| *** raildo is now known as raildo-afk | 00:33 | |
| *** sdague has quit IRC | 00:59 | |
| *** ebalduf has quit IRC | 01:38 | |
| *** ninag has quit IRC | 02:19 | |
| *** ebalduf has joined #openstack-meeting-cp | 02:53 | |
| *** persia has quit IRC | 03:40 | |
| *** persia has joined #openstack-meeting-cp | 03:50 | |
| *** _amrith_ is now known as amrith | 03:58 | |
| *** sheel has joined #openstack-meeting-cp | 04:09 | |
| *** amrith is now known as _amrith_ | 04:38 | |
| *** rakhmerov has joined #openstack-meeting-cp | 07:01 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 07:14 | |
| *** ebalduf has quit IRC | 07:20 | |
| *** Rockyg has quit IRC | 07:24 | |
| *** vilobhmm11 has quit IRC | 08:02 | |
| *** sheel has quit IRC | 08:55 | |
| *** vgridnev has joined #openstack-meeting-cp | 09:31 | |
| *** sdague has joined #openstack-meeting-cp | 10:08 | |
| *** markvoelker has joined #openstack-meeting-cp | 10:37 | |
| *** markvoelker has quit IRC | 10:42 | |
| *** markvoelker has joined #openstack-meeting-cp | 11:38 | |
| *** markvoelker has quit IRC | 11:42 | |
| *** mageshgv has joined #openstack-meeting-cp | 11:52 | |
| *** raildo-afk is now known as raildo | 11:58 | |
| *** _amrith_ is now known as amrith | 12:07 | |
| *** mtreinish has quit IRC | 12:18 | |
| *** markvoelker has joined #openstack-meeting-cp | 12:19 | |
| *** mtreinish has joined #openstack-meeting-cp | 12:21 | |
| *** xyang1 has joined #openstack-meeting-cp | 13:30 | |
| *** openstackstatus has joined #openstack-meeting-cp | 13:37 | |
| *** ChanServ sets mode: +v openstackstatus | 13:37 | |
| -openstackstatus- NOTICE: We have recovered one of our cloud providers, but there is a huge backlog of jobs to process. Please have patience until your jobs are processed | 13:40 | |
| *** sigmavirus24_awa is now known as sigmavirus24 | 14:01 | |
| *** ebalduf has joined #openstack-meeting-cp | 14:26 | |
| *** jungleboyj has joined #openstack-meeting-cp | 14:36 | |
| *** sheel has joined #openstack-meeting-cp | 14:36 | |
| *** mageshgv has quit IRC | 15:18 | |
| *** mageshgv has joined #openstack-meeting-cp | 15:30 | |
| *** vgridnev has quit IRC | 15:38 | |
| *** Rockyg has joined #openstack-meeting-cp | 15:57 | |
| *** sdake has joined #openstack-meeting-cp | 16:14 | |
| *** mageshgv has quit IRC | 16:19 | |
| *** anteaya has quit IRC | 16:20 | |
| *** sdake_ has joined #openstack-meeting-cp | 16:25 | |
| *** sdake has quit IRC | 16:29 | |
| *** mageshgv has joined #openstack-meeting-cp | 16:35 | |
| *** mageshgv has quit IRC | 16:46 | |
| *** tyr_ has joined #openstack-meeting-cp | 16:55 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 16:55 | |
| vilobhmm11 | #startmeeting quotas-wg | 16:59 |
|---|---|---|
| openstack | Meeting started Tue Apr 19 16:59:15 2016 UTC and is due to finish in 60 minutes. The chair is vilobhmm11. Information about MeetBot at http://wiki.debian.org/MeetBot. | 16:59 |
| openstack | Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. | 16:59 |
| *** openstack changes topic to " (Meeting topic: quotas-wg)" | 16:59 | |
| openstack | The meeting name has been set to 'quotas_wg' | 16:59 |
| vilobhmm11 | hello everyone | 16:59 |
| *** mageshgv has joined #openstack-meeting-cp | 16:59 | |
| amrith | hello everyone | 17:00 |
| nikhil | o/ | 17:00 |
| nikhil | vilobhmm11: can you please highlight us on the channel, I've the courtesy reminder list added to the agenda etherpad.. thanks in advance! | 17:00 |
| vilobhmm11 | hi so have updated the agenda for today | 17:00 |
| nikhil | #link https://etherpad.openstack.org/p/quotas-wg-meeting-agenda | 17:01 |
| vilobhmm11 | sure | 17:01 |
| vilobhmm11 | nikhil, vilobhmm, DuncanT, mc_nair, ninag, amrith | 17:01 |
| vilobhmm11 | nikhil : thanks for the link | 17:01 |
| nikhil | yw! | 17:01 |
| *** Qijing has joined #openstack-meeting-cp | 17:01 | |
| raildo | hey :) | 17:01 |
