean!) 12:30
scuttlemonkey Tamil: when we start I'm guessing a couple of these guys will switch to consume rather than be on the broadcast 12:30
*** houkouonchi_work has joined #ceph-summit 12:31
*** fooz has left #ceph-summit 12:31
joelio there's definitely room for improvement in the model, currently just supports one pool. I've not tested multiple datastores of ceph type yet though, but I seem to recall there being some niggles 12:32
jyluke joelio: thanks for the advice, we will try some of the ideas out and let you know, our team is definately interested in different pools, sort of having a premium pool with premium disks for premium users and so on 12:35
loicd leseb: the discussion i refer to http://www.mail-archive.com/ceph-devel@vger.kernel.org/msg15479.html 12:35
loicd regarding LFS & rgw 12:35
leseb loicd: thanks :) 12:35
wido https://wiki.ubuntu.com/ServerTeam/CloudArchive 12:36
ircolle1 wido - how's your Japanese? Ceph on Cloudstack getting a lot of buzz in Japan lately. 12:36
mikedawson wido: I am using qemu 1.52 and libvirt 1.0.6 12:37
rturk ** Session “Teuthology” starting now! || pad: http://bit.ly/191jGYa 12:37
rturk ** Please tell rturk or scuttlemonkey if you want to join the hangout 12:37
*** dpippenger has quit IRC 12:37
wido ircolle1: pretty bad, as in: none 12:38
wido mikedawson: Ah, cool. From the Ubuntu archive? 12:38
joshd joelio: jyluke: since grizzly you can have multiple volume backends and different volume types, which could use different rbd pools 12:39
mikedawson wido: no, we're on Raring, so we installed via ppa:jacob/virtualisation 12:39
sjustlaptop I can drop if we need another slot 12:39
wido mikedawson: Ah, ok. I try to stick with Precise 12:39
wido backport the raring kernel and use the Cloud Archive repo 12:39
jyluke joshd: can i assume to use havana now? 12:40
jyluke joshd: i am on stock precise 12:40
*** sjustlaptop has quit IRC 12:41
joshd jyluke: yes, it'll work in havana too. basically each backend has its own section in cinder.conf - leseb has a nice blog about it http://www.sebastien-han.fr/blog/2013/04/25/ceph-and-cinder-multi-backend/ 12:41
*** sjustlaptop has joined #ceph-summit 12:42
*** dpippenger has joined #ceph-summit 12:42
jyluke joshd: thanks. 12:46
nhm I came into the teuthology conversation late... SLURM! 12:46
* nhm runs 12:46
scuttlemonkey nhm: http://goo.gl/l8n5EW 12:48
yehudasa sagewk: you'll still want to have priorities, even as a single user, e.g., if you have tests running continuously and you want to get a single test running immediately 12:48
nhm scuttlemonkey: close! https://computing.llnl.gov/linux/slurm/ 12:48
scuttlemonkey hah 12:49
scuttlemonkey I like mine better :P 12:49
nhm scuttlemonkey: indeed 12:49
nhm Whatever we do with teuthology, the simpler the better. I spend too much time right now trying to debug why something is broken. :( 12:51
alfredodeza I would really like it to be a web service 12:52
alfredodeza with an optional CLI that can talk to the HTTP API 12:52
alfredodeza best of both worlds 12:53
alfredodeza but, no ssh-keys, no '~/.teuthology' etc... 12:53
nhm alfredodeza: So long as it's not SOAP I might be willing to get on board. ;) 12:53
alfredodeza SOAP! 12:53
* alfredodeza runs 12:53
nhm alfredodeza: I actually really like pdsh for this kind of thing 12:53
nhm alfredodeza: I had a really really bad experience with folks that were very into semantic web interfaces and soured on the whole thing. 12:54
alfredodeza better visibility, history of previous runs, see what machines are available/unavailable, etc... 12:54
*** leseb has quit IRC 12:54
*** leseb has joined #ceph-summit 12:55
alfredodeza nhm: I see it as a couple of things: one is an API where you can interact with it (for tools like a CLI) and as a UI to interact with it through the web 12:55
alfredodeza nhm: we should chat :) 12:56
