Monday, 2016-02-01

*** evgenyl has quit IRC00:08
*** sylwesterB has quit IRC00:09
*** evgenyl has joined #openstack-solar00:09
*** sylwesterB has joined #openstack-solar00:10
*** evgenyl has quit IRC00:16
*** evgenyl has joined #openstack-solar00:17
*** ikalnitsky_ has joined #openstack-solar05:42
*** pigmej_ has joined #openstack-solar05:48
*** ikalnitsky has quit IRC05:49
*** pigmej has quit IRC05:49
*** pigmej_ is now known as pigmej05:49
*** ikalnitsky_ is now known as ikalnitsky05:50
*** salmon_ has joined #openstack-solar07:27
openstackgerritLukasz Oles proposed openstack/solar: Allow to use any IPs with fuel-devops
openstackgerritLukasz Oles proposed openstack/solar: Use yaml to load data in solar create
openstackgerritLukasz Oles proposed openstack/solar: Allow to use any IPs with fuel-devops
openstackgerritLukasz Oles proposed openstack/solar: Use yaml to load data in solar create
openstackgerritLukasz Oles proposed openstack/solar: Allow to use any IPs with fuel-devops
*** dshulyak has joined #openstack-solar08:26
salmon_dshulyak: o/08:28
pigmejhey guys :)08:40
openstackgerritMerged openstack/solar: Use yaml to load data in solar create
openstackgerritMerged openstack/solar: Hardcode python 2 usage
*** openstackgerrit has quit IRC10:02
*** openstackgerrit_ has joined #openstack-solar10:02
*** openstackgerrit_ has quit IRC10:03
pigmejdshulyak: jenkins voted -110:04
dshulyakpigmej: you will go to ost summit?10:06
pigmejdshulyak: I'm not aware of this ;D10:07
pigmejbtw where are mike notes ?10:08
dshulyakpigmej: i thought that was the purpose of getting visa10:08
pigmejsome say that I will visit sunnyvale in march10:08
pigmejthat's all taht I know right now10:08
pigmejdshulyak: btw, short summary about solar things in poznan is "pretty much nothing changed from solar perspective"10:09
dshulyaki see, looks like nothing was changed except that we will fetch data from config service instead of nailgun10:12
salmon_dshulyak: no, ConfigService will become config api on top of solar10:12
pigmejlike UX10:13
salmon_that's the biggest change10:13
pigmejsure, but it's kinda unrelated for whole picture10:14
dshulyakok, i will wait for meeting10:14
pigmejone thing that changed, is that we will need immediate feedback from all DPs, so we will need to execute them frequently, we also discussed how DPs will be executed etc.10:15
pigmejlet's say that the original idea is the same, but we know more details about it :)10:15
dshulyakyeah, i remember Mike was talking about it in MV10:15
pigmejsalmon_: It looks like we did nothing in poznan :D10:16
salmon_we played some foosball ;)10:17
pigmejdshulyak: we missed you a lot in poznan10:17
pigmejnot only because of foosball :P10:18
openstackLaunchpad bug 1540304 in Solar "When solar worker is not running run-once doesn't return error" [High,New]10:50
dshulyakok, but it is push/pull and you dont know if there is someone on the other side10:52
salmon_dshulyak: ok, so worker when is starting should check db for tasks10:53
dshulyakthere is no information in db10:54
salmon_how so? report is taking info from db10:55
salmon_so the only way is to run restart?10:55
dshulyakno in this sense, there is no information in db if scheduler should execut workflow10:55
dshulyakthis message should come from queue10:56
salmon_ok, so start worker, restart deploy10:56
dshulyaki think that zero should allow late binding of server10:56
dshulyaki will check that10:56
pigmejdshulyak: better not10:57
pigmejbecause then we're not in control about messages10:57
pigmejand we may execute very old stuff10:57
dshulyakyou are right, thats possible, i can also change zero sockets from push/pull to other model, and acknowledge each event before processing it10:59
dshulyakwe will see error atleast10:59
pigmejyeah that may be better10:59
dshulyaki watched video from Rich Hickey about datomic..11:06
dshulyakand was wtf..11:06
pigmejwhy ?11:07
dshulyakwhat datomic is doing? except incremental history, additional consistency guarantees on top of common db11:09
pigmejpretty much nothing more11:12
pigmejbut it can act as external coordinator, but don't worry, we will not gonna use it :)11:12
pigmejdatomic is distaibuted datalog, it adds you kinda transactions though11:13
dshulyaknah, i am not, it just seemed like an interesting piece of software from clojure author11:13