| vilobhmm11 | will wait for few min | 17:02 |
| vilobhmm11 | hey raildo | 17:02 |
| amrith | hello all | 17:02 |
| nikhil | :) | 17:02 |
| vilobhmm11 | alrite lets start | 17:03 |
| DuncanT | Hi | 17:03 |
| vilobhmm11 | #topic Meetup at Summit | 17:03 |
| *** openstack changes topic to "Meetup at Summit (Meeting topic: quotas-wg)" | 17:03 | |
| vilobhmm11 | Nikhil bought up a good point last meeting or an e-mail conversation that we should meetup on Monday | 17:04 |
| vilobhmm11 | before we session on Tuesday 04/26 | 17:04 |
| nikhil | #link http://doodle.com/poll/e4nq9tc77bdv8zwr | 17:05 |
| vilobhmm11 | we can meet at room 4a reserved for brown bag | 17:05 |
| DuncanT | Seems like a good plan | 17:05 |
| nikhil | please vote! | 17:05 |
| vilobhmm11 | nikhil : thanks again ! :) | 17:06 |
| nikhil | np | 17:06 |
| amrith | sorry, am timeslicing with a phone call | 17:07 |
| vilobhmm11 | as per the timings on the doodle we won't be getting the room reserved for brown bag but we can all meet near that room and then find a common one | 17:08 |
| vilobhmm11 | otherwise it would be very hard to find each other | 17:08 |
| nikhil | vilobhmm11: yeah, true | 17:08 |
| *** Qijing_ has joined #openstack-meeting-cp | 17:09 | |
| nikhil | what's the best mode of communication here? email? | 17:09 |
| DuncanT | Email for me | 17:09 |
| amrith | email works for me, IRC won't most of the time | 17:09 |
| vilobhmm11 | nikhil : email works for me too | 17:09 |
| nikhil | cool | 17:09 |
| *** Qijing_ has left #openstack-meeting-cp | 17:09 | |
| vilobhmm11 | so please vote on the doodle | 17:10 |
| nikhil | Thanks all. will email then. | 17:10 |
| vilobhmm11 | we can gather votes and find a good time that works for everyone | 17:10 |
| vilobhmm11 | #topic Product group getting started with requirements. | 17:10 |
| *** openstack changes topic to "Product group getting started with requirements. (Meeting topic: quotas-wg)" | 17:10 | |
| nikhil | (it'd be great if harlowja can join us too given he is leading the CP session) | 17:11 |
| DuncanT | I've some substantial worries about this and scope creep | 17:11 |
| nikhil | :) | 17:11 |
| vilobhmm11 | nikhil : he will join; we discussed and i will be leading the CP session | 17:11 |
| *** vgridnev has joined #openstack-meeting-cp | 17:11 | |
| DuncanT | Things like reserving quota for heat and similar - at least for cinder, that's pointless | 17:11 |
| nikhil | vilobhmm11: ah k, thanks for the update. | 17:11 |
| vilobhmm11 | harlowja : ping | 17:12 |
| nikhil | DuncanT: reserving or enforcing ? :) | 17:12 |
| nikhil | or both | 17:12 |
| vilobhmm11 | nikhil : did request sdauge to for an update on the chair for the quota session ; should see an update soon… | 17:12 |
| nikhil | vilobhmm11: excellent, thanks. I was about to ping you offline. | 17:13 |
| DuncanT | nikhil: what's the point of a reservation if it is anything other than enforcing? | 17:13 |
| vilobhmm11 | DuncanT , nikhil : will come to reservation in a bit; its a long topic …lets cover smaller ones quickly…thanks! | 17:13 |
| nikhil | ok | 17:13 |
| vilobhmm11 | irc://irc.freenode.net:6667/#topic : Product group getting started with requirements. | 17:13 |
| vilobhmm11 | irc://irc.freenode.net:6667/#topic Product group getting started with requirements | 17:13 |
| DuncanT | vilobhmm11: This isn't reservations in the currnet code sense :-) | 17:13 |
| nikhil | but is this particular one tied to PW | 17:14 |
| vilobhmm11 | ok :) | 17:14 |
| vilobhmm11 | shamail and other have started gathering use-cases and requirement about how people use quota and do capacity management | 17:14 |
| nikhil | is there a reference document for the scope? and for worries from DuncanT? | 17:14 |
| DuncanT | We've lots of terms beign thrown around, and we've still got no glossary - each mailing list thread picks its own terminology | 17:14 |
