Friday, 2014-11-14

*** packet has quit IRC00:12
*** amalagon has joined #openstack-ceilometer00:20
*** fnaval has quit IRC00:24
*** david-lyle is now known as david-lyle_afk00:31
*** fnaval has joined #openstack-ceilometer00:43
openstackgerritSrinivas Sakhamuri proposed openstack/ceilometer: Internal error with period overflow
*** asalkeld has joined #openstack-ceilometer00:57
*** david-lyle_afk has quit IRC01:08
*** Viswanath has joined #openstack-ceilometer01:09
*** Viswanath has quit IRC01:12
*** rwsu has quit IRC01:18
*** ssakhamuri has quit IRC01:25
*** nosnos has joined #openstack-ceilometer01:43
*** Yanyanhu has joined #openstack-ceilometer01:58
*** _cjones_ has quit IRC02:23
*** mc__ has joined #openstack-ceilometer03:28
*** nosnos has quit IRC03:30
*** nosnos has joined #openstack-ceilometer03:30
*** nosnos has quit IRC03:35
mc__not clear to me: network.incoming.bytes.rate uses 1024bytes or 1000bytes as unit, when talking about usage of SI-Unit?03:38
*** Longgeek has joined #openstack-ceilometer03:39
*** jd__ has quit IRC03:44
*** jd__ has joined #openstack-ceilometer03:44
*** SergeyLukjanov has quit IRC03:45
*** Longgeek has quit IRC03:45
*** d0ugal has quit IRC03:45
*** SergeyLukjanov has joined #openstack-ceilometer03:47
*** d0ugal has joined #openstack-ceilometer03:48
*** d0ugal is now known as Guest3255603:48
*** ssakhamuri has joined #openstack-ceilometer03:52
*** ryanpetrello has joined #openstack-ceilometer04:06
*** ryanpetrello has quit IRC04:13
*** ryanpetrello has joined #openstack-ceilometer04:18
*** nosnos has joined #openstack-ceilometer04:27
*** ryanpetrello has quit IRC04:28
*** mc__ has quit IRC04:31
openstackgerritjustinsb proposed openstack/ceilometer: Support for Cassandra
*** yatin has joined #openstack-ceilometer04:42
*** deepthi has joined #openstack-ceilometer04:42
*** Longgeek has joined #openstack-ceilometer04:59
*** amalagon has quit IRC05:23
*** _cjones_ has joined #openstack-ceilometer05:24
*** mc__ has joined #openstack-ceilometer05:26
*** _cjones_ has quit IRC05:28
*** amalagon has joined #openstack-ceilometer05:44
*** ishant has joined #openstack-ceilometer05:55
*** exploreshaifali has joined #openstack-ceilometer06:02
openstackgerritOpenStack Proposal Bot proposed openstack/ceilometer: Imported Translations from Transifex
openstackgerritKai Qiang Wu proposed openstack/ceilometer: Initializing a longer resource id in DB2 nosql backend
*** _nadya_ has joined #openstack-ceilometer06:40
mc__Hey, I would like to know how ceilometer count memory / memory.usage. Does it include virtual and swap memory inside this VM or just count available memory part?06:45
openstackgerritjustinsb proposed openstack/ceilometer: Support for Cassandra
*** ssakhamuri has quit IRC06:56
*** ildikov has quit IRC06:56
*** yatin has quit IRC07:01
*** k4n0 has joined #openstack-ceilometer07:08
*** exploreshaifali has quit IRC07:22
*** ssakhamuri has joined #openstack-ceilometer07:36
*** Longgeek has quit IRC07:37
*** Longgeek has joined #openstack-ceilometer07:38
*** ildikov has joined #openstack-ceilometer07:46
*** nosnos has quit IRC08:00
*** exploreshaifali has joined #openstack-ceilometer08:00
*** nosnos has joined #openstack-ceilometer08:03
*** ala_ has joined #openstack-ceilometer08:10
*** _nadya_ has quit IRC08:17
*** llu has quit IRC08:18
*** llu has joined #openstack-ceilometer08:18
*** ifarkas has joined #openstack-ceilometer08:18
eglynnmc__: we report both memory and memory.usage meters08:36
eglynnmc__: "memory" is the total RAM allocated to the VM08:36
eglynnmc__: "memory.usage" is the difference between the available and unused numbers in the stats reported by libvirt (assuming that's the inspector you're using)08:37
eglynnmc__: not the caveat in on the libvirt/qemu versions required08:38
eglynn*note the caveat08:38
openstackgerritJulien Danjou proposed stackforge/gnocchi: carbonara: add full option to fetch()
openstackgerritJulien Danjou proposed stackforge/gnocchi: carbonara: fix the default fetch() behavior
openstackgerritJulien Danjou proposed stackforge/gnocchi: carbonara: fix archive back window
openstackgerritJulien Danjou proposed stackforge/gnocchi: storage: do not include to_timestamp in the range
openstackgerritJulien Danjou proposed stackforge/gnocchi: storage: expose full argument in get_measure()
*** ishant has quit IRC08:50
openstackgerritMerged stackforge/gnocchi: Remove assertEqual when request method has params for it
