15:00:09 <eglynn-office> #startmeeting ceilometer
15:00:10 <openstack> Meeting started Thu Oct 23 15:00:09 2014 UTC and is due to finish in 60 minutes.  The chair is eglynn-office. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:11 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:00:14 <openstack> The meeting name has been set to 'ceilometer'
15:00:20 <DinaBelova> o/ hello folks!
15:00:30 <eglynn> hey y'all
15:00:44 <llu-laptop> o/
15:00:50 <_nadya_> o/
15:00:55 <ildikov_> _o_
15:01:01 <gordc> o/
15:01:05 <fabiog> o/
15:01:36 <eglynn> let's get gnocchi out of the way first before we jump into the summit scheduling
15:01:42 <eglynn> #topic TSDaaS/gnocchi status
15:02:20 <cdent> o/
15:02:21 <eglynn> jd__: how goes it with the pasta maker?
15:02:34 <nealph_> o/
15:02:39 <idegtiarov> o/
15:03:15 <DinaBelova> eglynn, don't know about jd__'s update, but I have some about the dispatcher via API performance :)
15:03:27 <eglynn> DinaBelova: shoot!
15:03:42 <DinaBelova> quick hack to avoid API calling makes things x10 times better ;)
15:03:47 * EmilienM reminds it's jd__ birthday :)
15:04:10 <eglynn> EmilienM: a-ha, cool
15:04:11 <DinaBelova> eglynn, so direct DBs calls from the same dispatcher is 10x-12x times quicker
15:04:35 <DinaBelova> ityaptin is making nice pictures right now
15:04:44 <eglynn> DinaBelova: direct calls to the gnocchi storage driver?
15:04:57 <DinaBelova> and indexer in the needed places
15:05:00 <DinaBelova> eglynn, yeah
15:05:30 <DinaBelova> eglynn: so the thing is remove API calls -> gain 10x perf
15:05:46 <eglynn> DinaBelova: so we'd get the horizontal scaling by running multiple collectors as opposed to multiple gnocchi-api services
15:05:59 <DinaBelova> eglynn, indeed
15:06:29 <DinaBelova> in the way we're working now for not may collectors - three of them work definitely better than one
15:06:46 <eglynn> DinaBelova: out of interest, what was jd__'s reaction to that idea of by-passing the API layer?
15:06:55 <fabiog> DinaBelova: are you doing any batching for the writes?
15:07:16 <DinaBelova> eglynn, that was out own test actually
15:07:30 <DinaBelova> eglynn, jd__ did not react somehow on the idea of this trying
15:07:37 <DinaBelova> fabiog, both
15:08:02 <eglynn> IIUC gnocchi-api only really allows writes to the *same* entity to be batched?
15:08:06 <DinaBelova> fabiog, the thing I'm talking now about - 100 measures batches are written 10x times faster without API then with it
15:08:16 <DinaBelova> eglynn, hmmm
15:08:22 <DinaBelova> yes
15:08:22 <eglynn> whereas samples generally arrive together for a mix of entities
15:08:44 <DinaBelova> eglynn, so that's kind of syntethic situation of batching, yeah
15:09:05 <eglynn> DinaBelova: how did the perf testing go for the plugged in TSDBs?
15:09:15 <eglynn> (i.e. opentsdb/influx)
15:09:32 <DinaBelova> eglynn, we did not try influx yet :(
15:09:40 <DinaBelova> we were having deployment issues...
15:09:53 <DinaBelova> with OpenTSDB + gnocchi :)
15:10:02 <DinaBelova> so influx is the next step here
15:10:08 <eglynn> cool enough
15:10:44 <eglynn> so seems like this direct-to-storage idea will require some more discussion/buy-in
15:10:55 <eglynn> but the initial perf gain is hard to ignore
15:11:13 <DinaBelova> eglynn, well, at least it's making the Gnocchi POC performance close to current Ceilometer one
15:11:22 <eglynn> yeap
15:11:23 <DinaBelova> eglynn, indeed
15:12:01 <eglynn> in other gnocchi news ...
15:12:10 <eglynn> some iteresting discussion on mixed granularities in sileht's cross-entity aggregation patch https://review.openstack.org/125572
15:12:49 <eglynn> I'm thinking that we need to require an explicit choice of the required granularity on GET
15:13:14 <eglynn> i.e. never mix granularities in a query response
15:13:29 <sileht> o/
15:14:04 <eglynn> sileht: hey ... that make sense about the mixing?