| nikhil | a BIG +1 on "defining terms" | 17:15 |
| vilobhmm11 | this will help to strengthen the use-cases problems with current quota mgmt | 17:15 |
| DuncanT | (please excuse my typing, I've substantial damage to two fingers and so it's rather tricky to type well and fast) | 17:15 |
| nikhil | actually, that can be the additional info section in the spec and people _should_ refer to it while emailing | 17:16 |
| vilobhmm11 | coming back to nikhil and DuncanT :) | 17:16 |
| DuncanT | I'm a strong proponent of matching terms to existing usage, e.g. reservation should be the code construct we already have in three projects | 17:16 |
| vilobhmm11 | #topic No reservation | 17:16 |
| *** openstack changes topic to "No reservation (Meeting topic: quotas-wg)" | 17:16 | |
| vilobhmm11 | DuncanT : +1 | 17:16 |
| DuncanT | So first question: what do we mean by 'reservation'? | 17:17 |
| amrith | I'm curious to know how we can handle quotas without reservations; the risk is that quotas are not a constraint if you don't have reservations (I THINK, AND MAY BE WRONG) | 17:17 |
| amrith | DuncanT, if that's a quesiton to me, I described it in my mail to the ML | 17:17 |
| vilobhmm11 | having said that what it funcationaly translates to in every project and in delimiter can mean different things (if at all they are used) | 17:17 |
| * amrith shamlessly plugs email to the ML :) | 17:17 | |
| nikhil | heh | 17:17 |
| DuncanT | amrith: Got a link to your mail, please? | 17:18 |
| vilobhmm11 | amrith : first of all thanks for the ML | 17:18 |
| amrith | http://openstack.markmail.org/thread/at3phitv3oupxupi | 17:18 |
| amrith | I got one comment from Jay, didn't realize it till now | 17:18 |
| amrith | have to read it | 17:18 |
| amrith | his comments are always very insightful | 17:18 |
| vilobhmm11 | yup | 17:19 |
| vilobhmm11 | they are | 17:19 |
| vilobhmm11 | in short let me brief everyone | 17:19 |
| vilobhmm11 | on why i am saying that even going with no-reservation can be an option | 17:19 |
| nikhil | I think the advantage (and consequential disadvantages for "the other set") is that reservation is super useful if your assets/resources "DO have" a task potentially in process that is not highly elastic or distributed. | 17:19 |
| DuncanT | I strongly disagree with Jay about rate limiting not being a quota thing, for the record. I'll write a reply to him soon | 17:19 |
| amrith | vilobhmm11, I'm going to wait to hear what you say about no-reservations and hold off on reading Jay's comments | 17:20 |
| DuncanT | In the case of cinder, a reservation is just consumed quota attached to a request i nprogress. | 17:20 |
| DuncanT | in progress | 17:21 |
| nikhil | exactly DuncanT, it's the "in process" part that's significant | 17:21 |
| nikhil | because the race is outside of the DB | 17:21 |
| vilobhmm11 | DuncanT : rate limiting or rate of request mgmt can be done even by external sources like repose http://www.openrepose.org/ | 17:21 |
| nikhil | to elaborate on my point: Glance will not need reservation for storing images or number of images or members etc. I think the DB transaction semantics is enough. | 17:21 |
| DuncanT | nikhil: There shouldn't be a race | 17:21 |
| vilobhmm11 | amrith : sure…in a min going through duncanT and nikhils comments | 17:22 |
| DuncanT | nikhil: Assuming the semantics are implemented without bugs, there are no race differences between getting a reservation and getting commited quota before returning from the create request | 17:22 |
| nikhil | I see | 17:23 |
| nikhil | so, we are talking about reservation vs. quota-commit+rollback | 17:23 |
| nikhil | there's the third case | 17:23 |
| DuncanT | The only difference between a reservation and a commited quota usage is that the reservation can be cancelled if the consuming process dies (plus they are handy for some debugging, though not as handy as they could be) | 17:23 |