*** ishant has joined #openstack-ceilometer08:51
*** gordc has joined #openstack-ceilometer08:55
*** _nadya_ has joined #openstack-ceilometer08:57
*** safchain has joined #openstack-ceilometer08:58
*** mc__ has quit IRC09:13
*** _nadya_ has quit IRC09:15
openstackgerritMerged openstack/ceilometer: Add encoding to keys in compute_signature
*** exploreshaifali has quit IRC09:20
deepthieglynn, On similar lines, what about memory usage statistics of the physical host ? This is not reported currently, so any plans for the same ?09:24
eglynndeepthi: you could use SNMP polling to gather per-host data09:25
eglynndeepthi: however the problem currently with SNMP polling in ceilometer is the discovery of the SNMP daemons to poll09:26
eglynndeepthi: these can be hardcoded into the pipeline.yaml resources attribute09:26
eglynndeepthi: but that's not reasonable for anything but a very small and static deployment09:26
deepthieglynn, aha09:26
eglynndeepthi: in the special case of tripleO deployed hosts, we support auto-discovery via the nova API09:27
eglynnlsmola: ^^^ is there docco for that discovery extension?09:27
deepthieglynn, thank you so much for the info. Shall try to explore the same via SNMP agents.09:28
eglynndeepthi: np!09:28
lsmolaeglynn: not that I know of, would be worth to create one09:29
lsmolaeglynn: :-)09:29
lsmoladeepthi: this is the current list of SNMP stats we are collecting
lsmoladeepthi: it's kind of hardcoded here now09:32
lsmoladeepthi: but we should extract it to yaml conf, so you can define it per deployment, that should happen in K09:33
openstackgerritJulien Danjou proposed stackforge/gnocchi: carbonara: add full option to fetch()
openstackgerritJulien Danjou proposed stackforge/gnocchi: carbonara: fix the default fetch() behavior
openstackgerritJulien Danjou proposed stackforge/gnocchi: carbonara: fix archive back window
openstackgerritJulien Danjou proposed stackforge/gnocchi: storage: do not include to_timestamp in the range
openstackgerritJulien Danjou proposed stackforge/gnocchi: storage: expose full argument in get_measure()
deepthilsmola, I see all the memory stats I require, so tweaking the yaml file to enable collecting the same should resolve my issue.. right ?09:35
lsmoladeepthi: for TripleO I am using this pipeline
lsmoladeepthi: you can define there what you want to collect, we collect all with "hardware.*"09:36
lsmoladeepthi: it's also example of how to use the discovery and the new arithmetic transformers09:37
deepthilsmola, aha nice. Thanks for the pointers. Will use this as a reference and try out the same :-)09:37
openstackgerritgordon chung proposed openstack/python-ceilometerclient: sync with oslo and use oslo.i18n
*** Yanyanhu has quit IRC09:52
*** nellysmitt has joined #openstack-ceilometer09:54
*** nellysmitt has quit IRC09:59
*** alexpilotti has joined #openstack-ceilometer10:01
*** cdent has joined #openstack-ceilometer10:03
openstackgerritRomain Soufflet proposed stackforge/gnocchi: Add get_entity method in indexer
*** Guest32556 is now known as d0ugal10:14
*** d0ugal is now known as Guest9245510:14
*** Guest92455 is now known as d0ugal10:16
*** d0ugal has quit IRC10:16
*** d0ugal has joined #openstack-ceilometer10:16
*** ssakhamuri has quit IRC10:16
*** cdent has quit IRC10:18
*** cdent has joined #openstack-ceilometer10:30
*** nosnos has quit IRC10:43
*** nosnos has joined #openstack-ceilometer10:51
*** asalkeld_ has joined #openstack-ceilometer10:52
*** IvanBerezovskiy has joined #openstack-ceilometer10:52
*** asalkeld has quit IRC10:53
*** cmyster has quit IRC11:02
*** exploreshaifali has joined #openstack-ceilometer11:07
*** cmyster has joined #openstack-ceilometer11:10
*** cmyster has joined #openstack-ceilometer11:10
openstackgerritRomain Soufflet proposed stackforge/gnocchi: Add get_entity method in indexer
*** Longgeek has quit IRC11:13
*** Longgeek has joined #openstack-ceilometer11:21
*** nellysmitt has joined #openstack-ceilometer11:26
openstackgerritJulien Danjou proposed stackforge/gnocchi: rest: validate archive policies definitions
openstackgerritJulien Danjou proposed stackforge/gnocchi: rest: allow to have infinite retention in policies
*** nellysmitt has quit IRC11:31
*** zqfan has joined #openstack-ceilometer11:39
*** ishant has quit IRC11:51
*** ishant has joined #openstack-ceilometer11:52
*** asalkeld_ has quit IRC12:03
jd__eglynn: so InfluxDB drops all data points by default after aggregation? I thought it would keep them for ever. Or is it a limitation when running the continuous query?12:06
