12:05:02 <davemc> #startmeeting Weekly GlusterFS Community Meeting
12:05:02 <zodbot> Meeting started Wed Oct  8 12:05:02 2014 UTC.  The chair is davemc. Information about MeetBot at http://wiki.debian.org/MeetBot.
12:05:02 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic.
12:05:15 <davemc> So, who's here
12:05:25 * davemc is here
12:05:26 * lalatenduM is here
12:05:30 * xavih is here
12:05:31 * overclk is here
12:05:36 * jdarcy is ubiquitous, and therefore here.
12:05:42 <bene> here
12:05:49 <davemc> Etherpad: https://public.pad.fsfe.org/p/gluster-community-meetings
12:06:11 <davemc> #topic Action items from previous meetings
12:06:40 <davemc> I wasn't at the last meeting, so its all new to me
12:07:28 * kkeithley is here
12:07:32 * jdarcy doesn't see hagarth.  Can anyone else speak for his items, or should we table them?
12:07:42 <davemc> Humble actions for Soumya marked complete
12:08:12 <davemc> anyone speak for hagarth?
12:08:36 <davemc> tabling item till later
12:09:02 <davemc> JustinClift updated his items in the pad.
12:09:22 <davemc> Homebrew formula is done
12:09:45 <bene> I would like to discuss small-file perf. when u get to it
12:09:49 <davemc> Free BSD, FreeNAS communities contacted
12:09:58 <davemc> bene ok
12:10:48 <davemc> lalatenduM, anything new on dev@fedoraproject.org
12:10:59 <davemc> still waiting on 3.6.0?
12:10:59 <lalatenduM> davemc, nope
12:11:09 <lalatenduM> yeah , waiting till GA
12:11:15 <lalatenduM> as updated in the etherpad
12:11:19 <davemc> k
12:11:47 <davemc> JustinClift on Rackspace in progress
12:12:16 <davemc> overclk to loop in Joe|Dan on BitRots?
12:12:28 <davemc> overclk, any update?
12:12:41 <overclk> davemc: did the needful.
12:13:05 <overclk> davemc: currently, they're perf testing different db's to choose from.
12:13:05 <davemc> k
12:13:12 <overclk> davemc: will update the pad.
12:13:32 <davemc> thx
12:13:54 <davemc> no hagarth still.
12:14:19 <davemc> anyone got any idea on     hagarth to review cppcheck and http://review.gluster.org/#/c/8213/ this week
12:14:43 <davemc> silence
12:15:07 <davemc> ndevos and hagarth to discuss on gluster-devel (by 24th Sept) the outstanding 3.5.3 blocker BZ's, and which ones to move to 3.5.4
12:15:35 <davemc> I don't think either are on today
12:16:01 <davemc> davemc to get the GlusterFS Consultants and Support Company's page online
12:16:30 <davemc> Still not done. Have confirmed that the git clone and live site are at least in sync
12:16:41 <davemc> Have taught myself HAML
12:16:42 <hchiramm_> davemc, Soumya will be sending out an email  regarding this..
12:17:07 <davemc> HAve identified many, many other errors or incorrect information
12:17:27 <hchiramm_> Soumya Deb is helping us to improve gluster.org
12:17:39 <davemc> so we need to figure out how to get updates done regularly
12:18:16 <davemc> hchiramm_, great, it needs some definite improvement
12:18:29 <hchiramm_> yep..
12:18:44 <hchiramm_> he is making a draft of proposals and will be reaching out to gluster-infra soon on this.
12:18:53 <davemc> kkeithley to  put 3.6.0 beta / rc on the Longevity Cluster
12:18:56 <davemc> marked done
12:19:01 <kkeithley> yes, done
12:19:39 <kkeithley> it's running, you can see the results at
12:19:47 <kkeithley> one sec
12:20:04 <bene> ...drum roll....
12:20:08 <kkeithley> http://download.gluster.org/pub/gluster/glusterfs/dynamic-analysis/
12:20:34 <davemc> very cool
12:21:45 <davemc> I think I forgot to type topic before each of these. ouch
12:21:54 <davemc> #topic 3.4
12:22:52 <davemc> any updates or actions?
12:22:53 <kkeithley> nothing to report, still waiting for two sets of patches: Raghavendra's DHT patch set, and the memory leak. Anoop (?) was looking at the memory leak. I need to circle back with him
12:23:21 <raghug> kkeithley: davemc was busy with backporting patches to other downstream branches
12:23:23 <kkeithley> BLR devs have been working on the RHS 3.0 release. Now we hope to get some cycles to get these addressed
12:23:41 <raghug> we will try to close this by next gluster-meeting
12:23:44 <davemc> #action kkeithley to loop back on memory leak
12:24:06 <davemc> anything else on 3.4?