| nikhil | that involved simply quota-commit on success | 17:23 |
| DuncanT | You need to(Semantically) quota commit before you create the resource, or the quota is useless against DoS and similar | 17:24 |
| * amrith is having a hard time demultiplexing conversations | 17:24 | |
| nikhil | DuncanT: not necessarily | 17:24 |
| nikhil | for example: image-create call in glance | 17:25 |
| *** ericksonsantos has joined #openstack-meeting-cp | 17:25 | |
| nikhil | it's a simple record / row creation in the DB | 17:25 |
| nikhil | the only thing that we need is: | 17:25 |
| nikhil | 1. create a image , update quota table | 17:25 |
| DuncanT | Plus storage used.... | 17:25 |
| nikhil | 2. once quota is full, keep returning non availability to the users | 17:25 |
| nikhil | storage is different resource | 17:26 |
| nikhil | so I am breaking things down for sanity sake | 17:26 |
| nikhil | 1. # images | 17:26 |
| nikhil | 2. amount of data stored for all images in a cloud account | 17:26 |
| nikhil | 3. # members on an image / account (depending on usage) | 17:26 |
| nikhil | for #1, all we need quota commit on success | 17:27 |
| nikhil | #2 is image upload for which it's a "task" or potentially long running process | 17:27 |
| DuncanT | nikhil: If I can create db resources (even db records, in principle) without quota check, then in theory I can bomb the system with parallel requests | 17:27 |
| DuncanT | nikhil: Glance is a pretty weird API, but the principle stands. If I have a quota of 10, it should be impossible for me to create more than 10 'things' | 17:28 |
| DuncanT | nikhil: I should never have got back a 200 or a 202 from the 11th request | 17:28 |
| nikhil | Glance API is that way as Nova needs it so | 17:28 |
| DuncanT | nikhil: I can strongly disagree with that, but this isn't the venue -grab me over a beer | 17:29 |
| nikhil | and to your point of 10 resources, if we implement test and set for quota check in glance we don't need reservation (As the transaction can be limited to DB only) | 17:29 |
| nikhil | yeah, I don't know what you are referring to with the specifics of Glance API. If you ask me , cinder shouldn't be using glance (& use something like GLARE) in the first place but then again .. not the place for that :) | 17:30 |
| vilobhmm11 | alrite first of all DuncanT : I agree with you on the point of *timelines for reservation and quota commit* nikhil : just to elaborate more you can think of reservation and few quota related tables as *cache* tables for actual resource consumption. They are nothing more than that! | 17:31 |
| vilobhmm11 | amrith : ^^ | 17:31 |
| nikhil | vilobhmm11: exactly | 17:31 |
| vilobhmm11 | reservation and few quota tables are cache tables | 17:31 |
| nikhil | and that's a race for glance | 17:31 |
| amrith | vilobhmm11, I've totally lost the train of the conversation | 17:32 |
| vilobhmm11 | actual resource consumption happens in actual resource tables like nova.instances in nova or glance.images in glance | 17:32 |
| DuncanT | The one thing about cinder and nova reservations currently is that they can auto-uncommit if the system breaks down and the creation request is lost. That is badly done at the moment but might be a desirable property | 17:32 |
| amrith | I don't know who is replying to whom/what and I'm trying to untangle the threads | 17:32 |
| DuncanT | nikhil: Why is there a race? | 17:32 |
| vilobhmm11 | amrith : just read my last comment you will get a gist of it | 17:33 |
| amrith | also, I think we're talking about the specifics of a specific service or a set of services, rather than talking about what a generic set of requirements may be for a bunch of services. | 17:33 |
| amrith | vilobhmm11, I read it; I don't understand why we're talking about tables. | 17:33 |
| amrith | that's an implementation detail | 17:33 |
| amrith | what I want to understand is what we're going to QUOTA'ify. | 17:33 |