15:14:59 <sileht> eglynn, current the code bail out if sometimes is different between the timeseries dimensions
15:15:02 <jd__> o/
15:15:15 <eglynn> jd__: the birthday boy :)
15:15:29 * cdent raises a glass at jd__
15:15:35 <DinaBelova> jd__, happy b-day!
15:16:20 <eglynn> sileht: even in the case where the dimensions match, I'm thinking we should require the caller of GET measures to choose the granularity they'be interested in
15:16:20 <sileht> jd__, bonne anniversaire :p
15:16:34 <jd__> thanks guys!
15:16:48 <ildikov_> jd__: happy birthday :)
15:16:59 * cdent is a fan of explicit queries
15:17:06 <eglynn> sileht: ... so that we never mix say hourly and daily datapoints in a single response
15:17:11 <sileht> eglynn, instead of return the only values with have (in case of dimemsion match across entities) ?
15:17:25 <sileht> eglynn, oh I see
15:17:33 <jd__> DinaBelova: about API bypass, did you publish the patch?
15:18:03 <DinaBelova> jd__, not yet, as said, it was just a quick OpenTSDB related hack to measure the possible perf improvement
15:18:03 <eglynn> sileht: yeah, coz of the overwriting issue in the result dict and the difficulty in telling the granularity apart in the response
15:18:16 <sileht> eglynn, to not have more and more space between timestamp because of different archive policies
15:18:22 <DinaBelova> jd__, just saying it looks 10x times better without pecan apu
15:18:23 <DinaBelova> api*
15:18:51 <DinaBelova> jd__, we're still struggling to find reason for the / behavior
15:18:57 <DinaBelova> jd__, sorry, no results yet
15:19:18 <eglynn> jd__: what's your gut reaction to the API-bypass idea?
15:19:33 <jd__> DinaBelova: ok, I'm curious too
15:19:52 <jd__> eglynn: there's logic we need in the REST API so I'm not sure how it's bypass and if it is properly
15:20:05 <jd__> also I think the problem with the current performance is also tied to the sample model we have currently
15:20:11 <jd__> that is clearly not optimized for Gnocchi
15:20:13 <sileht> eglynn, hum what is the overwritting issue  ?
15:20:30 <jd__> it was designed for our way of collecting the measures
15:20:30 <DinaBelova> jd__, eglynn - it's a quick hack now with API bypassing, API is doing much checking work now...
15:20:51 <DinaBelova> so it won't work in every case - that was done only to measure the perf with this case
15:21:35 <eglynn> sileht: e.g. 1 minute data with say 2014-10-23T15:00:00 timestamp overwrites 5min or hourly data with the same timestamp in the response dict
15:21:43 <eglynn> (IIUC)
15:21:47 <DinaBelova> jd__, ityaptin and I may try to work on more preferrable API bypassing solution
15:21:58 <DinaBelova> that will keep flexibility with speed
15:22:17 <DinaBelova> jd__, we'll work on it during our tests for sure
15:22:19 <jd__> DinaBelova: ok, let me know how that goes
15:22:26 <DinaBelova> jd__, thanks!
15:22:37 <eglynn> sounds like good material for the "Bringing Gnocchi and Ceilometer together" session at summit
15:22:46 <DinaBelova> eglynn, for sure!
15:22:48 <eglynn> ... speaking of which
15:22:59 <eglynn> should we move on to summit scheduling?
15:23:07 <cdent> maybe we can use slow api performance as an excuse to trash (as in remove) pecan ;)
15:23:32 <sileht> eglynn, you think in case of both entity have the datapoint with the same timestamp ?
15:23:43 <eglynn> cdent: hmmm, the wind has been blowing in favour of pecan for a while
15:24:06 <cdent> object-dispatch is a broken model
15:24:20 <fabiog> cdent: I liked webOB ...
15:24:25 <cdent> but that's just my NSHO...
15:24:29 <eglynn> sileht: yes, I thought so, if the timestamp is a "shared" boundary
15:24:41 <DinaBelova> cdent, well, at least we may avoid it where we have other variants
15:24:50 * cdent is sorry for adding more weeds
15:24:51 <eglynn> sileht: e.g. 15:00:00 is a 1min, 5min, 1hr etc. boundary
15:24:52 <sileht> eglynn, I guess my code works in this case I will a unit tests
15:24:52 <cdent> scheduling?
15:25:17 * eglynn wonders what NSHO stands for?