12:24:18 <davemc> #topic 3.5
12:24:31 <davemc> Updates or actions
12:24:35 <davemc> ?
12:24:59 <hchiramm_> 3.5.3beta1 is released.. :)
12:25:18 <davemc> woo hoo!
12:25:42 <davemc> off topic, do we have a target release for 3.6.0?
12:25:54 <davemc> sorry, next topic
12:26:06 <davemc> anything more on 3.5?
12:26:15 <davemc> #topic 3.6
12:26:25 <hchiramm_> 3.6.0beta4 is about to release this week ..
12:27:18 <davemc> yep. the dates for the website are one of the errors we need to correct.
12:27:26 <davemc> still shows GA in September
12:27:47 <hchiramm_> we released 3 betas on 3.6.0 already , may be it pushed the dates..
12:28:10 <davemc> #action correct 3.6.0 dates in website
12:28:20 <hchiramm_> afaik , hagarth was planning to push 3.6 GA by second week of Oct.
12:28:30 <davemc> tkx
12:29:02 <davemc> If so, I can taalk about it at Strata/Hadoop World
12:29:27 <davemc> tracker issues?
12:29:38 <davemc> Open BZs?
12:30:02 <davemc> going, going,
12:30:15 <davemc> #topic small-file performance feature page
12:30:33 <jdarcy> I added a bit on dentry injection.
12:30:45 <bene> yes I saw that.
12:31:09 <bene> I had thought that this was covered by DHT V4 reference originally
12:31:43 <bene> or whatever you call next-gen DHT
12:32:17 <bene> but I see now it's not
12:32:23 <jdarcy> DHT:TNG?  NSDHT?
12:33:18 <jdarcy> (OT, but we need to get away from "DHT" terminology because it's not a real DHT as a CS person would use that term.)
12:33:57 <bene> I like DHT:TNG, since I'm a trekkie.   So are you talking about prefetching directory entries into FUSE somehow?  Or just speeding up readdir() in translators?
12:34:12 <bene> anyway I'll ask you later if everyone else understands
12:34:57 <davemc> fascinating page of information there
12:35:13 <bene> can we pop the stack and just see what items can & should get done in Gluster 3.7?
12:35:45 <jdarcy> Sounds good.
12:35:56 <davemc> I think that would be a good plan. In the last two months I'd heard more about small file/many files performance than any other topic
12:36:21 <davemc> Is that worthy of an action?
12:37:11 <davemc> if not, I'm assigning it as an idea, so we don't loose it
12:37:29 <bene> I'd said my opinions in the URL, I'd like to hear from people here about what is possible and practical.
12:37:50 <davemc> #idea  pop the small file performance stack and just see what items can & should get done in Gluster 3.7
12:38:03 <jdarcy> Multi-threaded epoll is doable, if we *force* SSL connections to use their own threads and thus remove themselves from the main polling infra.
12:38:39 <jdarcy> xattr/stat cache is doable, since md-cache already does that for some things.
12:38:41 <davemc> #action get wider review of small file ideas for doable and practical
12:39:13 <jdarcy> lookup-unhashed can and should be doable in that timeframe, along with the simpler tiering suggestions.
12:39:26 <jdarcy> Other than that, not so sure.
12:40:16 <davemc> any more comments? thoughts?
12:40:33 <kkeithley> one thought about 3.4.6
12:40:58 <kkeithley> maybe we should release with what we've got, and do 3.4.7 once the DHT and memleak are ready?
12:42:14 <davemc> remind me, we don't have ETA for DHT and memleak yet. That's in hagarth list?
12:42:41 <kkeithley> No, no ETA per se
12:42:46 <raghug> kkeithley: dht patches can be merged by next Wednesday
12:42:56 <bene> about smallfile, I think if we did the 3 things Jeff describes, this would make a big difference.    Can we start with just those?
12:42:57 <raghug> I am planning to work on them from tomorrow
12:42:57 * davemc needs more sleep/caffeine/adrenaline
12:42:59 <kkeithley> oh, excellent
12:43:09 <kkeithley> raghug: ^^^
12:43:23 <davemc> raghug++
12:43:56 <bene> who uses SSL sockets today?
12:44:17 <davemc> we should check into the memleak fix to decide on 3.4.6 or a .6/.7 splie
12:44:24 <davemc> s/splie/split
12:45:02 <davemc> any more before we hit next topic?