| amrith | is it (for example) some predetermined set of resources | 17:33 |
| amrith | or can each consumer of thelibrary define their own set of resources | 17:33 |
| vilobhmm11 | amrith : for existing project there are well defined set of resources right | 17:34 |
| DuncanT | amrith: Consumers need to define their own resource set. That's a hard requirement for cinder | 17:34 |
| amrith | DuncanT, I'll let you respond to that point that Jay makes (a static set of resources) | 17:34 |
| amrith | ok, then there's a question on the kinds of quotas we want | 17:34 |
| DuncanT | amrith: e.g. volume types (which are arbitrary in name and number) have individual quotas currently | 17:34 |
| amrith | upper bound | 17:34 |
| vilobhmm11 | DuncanT : for existing projects too …like nova has ram, cpu, mem as resources | 17:34 |
| amrith | lower bound | 17:34 |
| amrith | rate | 17:34 |
| vilobhmm11 | DuncanT : for existing projects too ? | 17:35 |
| amrith | then there's the question of what to do about resizes; should they be supported in the library or not | 17:35 |
| DuncanT | We need to support all of the quota operations used in projects today, or we won't get adopted | 17:35 |
| vilobhmm11 | amrith : let me go over jay's reply and get back | 17:35 |
| amrith | then, are these quotas to be multi-tenant or system wide | 17:36 |
| amrith | vilobhmm11, this isn't about jay's email | 17:36 |
| vilobhmm11 | amrith, DuncanT, nikhil :strongly agree on getting the scope right for the lib | 17:36 |
| amrith | I haven't read it | 17:36 |
| DuncanT | Resizes are supported in projects right now (e.g. cinder), ergo we need to wupport them | 17:36 |
| nikhil | vilobhmm11: yes, and we need to meet personally for that! :)) | 17:36 |
| vilobhmm11 | nikhil : ++ | 17:36 |
| amrith | then there's the question of how we intend dealing with the use-case where a user (say me) may have a quota of 5 instances and comes and makes two concurrent requests for 3 instances each | 17:36 |
| amrith | the first one we look at says 3 < 5, great | 17:37 |
| amrith | but since no instances have been actually created, the others says 3 < 5, great | 17:37 |
| amrith | and now we get 6 instances | 17:37 |
| amrith | and 6 > 5 | 17:37 |
| amrith | we can figure out what we want to call it; reservations, preallocations, intents, ... | 17:37 |
| DuncanT | amrith: That is already handled - creating a reservation consumes quota. The first to grab the quota gets created, the second doesn't | 17:37 |
| nikhil | vilobhmm11: I need 5 mins for open discussion | 17:38 |
| DuncanT | amrith: We can't go backwards from that sematic (we can fix the races that each project has that break those semantics) | 17:38 |
| vilobhmm11 | nikhil : sure | 17:38 |
| amrith | so DuncanT what you are saying is that this new library is constrained by what existing projects are doing? | 17:39 |
| vilobhmm11 | the timeline when the reservation is commited and when the quota commit happens are the same…also the validation applies at both places…amrith : the use case you mentioned above | 17:39 |
| DuncanT | amrith: That semantic *can* be implemented without any kind of reservation though, as long as you're on a single DB. You put the thing in the DB table in some 'creating' state (cinder and nova already do that) and count those as existing for quota purposes | 17:39 |
| amrith | DuncanT, depending on the isolation level of your database, that may or may not work | 17:40 |
| vilobhmm11 | DuncanT : yup +1 | 17:40 |
| amrith | sorry vilobhmm11, happy to sketch it out on a whiteboard and show you why not | 17:41 |
| vilobhmm11 | amrith : question for you…what benefits you think that reservation is bringing right now and we might end up loosing it if it is removed ? | 17:41 |
| vilobhmm11 | amrith : sure | 17:41 |
| amrith | I believe that the business that a service does is multi-step | 17:41 |