jd__not sure I'm clear :)12:07
eglynnjd__: in influx, the rentention and the aggregation (=downsampling in their lingo) are controlled by two separate mechanisms12:08
eglynnjd__: so continuous queries drive the aggregation, but don't discard the original full-res datapoints12:08
eglynnjd__: whereas shard-space deletion controls the rentention period12:08
eglynnjd__: so generally the full-res data can be kept around for as long as you want it to be, independent of the downsampling logic12:09
jd__so it sounds to me they shouldn't have any kind of limitation12:10
jd__if they never drop any point, adding a point back in time should trigger the recomputing of the aggregates12:10
eglynnjd__: the specific influx problem I was talking about in is that once the continuous query runs, slightly "late" datapoints for the last period are ignored12:10
jd__thought I guess the choise of having a pull based system (continuous query) and not a push system disallow that12:11
eglynnjd__: yeah, exactly the recompute is never triggered currently in influx12:11
jd__ bad design :p12:11
eglynnjd__: FYI the relevant issue in the influx tracker
jd__eglynn: so "back window" wouldn't make any sense in InfluxDB, e.g. it is 0 now and in the future it would be +∞ ?12:13
eglynnjd__: ... this one also relevant
eglynnjd__: well, it could be any finite value <∞ once those issues are fixed12:14
eglynnjd__: so currently it can't be >0, as the aggregate values are never recomputed12:15
jd__what would be the point to have something < ∞ since anyway everything is kept?12:15
eglynnjd__: well, if it was set <∞, we wouldn't keep everything12:15
*** _nadya_ has joined #openstack-ceilometer12:16
eglynnjd__: instead of keeping it forever (the default in InfluxDB) we'd use the shard-space deletion on the full-res timeseries to expire points older than the (block_size * back_window) from the last period boundary12:17
jd__got it12:18
jd__I like that plan :)12:18
eglynnjd__: so one other related point ... IIUC, while influx allows you to expire these older data, it just doesn't gaurantee that the discard is done eagerly12:18
eglynnjd__: i.e. there may be a degree of fuzziness in that it wait until an entire "shard" of data can be dropped at once12:19
jd__is that true for the result of continuous query also?12:19
eglynn"shard" meaning something smaller than the normal DB sense of the word12:19
eglynnno, I think the continuous queries are designed to run eagerly12:19
eglynn(i.e. when the period switches)12:20
*** _nadya_ has quit IRC12:20
jd__ok so that wouldn't be a problem for now as we never returned data from non-continous-query12:21
gordccdent: out of curiosity, does Rally run on bare-metal machines?12:22
cdentcan you be a bit more specific about "run" gordc ?12:22
jd__does it have legs?12:22
cdentWhen I've used it is with a bare-metal hosted devstack12:23
gordc... multiple legs.12:23
*** cmyster has quit IRC12:23
gordccdent: if we gate on it, i assume it's being benchmarked against some devstack on a vm?12:23
gordccdent: i don't know much about rally tbh.12:24
eglynngordc: as a CI job, it runs on normal CI virt nodes IIUC12:24
eglynngordc: but boris-42 also runs periodic jobs on dedicated baremetal IIUC12:24
cdentgordc: are you concerned about the unpredictable latency that vms intorduce?12:24
*** exploreshaifali has quit IRC12:24
gordceglynn: i see... can you really effectively performance test against something on a virt node12:25
gordccdent: exactly.12:25
eglynngordc: yeah, exactly ... see discussion here
cdentIf you look at my rally spec gordc there's a fair bit of discussion about that12:25
gordccdent: eglynn: cool cool. as you can tell, i read the title and that's as far as i got. :)12:25
cdentto be honest/fair/whatever my enthusiasm for having the gate job waned quite a bit as a result of that discussion12:26
eglynngordc: yeah, so the key problems are the noisy neighbors and the perf penalty seen on RAX versus HP test clouds12:27
cdentlargely because it revealed that the rally team is planning to keep historical data on all "integrated" projects themselves and that stuff is more interesting to me12:27
eglynn(or HP versus RAX, can't remember which was the slower)12:27
cdentthe per patch data is perhaps useful for flagging up "oh shit!"s12:27
