12:05:48 #startmeeting 12:05:48 Meeting started Fri Feb 6 12:05:48 2015 UTC. The chair is jdarcy. Information about MeetBot at http://wiki.debian.org/MeetBot. 12:05:48 Useful Commands: #action #agreed #halp #info #idea #link #topic. 12:06:02 krishnan_p, this is just the first meeting :) 12:06:08 #meetingname GlusterFS 4.0 12:06:08 The meeting name has been set to 'glusterfs_4.0' 12:06:15 #topic Roll call 12:06:26 * shyam1 is here 12:06:26 * hagarth is here 12:06:27 * kshlm is here 12:06:29 * xavih is here 12:06:32 * krishnan_p is here 12:06:32 * kkeithley_ is here 12:06:36 Dan is here 12:06:40 me is here (obviously) 12:06:46 * ndevos is here 12:06:52 #addchair hagarth 12:06:57 #addchair krishnan_p 12:07:13 Awesome turnout. :) 12:07:26 #topic Introduction and Goals 12:07:40 OK, just a couple of words here. 12:08:09 What I'm hoping here is first for "interested parties" to get to know each other. Making good progress there. :) 12:08:16 * telmich is here 12:08:57 Second, I hope to get a good picture of what people are *really* interested in working on, and can really hope to get done sometime this summer. 12:09:09 Anyone have any other goals they'd like to add? 12:09:23 yes 12:09:34 telmich: OK, go ahead. 12:09:54 if it is not too much out of scope, I would like to add a small topic of "running glusterfs in production" - status & best practices 12:10:23 this can be related to or unrelated to 4.0 and if it is too much out of scope, it's no trouble kicking it for this meeting 12:10:43 telmich: If we can, we'll try to address that, but it might need more of a follow-on in #gluster. 12:11:08 telmich: we can also talk about this in the community meeting on Wednesdays 12:11:20 hagarth: sure thing 12:11:43 * ndevos thinks its a great idea, many users will benefit from that 12:11:45 One caveat on goals: there's still some uncertainty about resource levels and availability. Hopefully what comes out of this will be *input* to the Powers That Be who control such things (both Red Hat and elsewhere). 12:12:07 Moving on... 12:12:12 ndevos: +1 12:12:17 #topic Sub-project Status 12:12:34 Working off the page at http://www.gluster.org/community/documentation/index.php/Planning40 12:13:09 Let's start with thousand-node glusterd. krishnan_p, can you fill us in on your current half-year-or-so vision? 12:13:22 OK, let me try 12:13:42 o/ 12:13:57 We have tried to come up with two proposals to adddress the consistent configuration store problem 12:13:59 gothos: Is that a raised hand? 12:14:11 jdarcy: that is a hello 12:14:28 Both can be looked up in gluster-devel archives. 12:14:33 gothos: Ah. Hi! :) 12:15:00 krishnan_p: You guys have lots of ideas. Which particular two? 12:15:01 The proposal involving an existing consistent store like etcd or an "in-house" implementation of the same seems like the way to go 12:15:14 jdarcy, let me share the links in a bit 12:15:20 OK. 12:16:10 1) http://www.gluster.org/pipermail/gluster-devel/2014-November/042944.html 12:16:54 I am having trouble locating the other one (the one worth pursuing) :( 12:17:12 jdarcy, can I add this during the course of this meeting and not make everyone wait on me? 12:17:26 Sure. 12:17:48 * kshlm will check if I can find the other. 12:17:51 We are planning to revive the discussion on gluster-devel after 3.7 12:18:01 So the other one is more along Paxos/Raft DB lines? 12:18:10 jdarcy, yes. 12:18:36 Currently, we have Kaushal and Atin working with me on this. I hope to see more joining us. 12:19:20 The other proposal http://www.gluster.org/pipermail/gluster-devel/2014-September/042121.html 12:19:40 krishnan_p: Do you think it's likely you'll have something prototyped (at least) this summer/fall? 12:19:46 Thanks kshlm. 12:20:06 jdarcy, that is what we aspire to do. 12:20:19 Yay! That seems to be one of our more solid sub-projects. 12:20:23 The prototype could use one of consul or etcd to do the heavylifting of consensus 12:20:48 jdarcy, fall seems more doable. Summer would be too ambitious 12:20:56 We began very basic tests with when the mail was sent, and have some idea of how to proceed. 