pigmejthe interesting part is closed source :)11:14
pigmejsalmon_: to  this is fixed, isn't it ?11:39
openstackLaunchpad bug 1539174 in Solar "Composer files may changes data type" [Critical,Confirmed]11:39
salmon_pigmej: nope11:39
salmon_pigmej: I just have workaround in nodes.yaml11:39
pigmejah right11:40
*** openstackgerrit has joined #openstack-solar11:53
*** openstackgerrit has quit IRC11:54
*** openstackgerrit_ has joined #openstack-solar11:54
*** openstackgerrit_ is now known as openstackgerrit11:55
*** openstackgerrit has quit IRC11:59
*** openstackgerrit has joined #openstack-solar12:07
openstackgerritJedrzej Nowak proposed openstack/solar: Removed all __init__ files from pep8 ignore
pigmejI really hate hacking + pep8 combo12:26
openstackgerritJedrzej Nowak proposed openstack/solar: Removed all __init__ files from pep8 ignore
pigmejsalmon_: fixed probably12:41
openstackgerritDmitry Shulyak proposed openstack/solar: Introduce documentation related to new worker
pigmejdshulyak: cool :)12:44
pigmejcan you guys review ?12:44
dshulyaki still need to work on it, but i want switch from documentation for now :)12:44
pigmejdshulyak: smart choice :D12:44
dshulyakbtw what about that meeting?12:48
dshulyaksalmon_: pigmej ^12:48
salmon_mihgen: tomorrow we are going to schedule a meeting with all interested guys and summarize Poznan meeting. Do you also want to attend? If yes, in which time zone are you now?12:51
pigmejsalmon_: tomorrow time table is a bit packed from what I see12:57
pigmejfrom what I see we have no common free slot for vova12:58
pigmejsounds like wednesday ~1PM polish time is free slot12:59
salmon_ok, scheduled for thursday... You are very busy guys...13:06
salmon_If I missed someone please add him13:06
pigmejbtw, then I think we should make fast introduction for dima tomorrow13:06
salmon_Pls review:
salmon_and merge in that order13:13
pigmejsalmon_: :(13:15
salmon_my was first :P13:16
pigmejyour is easier to rebase :P13:16
pigmejmy was reviewed first :P13:17
pigmejsalmon_: just check it, if it's fine, I can rebase  probably13:17
openstackgerritMerged openstack/solar: Removed ssh_ prefix from transports resources
openstackgerritLukasz Oles proposed openstack/solar: Allow to use any IPs with fuel-devops
pigmejsalmon_: can you fix + ?13:38
salmon_pigmej: done13:39
pigmejIt would be cool if we can have script that would create resoruces13:40
pigmejafter we merge this all examples and their usage will be not valid13:49
pigmejI would expect a script / python whatever that just creates nodes with default configuration13:52
salmon_you can always create nodes with: solar resource create nodes templates/nodes count=X13:52
pigmejwhere default is let's say vagrant env13:52
salmon_solar resource create nodes templates/nodes count=X13:52
pigmejdocs updates then are required :)13:53
pigmejI was pretty sure that we docummented it13:54
pigmejbut it turns out that no13:54
salmon_we just have links to the examples13:54
salmon_you may add it:P13:54
openstackgerritJedrzej Nowak proposed openstack/solar: Removed deprecated ansible handler documentation
pigmejI removed it because it's deprecated / obsolete13:58
pigmejw need new documentation about handlers
openstackLaunchpad bug 1540383 in Solar "Add documentation about handlers internals" [Undecided,New]13:58
dshulyakpigmej: have you saw such errors ?14:01
dshulyakfrom solar.orchestration.workers import tasks14:05
pigmejmaybe my recent changes about pep8 + ?14:09
dshulyakno, this is my patches14:09
dshulyakwith -1 from ci14:10
dshulyaki noticed that gevent works same way sometimes..14:10
dshulyaksometimes gevent.pool.Pool works :) and sometimes it fails with AttributeError, but i never wanted to debug it14:10
pigmejstupid import order probably then14:11
pigmejjust like ansible14:11
pigmejdshulyak: probably circural import14:12
pigmejcheck this beautifull hack :)14:12
dshulyaki see14:12
dshulyakbut i am not importing orchestration.workers at all14:14
dshulyakthats like first import of that module14:15
dshulyakoh crap14:21
dshulyakif one wants to use submodules like that - you should do imports of all submodules in __init__.py14:22
dshulyakimport solar.orchestration.workers.scheduler14:23
dshulyakimport solar.orchestration.workers.tasks14:23
dshulyakimport solar.orchestration.workers.system_log14:23