gordccdent: yeah, i'd assume that gate would eat dirt once we near any milestone12:27
eglynncdent: yeah, exactly the point I was getting at with the surfacing the trend-line comments12:28
*** ishant has quit IRC12:28
eglynngordc: yeap, the proposal to only fail on p99 or p99.9 means it would need to be a catostrophic "oh shit" moment12:29
eglynn... e.g. sleep(3600) inadvertantly added to record_metering_data ;)12:29
cdentI reckon we have three options:12:30
eglynnany "normal" perf regression is unlikely to top out over p99.9 (with all the outliers in the mix anyway)12:30
cdentgo with the plan as spec12:30
cdentgo with spec - voting/sla12:30
cdentdrop spec12:30
cdents/-/minus/ # for clarity12:30
eglynn"go with spec - voting/sla" seems like a decent plan to me, as I doubt the worth of the pass/fail criteria12:31
*** deepthi has quit IRC12:31
*** yatin has joined #openstack-ceilometer12:31
gordccdent: yeah, i'm not sure how much value this is going to be honestly but i guess we can give it a try and drop it if we find it's giving us nonsense.12:32
*** yatin has quit IRC12:32
gordci've no idea what voting/sla means.lol12:32
cdentI'm struggling to recall the who/what was the inspiration for making the spec. I feel like someone asked me to, but dunno12:32
cdentgordc: if you get deeper in the spec it might make sense12:33
*** nosnos has quit IRC12:33
*** ishant has joined #openstack-ceilometer12:33
*** cmyster has joined #openstack-ceilometer12:34
*** cmyster has joined #openstack-ceilometer12:34
* gordc reading comments12:34
*** yatin has joined #openstack-ceilometer12:37
cdenteglynn: has anyone ever said you're passive aggressive?12:43
gordccdent: i assume this is somewhere in the spec, but i take it tests only against sql backends?12:43
* cdent points at comment on review 13347612:43
eglynncdent: huh?12:43
cdenteglynn: I'm half kidding, but if you are, it's in the good way12:44
gordclol passive aggressive... in the good way12:44
cdentgordc: that's not stated and I guess that's a good question12:45
cdentprobably needs to be clarified somehow12:45
cdentit doesn't _have_ to be just sql12:45
eglynngordc: well both mongo and sqla would be useful12:45
gordceglynn: agreed. wasn't sure if i missed it in spec12:46
*** _nadya_ has joined #openstack-ceilometer12:47
eglynngordc: yeah, I don't remember that point being called out explicitly TBH12:49
cdentit's not mentioned, should be12:49
cdentplease leave a comment and when I do the revision to wipe out the sla and I'll add in some words about it12:49
cdentI'm not sure how it would be done (yet).12:50
cdentThat's one of the annoyances of this spec process: its seem to operate against iteration driven discovery, so much of it is mental12:50
gordccdent: added backend item... and another comment about changes by infra12:51
eglynnthe idea IIUC was to swing the pendulum back from kicking the can down the road on all the gnarly issues12:53
cdent"i just added this comment randomly somehwere"12:54
eglynn... instead promote more upfront consideration/discussion12:54
cdentthat assumes we can see the gnarly issues12:54
*** exploreshaifali has joined #openstack-ceilometer12:54
cdentI'm not opposed to the spec process directly.12:55
eglynnyep, in this case I think we could see some at least, as per the noisy neighbor discussion12:55
cdentI do think, however, that encourages people to bite off too much in one spec.12:55
cdentAnd not necessarily the author doing that.12:55
eglynnbut in other yeah, the doing reveals a lot more12:55
eglynnbut in other *cases yeah, the doing reveals a lot more12:55
cdentWhat seems to happen is: spec gets written and someone else sees it and is like "Oh, you're in there, can you also do X?"12:55
eglynne.g. that was the case certainly for the central agent scale-out12:55
gordccdent: it's  (slightly) better than before... where we just had bps in launchpad purgatory.12:56
* cdent wants more small more often12:56
cdentgordc: yeah, that sounds like it was probably horrible12:56
gordccdent: tried to clean up launchpad bps recently... i've no idea what the hell is active, clean, abandoned.12:57
cdentlaunchpad is chaos12:57
eglynnafter the experience with one cycle of the specs process, I think we need to use our collective judgement a bit more12:57
eglynn(as discussed in the weekly meeting yesterday)12:57
openstackgerritjustinsb proposed openstack/ceilometer: Support for Cassandra