12:21:00 * jdarcy is itching to write our own, but that's just being a codemonkey. Not necessarily a good idea. 12:21:39 Anyone want to speak up on http://www.gluster.org/community/documentation/index.php/Features/dht-scalability? shyam1 maybe? 12:21:47 yup 12:21:47 jdarcy, thousand-node-glusterd has lot at stake. Just to handle that amount of details makes me run away from writing my own consensus algorithm implementation :) 12:21:56 krishnan_p: Fair enough. 12:22:10 * jdarcy (Sorry for the aggressive pace. Hoping to overlap some topics.) 12:22:21 We have a design document that details the core idea for the next generation distribution and its benefits 12:22:28 Here, http://www.gluster.org/community/documentation/index.php/Features/dht-scalability#Status 12:22:48 The main goals are scalability and consistency in DHT and to change the base requirement of needing a directory on every subvol of DHT (part of the scale issue) 12:23:06 There are some further thoughts in some of the DHT problem areas as presented in the docs above, but no code/prototype yet 12:23:22 The next plan would be to, start a prototype and visualize the benefits that DHT2 can demonstrate over existing DHT 12:23:46 Currently myself and Du are sort of thinking about this (i.e not actively working, spending large amounts of time etc.) 12:24:01 This one also has a pretty long fuse. Any thoughts on pieces that might be separately prototype-able in a short/medium timeframe? 12:24:43 I would state, we can prototype the separation of DIRS and DATA, and start from there... Other areas can come in later as we expand the problem space 12:25:03 Excellent. 12:25:24 For example cluster expansion use cases can be prototyped/coded a little later 12:25:51 On timelines, I would really like to see this move out of concept and into a strong prototype by Summer 12:26:03 shyam1, by cluster-expansion are you referring to add-brick and rebalance? 12:26:09 yes 12:26:13 Yay! 12:26:19 Next is Next is http://www.gluster.org/community/documentation/index.php/Features/sharding-xlator 12:26:36 Lots of people consider this important, but AFAIK nobody has it at the top of their list to actually work on. Correct? 12:26:47 yes 12:26:53 no 12:27:04 Kruthika and Pranith have been working on this 12:27:16 Oh. Well, that's why we have these meetings then. ;) 12:27:26 Kruthika did add this to the 3.7 feature list - so I expect some action on sharding soon. 12:27:39 +1 12:27:41 I believe we need to consider this (sharding) in lieu with DHT2 so that we can get a more cogent story... 12:27:55 So should this be moved to 3.7? 12:27:56 I know xlators are independent etc. but just stating. 12:28:05 jdarcy: yes, we can park it in 3.7 for now 12:28:30 I consider pulling stuff forward into 3.7 to be awesome. 12:28:48 OK, http://www.gluster.org/community/documentation/index.php/Features/caching 12:29:03 xavih: Do you think that spec's ready for general consumption? 12:29:29 we have an initial approach, but haven't had time to complete it 12:29:43 I'm working on some changes for ec on 3.7 12:29:58 I expect to have something ready after Feb 10 to publish 12:30:24 xavih: Do you think you'll be in a position to focus on caching after that, or are there other demands that will take precedence? 12:30:48 is the cache invalidation done like what is being worked on with the upcall infrastructure? 12:31:01 its what we need for nfs-ganesha... 12:31:14 like this: http://www.gluster.org/community/documentation/index.php/Features/Upcall-infra 12:31:18 ndevos: that is one of the things I've seen. It seems there are duplicate features... 12:31:41 xavih: ah, right, the upcall work is targetted for 3.7 12:31:57 jdarcy: I'll have some work on ec before 3.7. After that we are considering caching and dfc v2 (new design) 12:32:14 dfc? 12:32:30 ndevos: an old idea we tried to implement long ago 12:32:34 I look forward to seeing that. Should we add that to the 4.0 page perhaps? 12:32:45 ndevos: we need more time to have a good proposal 12:32:51 xavih: ah, ok 12:33:02 jdarcy: I'll try to finish a design document a publish it 12:33:14 All great. Thanks, xavih. 