15:25:26 <eglynn> cdent: yep, let's do that
15:25:31 <cdent> not so humble opinion
15:25:32 <jd__> Not So Humble Opinion I'd say
15:25:36 <cdent> jinx
15:25:46 <eglynn> a-ha :)
15:25:49 <eglynn> #topic finalize kilo summit track
15:26:11 <sileht> eglynn, oh I got it, I see, fetch return the smaller granularaty
15:26:13 <eglynn> the boilerplate schedule is here
15:26:25 <eglynn> #link http://kilodesignsummit.sched.org/
15:26:36 <sileht> eglynn, I understand why this is more safe to add the granularaty  to the API param
15:26:37 <eglynn> not filled in for most projects yet
15:26:49 <eglynn> sileht: cool
15:27:12 <eglynn> the hard deadline to have the final schedule pushed is Tuesday next
15:27:40 <eglynn> but initial version of our track should go up tmrw if we can agree now
15:27:51 <jd__> is there a draft?
15:28:08 <eglynn> well I tallied the votes on http://bit.ly/kilo-ceilometer-summit-topics
15:28:21 <eglynn> and the following had most support
15:28:29 <eglynn> Configuration (pipeline) using persistent data store (20mins)
15:28:36 <eglynn> Bringing Gnocchi and Ceilometer together
15:28:49 <eglynn> Functional areas to target for collaboration with Monasca
15:28:56 <eglynn> Notifications {as-a-contract, standardization, maximization}
15:29:09 <eglynn> Switchover to in-tree functional tests (20 mins?)
15:29:16 <eglynn> Mapping gnocchi semantics to InfluxDB/OpenTSDB
15:29:25 <eglynn> Kafka Publisher (20mins)
15:30:07 <jd__> that makes 5.5 slots on 6, amirite?
15:30:19 <fabiog> 5
15:30:19 <eglynn> jd__: well I'm not 100% sure 20mins will do it for the in-tree tests
15:30:31 <jd__> ok
15:30:47 <jd__> looks good to me I guess
15:30:51 <eglynn> so if we were to promote that to a full slot, we'd have 6
15:30:58 <ildikov_> testing is always a long discussion, IMHO
15:31:03 <DinaBelova> ildikov_ ++
15:31:09 <eglynn> what do folks think of that list?
15:31:21 <DinaBelova> eglynn, looks nice imho
15:31:25 <cdent> +1
15:31:30 <ildikov_> eglynn: it looks ok to me
15:31:44 <eglynn> fabiog: are you good to proxy for nealph_ for the pipeline config session?
15:31:53 <fabiog> eglynn: yes
15:32:14 <eglynn> fabiog: cool, thank you sir!
15:32:18 <eglynn> ildikov_ also has ideas on that one BTW
15:32:23 <fabiog> eglynn: I took the liberty of adding a column to the google docs for the friday topics
15:32:29 <ildikov_> eglynn: fabiog: I will try to give some support too regarding to this session
15:32:35 <eglynn> fabiog: yes, I liked that
15:32:46 <fabiog> eglynn: can we also have a list of topics for friday?
15:33:09 <eglynn> so we have a full day on Friday, so I'd hope we could accomodate most or pretty all the remaining topics
15:33:14 <fabiog> eglynn: I think will help structure the discussions. IMHO :-)
15:33:22 <ildikov_> and maybe save some slots there for those discussions that we couldn't finish during the design sessions?
15:33:25 <nealph_> thumbs up
15:34:10 <eglynn> fabiog: for Friday the intent is to be free-flowing, but I can put together a list on an etherpad or some-such
15:34:26 <fabiog> eglynn: I would appreciate it
15:34:32 <eglynn> fabiog: np, will do!
15:34:49 <_nadya_> eglynn: I have one question about Kafka publisher. I think that Kafka is only one option. Separate publishers may be created for OpenTSDB or Graphite. I will not be at summit, so I hope you will discuss the topic like 'Additional publishers' :)
15:35:19 <eglynn> _nadya_: cool, please add that to the topic etherpad when it goes up (to remind us)
15:35:29 <_nadya_> eglynn: ok!
15:35:41 <eglynn> BTW I also got a half-session on the cinder track, on surfacing "capacity headroom"
15:36:00 <eglynn> plus a similar topic for neutron at their Friday session
15:36:11 <eglynn> (WtE WG related)
15:36:59 <llu-laptop> eglynn: so that are mainly about notify ciner/neutron's own 'capacity'?