eglynni.e. judge what needs a fully detailed spec review and what doesn't12:58
cdentI think we need to put less in ceilometer. For example that ^ (cassandra). Why is that not an external optional package?12:59
eglynnyes, we've discussed that very point in relation to 3rd party CI of non-core things that can't be CI'd easily upstream13:00
eglynne.g. VMware inspector, IPMI Intel node manager stuff13:00
cdentThen again I think nearly everything should be an externally packaged plugin13:00
* cdent nods13:00
eglynnmove out of tree, and get maintainers to put arms-length CI in place13:01
cdentThat would be _very_ good for ensuring and enhancing boundaries13:01
cdentand good boundaries make for good testability13:01
*** russellb is now known as rustlebee13:02
eglynn.... it's casual-nick Friday :)13:03
* eglynn needs a more casual nick13:04
* eglynn searches for clues to self-knowledge in
cdentA.t some point soon my lovely wife will return and I'll have to take the car for its mot13:05
cdentYou do understand which comment I'm talking about, don't you?13:05
eglynn"... dishonest, backstabbing saboteurs behind the scenes" LOL :)13:05
eglynnyeah I think so13:05
eglynnsmall, self-contained patch ... or?13:06
cdentpassive aggressive is probably the wrong term13:06
cdent"should be non-controversial"13:06
cdentcut off objections at the pass13:06
*** jmatthew_ has joined #openstack-ceilometer13:07
eglynnyeah, the repetition of same ... lazy copy'n'paste on my part13:07
openstackgerritjustinsb proposed openstack/ceilometer: Support for Cassandra
*** ishant has quit IRC13:08
eglynnright-o ... mahlzeit13:09
* cdent clones packstack13:09
*** eglynn is now known as eglynn-lunch13:10
*** ryanpetrello has joined #openstack-ceilometer13:20
*** zqfan has quit IRC13:22
*** yatin has quit IRC13:25
*** Longgeek has quit IRC13:28
*** ryanpetrello has quit IRC13:30
*** _nadya_ has quit IRC13:31
*** ryanpetrello has joined #openstack-ceilometer13:33
*** ildikov has quit IRC13:36
*** zqfan has joined #openstack-ceilometer13:41
*** jmatthew_ has quit IRC13:44
openstackgerritZhiQiang Fan proposed openstack/python-ceilometerclient: Fix timeout argument not treated as integer
eglynn-lunchjd__: time for a (hopefully) quick/simple carbonara question?14:30
eglynn-lunchjd__: ... just checking my understanding re. the purpose of the various TimeSerie classes14:30
eglynn-lunchBoundTimeSeries: the full-resolution buffer of incoming raw datapoints (one per entity)14:30
eglynn-lunchAggregatedTimeSeries: downsampled timeseries for a single granularity/aggregation-function pair supported by an entity14:31
eglynn-lunchTimeSerieArchive: collection of AggregatedTimeSeries for all granularities supported an entity (for a particular aggregation function)14:31
eglynn-lunchjd__: am I on-track or off-piste? ^^^14:32
*** eglynn-lunch is now known as eglynn14:32
*** _nadya_ has joined #openstack-ceilometer14:34
*** alexpilotti has quit IRC14:39
*** alexpilotti has joined #openstack-ceilometer14:40
eglynnfolks, last call for for anyone who hasn't already voted and is interested in learning from the pasta-maker14:43
eglynn_nadya_: ^^^ poll on gnocchi bootstrap hangout if you're interested14:43
_nadya_eglynn: ok! thanks14:45
*** ryanpetrello has quit IRC14:49
openstackgerritSrinivas Sakhamuri proposed openstack/ceilometer: Internal error with period overflow
openstackgerritJulien Danjou proposed openstack/ceilometer: Switch to oslo.concurrency
eglynnjd__: looks like 1500UTC Monday is our winner for the gnocchi hangout14:56
*** dnalezyt has quit IRC14:58
*** ddieterly has joined #openstack-ceilometer15:03
*** ddieterly has quit IRC15:05
*** ddieterly has joined #openstack-ceilometer15:05
*** ddieterly has quit IRC15:06
*** ddieterly has joined #openstack-ceilometer15:06
*** rbak has joined #openstack-ceilometer15:07
*** fnaval has quit IRC15:09
*** ryanpetrello has joined #openstack-ceilometer15:12
jd__eglynn: ok let me know once you close the poll15:15
jd__eglynn: sorry missed your question, so:15:16
jd__BoundTimeSeries: the full-resolution buffer of incoming raw datapoints (one per entity) -> yes + it's bound to a certain number of block (back window etc) so the size is limited15:16
jd__TimeSerieArchive: yes + it has a BoundTimeSeries to feed the AggregatedTimeSeries with raw data15:17
*** ryanpetrello has quit IRC15:17
openstackgerritJulien Danjou proposed openstack/ceilometer: Switch to oslo.concurrency