12:33:16 jdarcy: after that we can decide if it can be included in 4.0 12:33:28 Next up: http://www.gluster.org/community/documentation/index.php/Features/data-classification 12:34:03 cache tiering work will provide a good foundation, 12:34:03 dlambrig , I know it's not the same, but can you briefly tell us where we are with tiering? 12:34:11 * jdarcy was too slow typing. 12:35:12 I think volume level tiers are a good direciton once cache tiering is completed. 12:35:43 Also, using python in glusterd seems like a good implementation choice, that can be investigated. 12:36:13 You mean for the volfile-generation stuff? 12:36:20 cache tiering is on track for 3.7, this includes the fixes to readdirp aforementioned in the gluster-dev discussions. 12:36:47 Jeff, yes, monkeying with volfiles using python is easier than C, IMO 12:37:04 * krishnan_p wonders why not Go/Rust, but that opens a can of worms, or not ... 12:37:31 KP, I do not know the string manipulation libraries in Go but definately worth a look 12:38:00 dlambrig: Post 3.7, do you think you'll be available/interested to push toward the more general kind of data classification, or is that still in need of volunteers? 12:38:42 I will stay involved with tiering for gluster. I may not be full time, but will stay involved and would like to see the feature get better and better. 12:38:55 krishnan_p: I think for something occasional and well bounded like generating volfiles, we have more flexibility, and I already have a library (from HekaFS) for manipulating volfiles in Python. 12:39:20 Yay, thanks! 12:39:29 jdarcy, I started typing that before I knew it was only for volfile manipulation :( 12:39:52 python +1 12:39:52 All I can say about http://www.gluster.org/community/documentation/index.php/Features/SplitNetwork is that it's important but still looking for volunteers. I'd love to do it myself but don't think I'll be able. 12:40:03 Next up: http://www.gluster.org/community/documentation/index.php/Features/new-style-replication 12:40:41 I've put together a few pieces, including a new code-generation infra (so we don't have several thousand lines of repetitive code for every fop) and a full data-logging xlator. 12:41:08 http://review.gluster.org/#/c/9411/ is the codegen infra. 12:41:24 https://forge.gluster.org/~jdarcy/glusterfs-core/jdarcys-glusterfs-data-logging is the data-logging part 12:41:28 * krishnan_p thinks codegen is awesome! 12:41:48 Hoping to get NSR as it was a year ago revived in the next few weeks, and fully (?) done this summer. 12:42:02 * hagarth will drop off in about 10 minutes from now 12:42:16 krishnan_p: Yeah, I want to do stubs, syncops, RPC, etc. that way instead of not-quite-identical separate code for every fop. 12:42:28 jdarcy: that would be fantastic 12:42:37 jdarcy, I am waiting for that day! 12:42:49 Also, adding new fops might require touching fewer than twenty files. 12:43:05 Next up: http://www.gluster.org/community/documentation/index.php/Features/better-brick-mgmt 12:43:08 jdarcy, that requires a separate feature page IMO 12:43:15 krishnan_p: OK. 12:43:29 #action jdarcy to add separate feature page for codegen infrastructure 12:43:53 * krishnan_p thanks jdarcy for adding a feature page for codegen 12:44:22 BBM is also a bit of an orphan. Again, I think it's really important but probably won't be able to do it myself. Anyone have any thoughts on whether to keep it? 12:44:56 I think its a much needed featuer 12:45:03 feature even 12:45:13 Overlapping with that, I think the other items (small-file perf and other, except for maybe stat/xattr cache) are also orphans. 12:45:36 jdarcy: maybe we can ciruclate a list of features needing volunteers on -devel 12:45:43 Anybody know of any activity or interest in picking those up? 12:45:43 hagarth, +1 12:45:47 hagarth: Good idea. 12:45:59 was small file mostly a wish list of small fixes from Ben , or a large stand alone project 12:46:04 #action jdarcy to poll gluster-devel for volunteers on "orphan" subprojects 12:46:15 dlambrig: It was culled from Ben's list. 12:46:48 jdarcy: I will try to keep knocking off ideas in small-file-perf. I can be an interim owner for that. 12:47:13 hagarth: I think there has been *some* progress on stat/xattr cache (md-cache on server). Who's doing that? 12:47:24 jdarcy: I have patches sitting in my local repo 12:47:47 krishnan_p: merging volgen re-factor will help expedite this ;) 12:48:17 Volgen refactor? 