| amrith | and that is surely the case for something like trove | 17:41 |
| amrith | and therefore, being able to support multiple steps is essential | 17:42 |
| amrith | and having a database transaction be long lived and assuming that this will work is fraught with danger | 17:42 |
| amrith | so I'm viewing reservations as short lived database operations that help you break up the long lived database transaction | 17:42 |
| DuncanT | amrith: That is *not* what reservations (as the term is currently used in the code) is all about | 17:42 |
| amrith | ok, what does a reservation mean in the current usage in code | 17:43 |
| DuncanT | amrith: reservations are not an externally visible thing | 17:43 |
| DuncanT | amrith: They're a handle on a block of consumed quota | 17:43 |
| DuncanT | That's it | 17:43 |
| amrith | I think I'm describing the same thig | 17:43 |
| amrith | thing | 17:43 |
| amrith | not externally visible: yes | 17:43 |
| amrith | a handle on a block of consumed quota: yes | 17:44 |
| amrith | something that is 'in flight' | 17:44 |
| DuncanT | Ok, I'm just not understanding you then | 17:44 |
| amrith | to make a database instance I need an instance and a cinder volume | 17:44 |
| amrith | i make a reservation for cinder | 17:44 |
| amrith | then a reservation for an instance | 17:44 |
| DuncanT | No. | 17:44 |
| amrith | if the latter fails, I release the former | 17:44 |
| DuncanT | Definitely not. | 17:44 |
| amrith | sorry, I don't know what you mean by no; that's the use-case in Trove :) | 17:44 |
| DuncanT | There is no concept of trove 'making a reservation' for cinder | 17:45 |
| DuncanT | Currently | 17:45 |
| DuncanT | Reservations are just a construct inside cinder - no external consumer sees them | 17:45 |
| amrith | actually there is, but ... | 17:45 |
| amrith | here is what I mean | 17:45 |
| amrith | You do what you wish in cinder, that's a black box to me | 17:45 |
| DuncanT | Trove asks cinder to create a volume. That's the only operation it has | 17:45 |
| amrith | here are the limits that trove has | 17:45 |
| amrith | no user may have more than 10 db volumes | 17:46 |
| amrith | no user may have more than 10 db instances | 17:46 |
| amrith | remember that the user has a nova quota of 1000 instances | 17:46 |
| amrith | and a cinder quota of 500 volumes | 17:46 |
| amrith | trove has its own limits ON CINDER RESOURCES | 17:46 |
| amrith | that Trove has to enforce | 17:46 |
| DuncanT | MY point being, the word 'reservation' has a meaning in the code now. You're using it to mean something slightly different, and that is causing communication difficulties | 17:46 |
| amrith | sorry, I'm using it to mean the same thing | 17:47 |
| amrith | in the trove context | 17:47 |
| amrith | I need to make a reservaton for a cinder volume | 17:47 |
| amrith | and a reservation for an instnace | 17:47 |
| amrith | I don't tell cinder about it | 17:47 |
| amrith | you don't know | 17:47 |
| amrith | I make my own reservations | 17:47 |
| amrith | once I make my reservations and they all succeed, then it means the user has the quotas | 17:47 |
| amrith | then I go call cinder and ask for a volume | 17:47 |
| amrith | you could still fail the request | 17:47 |
| amrith | no guarantees | 17:47 |
| amrith | then I go ask nova | 17:48 |
| DuncanT | Gah, ok, so you're talking about trove's counting of cinder volumes? That was entirely not clear | 17:48 |
| amrith | it could succeed or fail | 17:48 |
| amrith | sorry | 17:48 |
| amrith | that was Trove's reservation for a cinder resource; not something cinder gets to know about | 17:48 |
| amrith | as I said in my email, "Such a hypothetical service may also consume resources from other services that it wishes to track, and impose limits on." | 17:48 |