eglynnjd__: cool, thanks!15:19
*** _nadya_ has quit IRC15:19
eglynnjd__: ... reason I doubted my understanding was the use of resample in BoundTimeSeries._first_block_timestamp() in the back window patch15:19
openstackgerritJulien Danjou proposed openstack/ceilometer: Remove module not really used by Ceilometer
jd__eglynn: ok, I hope it's clearer :)15:20
jd__my Gnocchi patches start to pile up15:21
eglynnjd__: yeah, so I'm in review mode for the rest of today, hoping to get thru' them all15:21
jd__thanks eglynn15:22
eglynnjd__: TBH I'm still a bit flumoxed by the BoundTimeSeries being a full-res buffer, yet resampling with the default how='mean' aggregation function15:22
*** _nadya_ has joined #openstack-ceilometer15:22
eglynnjd__: I feel I'm missing something really obvious on that point15:22
jd__eglynn: it doesn't store the resampling15:22
*** _nadya_ has quit IRC15:23
jd__eglynn: it's just used to have the boundaries timestamp of the aggregation15:23
jd__it's dropped on the floor then15:23
eglynnyeap, I was missing something really obvious :)15:23
jd__I guess that was that? :)15:23
eglynnyeap :)15:23
*** _nadya_ has joined #openstack-ceilometer15:24
eglynnjd__: BTW just closed that poll (looks like every who's likely to vote has done so)15:24
jd__eglynn: ok I'm gonna create an Hangout event15:24
eglynnjd__: coolness, thank you sir!15:25
*** cdent has quit IRC15:31
eglynnjd__: thanks!15:31
*** fnaval has joined #openstack-ceilometer15:36
*** fnaval_ has joined #openstack-ceilometer15:37
*** fnaval has quit IRC15:37
openstackgerritMerged stackforge/gnocchi: carbonara: fix archive back window
openstackgerritMerged stackforge/gnocchi: carbonara: fix the default fetch() behavior
idegtiarovHi! Folks what do you think about patch it is seems to be reviewed enough, and could be landed?15:43
eglynnamalagon: can you respond to jd__'s last comment in ?15:43
amalagoneglynn: responding now :)15:43
eglynnamalagon: i.e. "If that's not needed I can also drop this patch entirely"15:43
eglynnamalagon: cool, thanks!15:44
eglynnidegtiarov: looking now15:44
idegtiaroveglynn: Thank you sir!15:44
eglynnidegtiarov: looks good, other than a small test coverage suggestion and doc update15:58
*** cmyster has quit IRC15:58
*** zqfan has quit IRC15:58
idegtiaroveglynn: will improve, thank you for fast review15:59
eglynnidegtiarov: thank you sir!15:59
idegtiaroveglynn: :)15:59
eglynnamalagon: thanks for the update on gerrit, agree that basing computations on that full-res buffer is problematic16:02
eglynnamalagon: ... since the quantum of data available totally depends on the current time, and where that falls within the coarsest period for the entity16:03
amalagoneglynn: yeah :/16:03
amalagoneglynn: I totally misunderstood the buffer window though at first16:03
*** packet has joined #openstack-ceilometer16:03
eglynnamalagon: good to explore the idea in any case, though16:03
amalagonthought it was anchored on the last timepoint16:03
amalagoneglynn: I'll go back to aggregating on finest granularity data16:03
amalagonand hopefully the 0sec granularity option in archive policy will become available16:04
eglynnamalagon: cool, that makes sense16:05
amalagoneglynn: stupid question alert: I was wondering if it made more sense to load the plugins with something other than stevedore16:06
eglynnamalagon: is stevedore providing problematic for you?16:06
eglynnamalagon: e.g. the entry_points issue you saw earlier in the week16:06
amalagoneglynn: no, that's working now, but it seems like extra work to me that each custom aggregate function has to be specified in the setup.cfg file16:07
eglynnamalagon: since we already use stevedore to load the storage drivers, seems logical to follow that model for custom aggregations also16:07
amalagoneglynn: and I was reading that using abc, one can find out all concrete implementations of the abstract CustomAggregator16:08
dhellmannamalagon: discovering subclasses only works if the code is already imported, so you can't use that to find modules you need to import16:08
dhellmannit works by looking at the __subclasses__ attribute of the class16:09
eglynnamalagon: also I'm not sure that users would be writing their own custom aggregators (i.e. would need to be modifying the setup.cfg and rerunning
amalagoneglynn: that's true too16:10
dhellmannif they have their own aggregators, they would go in a separate package anyway16:10