nhm alfredodeza: definitely has some advantages, especially if it can be done right. 12:56
alfredodeza getting it right sounds to me like an iterative process :D 12:57
nhm alfredodeza: That's usually the better way to go! 12:57
*** sjustlaptop has quit IRC 12:57
alfredodeza e.g. "here is v0.0.1, try it out and give me some feedback to get 0.0.2 out immediately" 12:57
dmick date quickly, marry slow :) 12:57
alfredodeza lol 12:58
*** jyluke has quit IRC 12:58
*** sjustlaptop has joined #ceph-summit 12:58
nhm alfredodeza: here's what I use for a lot of my testing now. It's not really designed great (I tend to way overuse singletons), but it's easy and mostly doesn't break. :) https://computing.llnl.gov/linux/slurm/ 12:58
nhm doh, wrong url 12:58
nhm https://github.com/ceph/ceph-tools/tree/master/cbt 12:58
nhm alfredodeza: heart of it is the pdsh wrapper in common.py 12:59
* alfredodeza looks 12:59
loicd :-) 12:59
alfredodeza very nice 13:00
nhm alfredodeza: was very much geared for low external dependencies and easy cluster creation. With ceph-deploy a lot of this is probably obsolete or becoming obsolete. 13:00
alfredodeza sure 13:00
alfredodeza but the fact that this exists asserts we could totally leverage an HTTP service 13:00
*** dstroppa has left #ceph-summit 13:00
nhm alfredodeza: so part of my job at some point will be to move some of this into teuthology in a sane way. 13:01
rturk Great summit, everyone!!! 13:02
ircolle1 some point…soon 13:02
loicd thanks everyone, great summit :-) 13:03
rturk Let 13:03
loicd 's 13:03
rturk Let's do this again in three months, k? 13:03
nhm ircolle1: :) 13:03
loicd :-D 13:03
loicd +1 13:03
scuttlemonkey second verse, same as the first! 13:03
rturk did the two-day format work? 13:03
loicd yes 13:03
scuttlemonkey rturk: might be useful to send out a survey link 13:03
rturk scuttlemonkey: yeah 13:03
scuttlemonkey complete with cheese at the end 13:04
rturk my thoughts exactly 13:04
dmick \o/ 13:04
rturk survey form -> t-shirt 13:04
scuttlemonkey ok, time to grab some food 13:04
rturk enjoy 13:04
scuttlemonkey thanks again everyone, another wicked-product summit 13:04
rturk I'm going to go lie down 13:04
rturk <- still sick 13:04
* loicd likes t-shirts, still wearing oscon t-shirts ;-) 13:04
*** sjustlaptop has quit IRC 13:06
*** rturk is now known as rturk-away 13:06
*** dmick has left #ceph-summit 13:14
*** joshd has left #ceph-summit 13:19
*** dosaboy has joined #ceph-summit 13:21
*** dalgaaf has quit IRC 13:26
*** dosaboy_ has quit IRC 13:27
*** buck has left #ceph-summit 13:30
*** davidzlap has left #ceph-summit 13:37
*** davidzlap has quit IRC 13:37
*** Vincent_Valentine has quit IRC 13:38
*** nhm has left #ceph-summit 13:51
*** ircolle1 has quit IRC 14:02
*** mikedawson has left #ceph-summit 14:06
*** dpippenger has quit IRC 14:19
*** ircolle has joined #ceph-summit 14:24
*** dosaboy_ has joined #ceph-summit 14:27
*** dosaboy has quit IRC 14:32
*** loicd has quit IRC 14:42
*** loicd has joined #ceph-summit 14:43
*** dpippenger has joined #ceph-summit 14:52
*** sjust has left #ceph-summit 14:54
*** dosaboy_ has quit IRC 14:56
*** dosaboy has joined #ceph-summit 15:04
*** ircolle has left #ceph-summit 15:09
*** nwl has left #ceph-summit 15:36
*** nwat has joined #ceph-summit 15:56
*** scuttlemonkey has quit IRC 16:26
*** nwat has quit IRC 16:33
*** zhu_ has quit IRC 16:48
*** haomaiwa_ has quit IRC 16:48
*** Tamil has left #ceph-summit 16:53
*** houkouonchi_work has quit IRC 17:04
*** dosaboy_ has joined #ceph-summit 17:22
*** dosaboy has quit IRC 17:23
*** alfredodeza has quit IRC 18:06
*** alram has quit IRC 18:09
*** joao has quit IRC 18:50
*** dpippenger has quit IRC 19:25
*** dosaboy has joined #ceph-summit 20:43
*** dosaboy_ has quit IRC 20:50
*** haomaiwang has joined #ceph-summit 21:42
*** rongze has joined #ceph-summit 23:40