pigmejand because of pep8 / hacking you cannot `*`14:26
openstackgerritDmitry Shulyak proposed openstack/solar: Every extension should be able to subscribe to events of other ext
pigmejdshulyak: jenkins dislikes you :p14:45
dshulyaki dont really like gerrit workflow :( i think i will merge 2 patches into 114:46
pigmejI dislike it too ;P14:47
pigmejbut, you can work on nested commits14:47
pigmejbut the tests are problem14:47
dshulyakit is ok when you are doing simple change14:52
openstackgerritDmitry Shulyak proposed openstack/solar: Plugable constructors, runners and hooks
openstackgerritDmitry Shulyak proposed openstack/solar: Introduce documentation related to new worker
pigmejdshulyak: that is still wip, isn't it?15:22
pigmejbtw how do you like stevedore?15:24
openstackgerritJedrzej Nowak proposed openstack/solar: Added info that our examples require nodes
pigmejdshulyak: have you tested your zmq thingy with multiple processes and let's say tcp transport?15:30
dshulyakpigmej: i expect it to fail15:31
pigmejdshulyak: why?15:31
pigmejwe should support it15:31
pigmejotherwise zmq thingy makes small sense for us15:31
dshulyakbecause multiple servers shouldnt be able to bind to the same endpoint15:31
pigmejSO_REUSEADDR :P15:32
pigmejbut yeah, some router pattern will be needed then15:32
dshulyakem, no15:33
dshulyakit wont allow to bind 2 servers to exactly same socket15:33
pigmejsure it will not by default15:33
pigmejbut still, it should be supported by solar worker to have multiple processes15:33
pigmejotherwise what's the point of having zmq ?15:34
pigmejif the same could be archived with let's say pipe / gevent queue?15:34
dshulyakyes, but we will need router with zero15:36
pigmejI wanted to know in theory if code let's say supports it15:37
pigmejdshulyak: so let's say I want (not that I want to do it) plug some big message system behind scenes, it should be possible then ?15:37
dshulyakthere should be no problem to put zmq router between client and server, and for example with rabbitmq it is the same, server will declare all required exchange/queue options, and client will communicate the same way15:40
pigmejdshulyak: k, then the answer for me is 'yes' :)15:42
dshulyakbut we wont be able to use only soe gevent queue, because scheduler should be available for external clients15:44
dshulyakpigmej: how rollback different from update?15:55
pigmejI don't understand context :)15:56
dshulyakyou said that you found some issues with rollback15:56
dshulyaki am interested how it is different from update15:57
pigmejin poznan?15:57
dshulyaki think so :)15:57
pigmejfrom solar side it's thesame15:57
pigmejfrom bigger perspective it's not15:57
pigmejyou cannot rollback when you don't own the data15:57
dshulyakfrom nailgun side lets say15:57
dshulyakany client of solar15:58
pigmejso, if you assigned ip first, then you released it, someone other may take ip for other purpose, then you can't rollback to
pigmejwe can set it in solardb but any automatic rollback will fail15:59
pigmejand that's the point why "we cannot support rollbacks"15:59
dshulyakand this is quite easy for nailgun to decide15:59
pigmejfrom solar perspective it's easy, from bigger picture it's not15:59
dshulyakit is just update15:59
pigmejBUT would you expect that rollback in your environment affects another environment ? Even when it's totally isolated ?16:00
pigmejnot to mention that changes in the second environment may affect other envs, or other parts in first env16:01
dshulyakif 2 envs will share some network - they are not isolated, but is is actually not important16:01
dshulyakthe importatnt part is that you can validate this16:01
pigmejthe same situation with bind ports, and pretty much all "not solar" managed stuf :)16:01
pigmejyeah we can validate it. But that's all16:01
pigmejbut the result is: "we *may* support rollbacks if you're lucky"16:02
dshulyaknot lucky..16:02
pigmejkinda lucky,16:03
pigmejwe could also throw an error / message that manual invervention is needed16:03
dshulyakok, my point was that rollback is nothing special from update16:03
pigmejand "please resume when you're set"16:03
pigmejdshulyak: sure, it changes nothing :)16:03
dshulyakand saying that rollback of ips is not possible means that even update isnt possible16:03
pigmejno, because update is new fresh state16:03
pigmejlike you cannot acquire used ip16:04