15:37:28 <eglynn> llu-laptop: yeah, unallocated space/pyhsical-resources basically
15:37:47 <eglynn> how much more stuff can I fit in my cloud
15:38:04 <eglynn> operators want to know such things, so I'm told
15:38:33 <eglynn> BTW looking at the scheduled slot for the ceilo track, does anyone leading a session have any conflicts?
15:38:44 <eglynn> i.e. times they *can't* do?
15:40:57 <eglynn> k, that makes things simpler :)
15:41:44 <eglynn> is that it for the summit?
15:42:19 <eglynn> #topic open discussion
15:42:20 <DinaBelova> eglynn, I suppose yep
15:43:35 <cdent> you guys gonna get me, nealph_ and _nadya_ pressies while in paris?
15:44:33 <eglynn> cdent: I'll get ya a bottle of the finest wine known to humanity :)
15:44:58 <gordc> cdent: urbandictionary gives wildly differing definitions of "pressies"
15:45:02 <DinaBelova> eglynn, french rose one made me really crazy this summer
15:45:04 <DinaBelova> gordc ++
15:45:11 * DinaBelova was looking there just now
15:45:11 <eglynn> ... you'll have to drop over to Dublin to drink it though
15:45:12 <cdent> uh oh
15:45:29 * _nadya_ smiles :)
15:45:55 <cdent> wow gordc, I just don't know what to say about that
15:46:17 <cdent> I heard the enovance party was pretty wild though, so who knows
15:46:29 <gordc> cdent: lol
15:46:30 <DinaBelova> cdent, lol
15:48:40 <eglynn> speaking of parties ... the Red Hat / eNovance shindig clashes with the cores' party yet again :(
15:49:15 <nealph_> eglynn: thinking it's a conspiricy by non-cores now?
15:49:35 <nealph_> s/conspiricy/conspiracy
15:50:47 <cdent> tower, mayday, we're spiralling out of control
15:50:54 <DinaBelova> eglynn, it's kind of tradition
15:51:04 <eglynn> yeap, all the above :)
15:51:11 <jd__> eglynn: yeah :( and a concert I'm going to… :(
15:51:19 <eglynn> jd__: Damien Rice?
15:51:24 <jd__> eglynn: yeah <3
15:51:41 <eglynn> coolness, I've seen him live a few times in Dublin
15:51:55 <cdent> my lovely wife will be jealous
15:51:56 <eglynn> ... back in the +LisaHanigan days
15:52:03 <jd__> he might be my favorite artist :)
15:52:04 <jd__> hehe
15:52:39 <jd__> first gig I went in Paris they would have just broke a week before…
15:52:57 <eglynn> jd__: is that gig sold out now?
15:53:10 <jd__> eglynn: it was sold out in an hour
15:53:25 <eglynn> wowser!
15:53:40 <eglynn> (markmc was talking about going, obviously missed the boat on that)
15:54:45 <cdent> summit is not next week but the week after, yes?
15:54:50 <jd__> yep
15:54:57 <eglynn> cdent: Nov 3-7
15:55:00 * cdent is planning his takeover of he codez
15:55:05 <eglynn> cdent: LOL :)
15:56:02 <eglynn> k, let's call it a wrap a few mins early ... thanks for your time, folks!
15:56:08 <eglynn> #endmeeting ceilometer
15:56:11 <DinaBelova> eglynn, thanks!
15:56:23 <fabiog> eglynn: thanks
15:56:26 <DinaBelova> eglynn, it seems bot forgot about us
15:56:28 <DinaBelova> :-\
15:56:57 <DinaBelova> eglynn, no openstack bot around
15:57:50 <llu-laptop> bot is thinking its plan for summit :)
15:58:00 <jd__> haha
15:58:05 <eglynn> DinaBelova: eekk, an unlogged meeting!
15:58:06 <jd__> openstack: coucou
15:58:11 <DinaBelova> one moment
15:58:16 * DinaBelova in infra channek
15:58:23 <eglynn> ... like a tree falling unheard in the forest
15:58:46 <jd__> then, did it fell if nobody heard?
15:58:49 <jd__> you got 4 hours.
15:58:55 <eglynn> exactly :)
16:00:21 <DinaBelova> eglynn: [19:59:48]  <fungi>	 ping timeout at 15:07:36 utc
16:00:27 <DinaBelova> eglynn, bot died really early