amalagondhellmann: good point16:10
eglynndhellmann: so a completely separate entry_points.txt to drive the loading?16:10
amalagonalright, I'm convinced - sticking with stevedore16:11
dhellmanneglynn: right -- one key benefit of entry points is that they use a different namespace from the python package namespace, so the code can live somewhere else and still be discoverable16:11
eglynndhellmann: a-ha, coolness16:11
dhellmannin practice we tend to use the same string for both packages and ep namespaces, but that's a convention not a rule16:12
openstackgerritSrinivas Sakhamuri proposed openstack/ceilometer: Internal error with period overflow
*** Longgeek has joined #openstack-ceilometer16:25
gordcjd__: does watch_join_group only for zookeeper in tooz?16:27
jd__gordc: memcached and redis supports it too16:27
gordcjd__: guess i'm doing it wrong then.16:28
jd__gordc: got some code?16:28
*** ala_ has quit IRC16:29
gordcjd__: one sec. i'll copy/paste it somewhere.16:29
gordcit reuses the current coordination code in ceilometer. seems like the callback never gets triggered16:30
gordcalthough, based on subset extracted by second worker, it does seem to know that there are multiple agents16:31
jd__gordc: do you have another node joining the group?16:31
jd__gordc: you  need to call run_watchers() once in a while16:32
*** Longgeek has quit IRC16:32
jd__gordc: that's the one really calling the callback functions16:32
gordcjd__: ok. i'll take a look at run_watchers16:32
gordci have two agents which should be in same group since i'm using static group_id16:32
eglynngordc: would cdent's inspector tool be any use to you for checking if the agents are members of the expected group?16:34
eglynnDinaBelova: \o/16:36
eglynn"Nelly Kuznetsova (nellysmitt), Voronezh, Russia / Lier, Belgium - Ceilometer - Dina Belova"16:36
*** _cjones_ has joined #openstack-ceilometer16:36
eglynn^^^ presumably to be working on a ceph driver for gnocchi16:36
gordceglynn: good suggestion. i'm pretty sure it's in same group since the second agent does pick up a smaller set than the first.16:36
gordcjd__: cool. it seems to pick up join_group when i add run_watchers... doesn't see leave_group... i'll look into how it cleans up group16:41
jd__gordc: it can misses event if join/leave is done between 2 runs of run_watchers()16:42
jd__gordc: basically it just does a diff between the group state between each run_watchers() call for memcached or redis16:42
gordcjd__: i have run_watchers sync'd with heartbeat so it's every second16:44
gordcjd__: yeah, way more members register in tooz than real agents running. i'm assuming i'm not cleaning it up properly.16:46
openstackgerritJulien Danjou proposed stackforge/gnocchi: Remove custom 204 response code setting
openstackgerritJulien Danjou proposed openstack/ceilometer: Remove module not really used by Ceilometer
openstackgerritJulien Danjou proposed stackforge/gnocchi: storage: do not include to_timestamp in the range
*** exploreshaifali has quit IRC16:54
*** _nadya_ has quit IRC16:54
*** ifarkas has quit IRC16:56
gordcjd__: cool leave_group missing on stop. /me pats myself on the back.16:58
jd__gordc: or it sounds like it should be implicittely done by tooz16:59
gordcjd__: yeah maybe... it's definitely triggering callback when it cleans group of dead members.17:01
gordcdefinitely not*17:01
jd__which backend are you using?17:02
jd__gordc: would you have enough bravey to write a unit test in tooz at least?17:02
jd__should be pretty straight forward, I could fix it then17:02
gordcjd__: now or monday when i'm back in montreal?17:02
gordci have different answers depending on when. :)17:03
jd__gordc: whatever suits you, or at least report a bug or something :)17:03
jd__just don't forget about it17:03
gordcok. i'll open a bug for now17:03
gordci'm assuming that's the right project17:04
openstackgerritMerged stackforge/gnocchi: rest: validate archive policies definitions
jd__gordc: sure17:06
*** david-lyle has joined #openstack-ceilometer17:10
*** ryanpetrello has joined #openstack-ceilometer17:15
eglynnamalagon: so IIUC should do away with the need to subtract 1msec off the right-side timestamp in your MA calculation?17:18
amalagoneglynn: hmm I hadn't thought about that..17:19