12:48:18 hagarth, I will merge it. I got carried away by some 3.7 related activities 12:48:30 jdarcy: a basic brick graph generation refactor 12:48:40 jdarcy, http://review.gluster.com/#/c/9521/ 12:48:42 could not withstand the ugliness that existed there 12:48:47 Most excellent. That code's getting really hard to work on. 12:49:28 I'll have to see how that interacts with http://review.gluster.org/#/c/9387/ 12:49:40 * krishnan_p appreciates any help anyone can offer in cleaning up code in glusterd. Its not a one person's work. 12:49:54 +1 12:50:05 In the interests of time . . . meeting times and formats. 12:50:14 #topic Meeting Times and Formats 12:50:20 Does this time work for people? 12:50:36 I'm concerned that it's inconsiderate for BLR people who might want to start their weekends. 12:50:48 move it to Thursday maybe or setup a doodle poll? 12:51:07 jdarcy, Any day other than Friday should be OK for me. 12:51:14 * jdarcy is tempted to do a poll on whether to use Doodle. A meta-poll. 12:51:19 jdarcy: lol 12:51:28 #action jdarcy to set up a poll for meeting times 12:51:38 with that, I shall drop off now. need to head to my talk in devconf.cz. 12:51:41 later everyone 12:51:49 hagarth, bye 12:51:52 hagarth: Have fun! 12:52:07 How about frequency? Every other week, maybe? 12:52:20 sounds good to me 12:52:28 sounds good to me too 12:52:35 sounds good to me too. 12:52:41 Also,, I'd say let's stick with IRC for now. Seems to be working, and leaves a record. 12:52:50 Nicely searchable, and everything. 12:53:23 #topic Collaboration Tools 12:53:26 jdarcy, IRC works 12:53:39 Are people reasonably satisfied with using the wiki, or do we need/want something else? 12:54:04 https://freedcamp.com/ looks really nice, but non-OSS and needs each person to be invited 12:54:20 Haven't found a way to make a project fully public. 12:54:26 wiki should do, right? have all the information at one place? 12:54:36 Well wiki is useful, when better things are needed the wiki can serve as a link redirection point 12:55:12 I would like to know future aims of this meeting, operational/technical discussions, based on which collaboration medium may change 12:55:39 shyam1: I think mostly sharing status and issues that have arisen during the last $interval 12:55:45 I mean if we were presenting a proposal, IRC would not cut it 12:55:52 Ah ok, then IRC is just fine 12:56:32 Also calls to action - please review this proposal/patch, that kind of thing 12:56:36 whatever media, everything should be recorded and made available for everyone in the community 12:56:43 ndevos: +10 12:57:07 * ndevos *hates* phone conferences other project(s) do 12:57:18 OK, so wiki for now, people are *always* welcome to bring other alternatives to our collective attention. 12:57:38 #topic Open Floor 12:57:53 Anyone have anything else they want to discuss before we close 12:57:55 ? 12:58:05 telmich's topic? 12:58:58 A good topic, but probably more than a few minutes' worth. Move to #gluster perhaps? 12:59:01 maybe we should have something like a "best practise" or "advised usage" for each of the features? 12:59:07 Or #gluster-somethingelse? 12:59:29 User stories? ;) 12:59:47 user stories with techincal details on the configuration and all 13:00:19 That's a pretty good idea. Should I start a wiki page to collect those? 13:00:43 I guess we need something lika that, yeah 13:00:47 We *could* put those in the existing feature pages, but that might be kind of fragmented. 13:01:02 yes, and not each feature is really user visible 13:01:09 ndevos: Would you mind taking the AI for that? 13:01:16 or, features need to be combined somehow 13:01:35 hmm, not really, I do not have time for that in the next few wekks 13:01:37 *weeks 13:01:52 #action jdarcy to create wiki page for user stories, send out CFP 13:02:29 thanks! 13:02:30 Anything else? 13:02:45 nothing else from me 13:03:02 no, nothing 13:03:04 OK then, thank you thank you thank you everyone. :) 13:03:13 #endmeeting