pigmejbecause network manager will take care about it16:04
dshulyaksee example16:04
pigmejBUT with rollback / revert we cannot guarantee it16:04
dshulyakx = y , swap them with update16:04
dshulyaksame sequence of actions16:04
pigmejdshulyak: sure but it's good case16:05
pigmejI mean "the working one"16:05
dshulyakyes, it is both rollbackable and updatable because 1 == 2 :)16:05
dshulyakcan you show me bad one?16:06
dshulyakwith x and y :)16:06
pigmejimagine, x =, then you update network to, so you regenerated ip to (let's say), you kinda refreshed this info with solar update (or whatever), and well, you cannot rollback this change easily16:06
pigmejbecause something other may need network, and we can't just set in a place of because it will be also invalid16:06
dshulyakbut can you update x to :) ?16:07
pigmejsame happens when you change IP like x= then x =, what happens is that you acquire new ip, then release, so other service may take (it *may* be not managed by solar / fuel), and.... you have a problem16:07
pigmejdshulyak: sure, in solar and fuel sure, BUT we may be unable to set these values back to 3rd party service16:08
pigmejthere is small chance that having versioned 3rd party service storage would help there, but then we may endup with pretty complicated chained revert/rollback like thingy16:08
dshulyakfor the data managed by some 3rd party - update should be made there atfirst16:09
dshulyakif there is no history16:09
dshulyakthen ofcourse we cant call it rollback16:10
dshulyakbut update itself it possible16:10
pigmejand that's the message :)16:10
pigmejit's kinda like a rollback "if you're lucky enough" (because even rollback of IP may  succeed) BUT in general rollback will require manual intervention :)16:11
dshulyakidk, it seems pretty ok to me, if service that manages this data is reliable ofc16:11
pigmejwell, it changes pretty much nothing, because still, this service can be source not only for solar/fuel16:12
pigmejor 2 separate environments within solar/fuel16:12
pigmejif everything would be managed by solar, and all data would be in solar db then we could support very cool rollbacks, but even then not everything would be possible16:14
dshulyakbut its ok i guess, it is responsibility of that other service to react on changes correctly16:15
dshulyakalso there is not that much components in fuel right now16:15
pigmejand what about beloved cmdb ?16:15
dshulyakconfigdb you mean?16:16
dshulyakor smth else?16:16
pigmejno, cmdb16:16
pigmejexternal configuration database like thingy16:16
pigmejconfigdb is not a big deal, it seems that we can make it on top of solar api16:16
pigmejso for solar it will change nothing16:16
pigmejand for user it may be just another way to work with solar resource connections16:17
pigmejthat's also quite good solution for both words16:17
dshulyakso whats with that cmdb? i think it is the same, if they have history - they can try to do rollback, if they dont - updates, but thats like competely separate workflow16:19
pigmejdshulyak: even when it has history, it *may* be a problem, because that change will require to be somehow propagated for whole system16:20
pigmejand it maybe *not* only solar managed stuff16:20
pigmejbecause if we will assume that we have only solar, then it's super coll16:20
pigmejbut it turned out that we can't assume that16:20
pigmejdshulyak: I reviewed your changes, it works :)16:27
pigmejhmm ansible 2.0 changed a lot on API level17:15
pigmejhmm I think we should switch to use ansible as separate process (even local one), it's way easier to integrate (speed-- though) but... even 2.0 ansible cli API is compatible with 1.9.x17:34
openstackLaunchpad bug 1534145 in Solar "Make ansible handlers working with ansible 2.0" [High,Confirmed]17:57
pigmejthat's my comment salmon_ dshulyak please tell what do you think about it17:59
*** dshulyak has quit IRC19:30
salmon_pigmej: +1 from me20:13
*** nihilifer has quit IRC20:26
*** nihilifer has joined #openstack-solar20:30
openstackgerritMerged openstack/solar: Add support for configurable executor
openstackgerritMerged openstack/solar: Extensions mechanism for orchestration components
openstackgerritMerged openstack/solar: Plugable constructors, runners and hooks
*** angdraug has joined #openstack-solar21:52

Generated by 2.14.0 by Marius Gedminas - find it at!