amalagoneglynn: but my feeling is that I would still need to subtract off the msec? Because although that makes the retrieved measure non-inclusive of the stop, I still am slicing over small windows within the retrieved measure and have to make those small windows non-inclusive of the right endpoint17:21
eglynnamalagon: actually yep, you're right, ignore me ... I'm talking rubbish again :)17:21
*** IvanBerezovskiy has left #openstack-ceilometer17:22
amalagoneglynn: ha no17:22
*** _nadya_ has joined #openstack-ceilometer17:24
eglynnamalagon: since this was abandoned
eglynnamalagon: presumably there's another patch on the way to partner up with ?17:25
amalagoneglynn: yep, absolutely17:25
eglynnamalagon: coolness :)17:26
amalagonI am rebasing at the moment17:26
*** _nadya_ has quit IRC17:26
*** _nadya_ has joined #openstack-ceilometer17:27
*** _cjones_ has quit IRC17:29
*** _cjones_ has joined #openstack-ceilometer17:29
openstackgerritMerged stackforge/gnocchi: storage: do not include to_timestamp in the range
openstackgerritMerged stackforge/gnocchi: Remove custom 204 response code setting
*** _nadya_ has quit IRC17:34
*** harlowja has joined #openstack-ceilometer17:35
*** prad has joined #openstack-ceilometer17:37
*** harlowja has quit IRC17:37
*** harlowja has joined #openstack-ceilometer17:37
*** rwsu has joined #openstack-ceilometer17:38
*** _nadya_ has joined #openstack-ceilometer17:49
*** _cjones_ has quit IRC17:52
openstackgerritMerged stackforge/gnocchi: storage: multi-thread add_measure in Carbonara based drivers
*** _cjones_ has joined #openstack-ceilometer17:55
*** ryanpetrello has quit IRC17:59
openstackgerritMerged stackforge/gnocchi: Allow to filter resources on NULL values
*** ryanpetrello has joined #openstack-ceilometer18:06
*** _nadya_ has quit IRC18:07
openstackgerritAna Malagon proposed stackforge/gnocchi: extension for moving aggregates
*** amalagon has quit IRC18:15
*** gordc has quit IRC18:34
*** safchain has quit IRC18:36
openstackgerritEoghan Glynn proposed stackforge/gnocchi: Minor readability imporvements to carbonara
*** alexpilotti has quit IRC18:46
*** ryanpetrello has quit IRC18:47
*** amalagon has joined #openstack-ceilometer18:51
*** ryanpetrello has joined #openstack-ceilometer18:52
*** harlowja is now known as harlowja_away18:58
*** nellysmitt has joined #openstack-ceilometer19:10
*** harlowja_away is now known as harlowja19:11
*** nellysmitt has quit IRC19:14
*** ryanpetrello has quit IRC19:17
*** ryanpetrello has joined #openstack-ceilometer19:22
*** joshw42 has quit IRC19:34
*** exploreshaifali has joined #openstack-ceilometer19:36
*** prad has quit IRC19:39
*** imormo has joined #openstack-ceilometer19:45
*** amalagon has quit IRC19:50
*** ddieterly has quit IRC19:51
*** ddieterly has joined #openstack-ceilometer19:52
*** packet has quit IRC20:09
*** eglynn has quit IRC20:10
*** nellysmitt has joined #openstack-ceilometer20:12
*** _cjones_ has quit IRC20:15
*** ildikov has joined #openstack-ceilometer20:16
*** ryanpetrello has quit IRC20:18
*** amalagon has joined #openstack-ceilometer20:21
openstackgerritAna Malagon proposed stackforge/gnocchi: extension for moving aggregates
*** eglynn has joined #openstack-ceilometer20:28
*** promulo has joined #openstack-ceilometer20:31
*** promulo__ has quit IRC20:32
*** promulo has quit IRC20:35
*** promulo has joined #openstack-ceilometer20:35
*** _nadya_ has joined #openstack-ceilometer20:35
*** promulo has quit IRC20:42
*** imormo has quit IRC20:53
*** _nadya_ has quit IRC20:55
*** _cjones_ has joined #openstack-ceilometer21:00
*** ryanpetrello has joined #openstack-ceilometer21:00
*** rustlebee is now known as russellb21:03
*** ildikov has quit IRC21:08
*** alexpilotti has joined #openstack-ceilometer21:19
*** ryanpetrello has quit IRC21:22
*** ryanpetrello has joined #openstack-ceilometer21:25
*** ryanpetrello has quit IRC21:42
*** bklei has joined #openstack-ceilometer22:05
*** bklei has quit IRC22:09
*** bklei has joined #openstack-ceilometer22:10
openstackgerritAna Malagon proposed stackforge/gnocchi: extension for moving aggregates
*** bklei has quit IRC22:38
*** bklei has joined #openstack-ceilometer22:39
*** amalagon has quit IRC22:42
*** Viswanath has joined #openstack-ceilometer22:42
*** bklei has quit IRC22:44
*** Viswanath has quit IRC22:45
*** exploreshaifali has quit IRC22:59
*** _cjones_ has quit IRC23:09
*** _cjones_ has joined #openstack-ceilometer23:09
*** eglynn has quit IRC23:14
*** _nadya_ has joined #openstack-ceilometer23:56

Generated by 2.14.0 by Marius Gedminas - find it at!