12:45:14 <davemc> #topic other agenda items
12:45:26 <davemc> anything we need to bring up?
12:45:26 <raghug> kkeithley: davemc do you the bug-id for mem-leak bug?
12:45:29 <jdarcy> bene: I don't have the names off the top of my head, but I know of several.
12:45:52 <raghug> just wanted to understand what issue are you speaking about
12:46:13 <davemc> Is it 119894? not sure
12:46:26 <davemc> or 1127140?
12:46:38 <jdarcy> bene: There was a question about it on an RH internal list this past week, with an implication that many users in the "federal" (US government) space would consider it a strong differentiator.
12:46:58 <hchiramm_> https://bugzilla.redhat.com/show_bug.cgi?id=1144792
12:47:06 <hchiramm_> may be ^^^ ?
12:47:16 <kkeithley> 1127140 I think
12:47:41 <davemc> https://bugzilla.redhat.com/show_bug.cgi?id=1127140
12:47:56 <hchiramm_> yeah , looks like 1127140
12:48:36 <hchiramm_> raghug, https://bugzilla.redhat.com/show_bug.cgi?id=glusterfs-3.4.6 -> is tracker bug, its one of it..
12:48:43 <bene> jdarcy, thx, I wasn't aware, won't SSL sockets benefit from multi-thread epoll as well?  SSL is more CPU intensive.   I know it makes it harder to do
12:49:15 <raghug> seems like we don't have an RCA yet on 1127140
12:49:55 <raghug> and not sure whether anyone is actively working on it too
12:50:17 <kkeithley> raghug: correct. Anoop, I think, contacted me about it.
12:50:46 <davemc> maybe we should then split 3.4.6 for the DHt stuff, and promote memleak to a 3.4.7
12:51:00 <kkeithley> correct, as in we don't have a root cause analysis of the leak yet
12:51:11 <raghug> even I think so
12:51:39 <raghug> and steps to reproduce the issue are not there too
12:51:56 <davemc> #action look into releasing 3.4.6 with DHT and promoting memleak fix to a 3.4.7 release
12:51:56 <raghug> so, am not sure how easily this leak can be hit
12:52:22 <davemc> raghug, in fact, the reporter states he isn't sure how to reproduce
12:52:45 <jdarcy> bene: As I've said *dozens of times before* over the last two years, we already have own-thread to deal with SSL's CPU needs.
12:52:47 <kkeithley> no, reproducer, but when I was running 3.5.2 on longevity the client glusterfs process was growing. I suspect it might be the same leak? (But it's only a suspicion.)
12:53:10 <davemc> kkeithley, ah
12:53:23 <raghug> kkeithley: there are some bugs related to leaks
12:53:37 <raghug> may be they are same or they are not
12:53:49 <raghug> even on RHS 3.0 there was an issue
12:53:53 <bene> jdarcy, I'll ask you later...  So basically it's fine with me if we segregate SSL sockets, how badly does that break the code
12:54:09 <davemc> 5-ish minutes before my next meeting. Any other topics?
12:54:47 <jdarcy> bene: I don't think it breaks the code at all.  Using SSL without own-thread is kinda dumb anyway, no reason to continue allowing it.
12:55:49 <jdarcy> In other news, I have the configuration and I/O paths of NSR working against current master.
12:55:59 <bene> so am I hearing that we go ahead with 3 items for Gluster 3.7, multi-thread-epoll, stat+xattr-caching, and lookup-unhashed=auto?  We could see significant gains with just those for now
12:56:01 <jdarcy> No reconciliation yet.
12:56:22 <kkeithley> so maybe agreed for 3.4.6beta with the DHT and we'll do 3.4.7 when we get a fix for the memleak?
12:56:24 <davemc> bene agreed that could be a great win
12:56:40 <davemc> kkeithley, also agreed by me
12:57:48 <davemc> objections to a 3.4.7 plan?
12:58:13 <davemc> #agreed release a 3.4.6beta with the DHT, do 3.4.7 when we get a fix for the memleak
12:58:41 <bene> btw, re smallfile,  i assumed that we're continuing with data classifier work started by jdarcy and Lambright.  smallfiles and erasure coding don't mix
12:58:47 <davemc> okay, folks. I'm going to end the meeting with your kind permission
12:58:59 <bene> thanks davemc, bye
12:59:06 <jdarcy> Thanks davemc.
12:59:12 <raghug> thanks davemc bye
12:59:20 <davemc> #endmeeting