| DuncanT | You could just commit all those reservations in trove, and release them later in exactly the same way as a delete - there's no need for a reservation | 17:48 |
| nikhil | vilobhmm11: my point on defining terms and the ill consequences of over use of the same word .. :) (in practice) (no offense to anyone, just pointing out example of mis communications) | 17:49 |
| DuncanT | Just consume the resource | 17:49 |
| DuncanT | BAcking out a reservation is semantically the same as deleting it | 17:49 |
| vilobhmm11 | nikhil : sure…but few terms which come over time should def be documented in glossary..point noted will take care | 17:49 |
| amrith | Sorry DuncanT how do you intend that I do that? | 17:50 |
| amrith | "just commit all those ..." | 17:50 |
| DuncanT | amrith: start transaction; count (volumes where tenant id = me); if count < quota insert into volumes table state = creating | 17:51 |
| DuncanT | Then do the same for the nova table | 17:51 |
| amrith | DuncanT, great | 17:51 |
| amrith | now let me tell you why that won't work | 17:51 |
| nikhil | vilobhmm11: yes, agreed on glossary (true!, we can't really predict the future) :) | 17:51 |
| amrith | to begin, what's the database isolation level? I assume read-committed? | 17:51 |
| amrith | or would you like more than that? | 17:51 |
| amrith | would you like repeatable read | 17:51 |
| amrith | or serializable? | 17:51 |
| amrith | take your pick, let me know | 17:52 |
| amrith | I'm assuming read committed | 17:52 |
| amrith | do we agree? | 17:52 |
| * amrith waits | 17:52 | |
| DuncanT | amrith: Honestly, I can only pick what is currently used in cinder, I've not looked the the terms for teh semantics I want for some time | 17:52 |
| vilobhmm11 | amrith : read-commited means read whatever is written last right | 17:52 |
| amrith | NO | 17:52 |
| amrith | what ever was committed last | 17:52 |
| amrith | let me explain | 17:52 |
| amrith | starting state | 17:53 |
| amrith | 0 instances | 17:53 |
| amrith | 0 volumes | 17:53 |
| amrith | quota 5 of each | 17:53 |
| vilobhmm11 | by written i mean commited | 17:53 |
| amrith | request 1: 3 instances | 17:53 |
| vilobhmm11 | ok | 17:53 |
| amrith | begin transaction request 1 | 17:53 |
| amrith | transaction 1: how many instances: answer 0 | 17:53 |
| amrith | 0 is less than 5, insert 3 in table | 17:53 |
| amrith | request 2: 3 instances please | 17:53 |
| amrith | begin transaction request 2 | 17:53 |
| amrith | transaction 2: how many instance? | 17:54 |
| amrith | answer: 0 | 17:54 |
| amrith | please give me 3 ... | 17:54 |
| amrith | insert 3 in table | 17:54 |
| amrith | transaction 1: do something about volumes | 17:54 |
| amrith | commit | 17:54 |
| amrith | tranaction 2: do something about volumes | 17:54 |
| amrith | commit | 17:54 |
| amrith | net result | 17:54 |
| amrith | 6 instances | 17:54 |
| DuncanT | amrith: It is my understanding that one of those transations should fail | 17:54 |
| amrith | not if you run read committed | 17:54 |
| DuncanT | amrith: Or at least can be constructed that way | 17:54 |
| amrith | for that you increase isolation level for the db | 17:55 |
| amrith | to repeatable read | 17:55 |
| nikhil | amrith: yes, that was the argument I was making earlier but in non-defined terms | 17:55 |
| amrith | or serializable | 17:55 |
| vilobhmm11 | amrith : yes won't compare and swap or test and set happen at the db level | 17:55 |
| amrith | then performance issues begin | 17:55 |
| DuncanT | amrith: Ok, I now understand your point. Thank you. | 17:55 |
| amrith | vilobhmm11, yes, if you have the right isolation level | 17:55 |
| DuncanT | amrith: I also need to go examine some cinder db atomics code | 17:55 |
| nikhil | I think the openstack general rule on sql-like DBs is read committed | 17:56 |
| amrith | and please read: http://dev.mysql.com/doc/refman/5.7/en/innodb-transaction-model.html | 17:56 |
| nikhil | barring nova in some cases | 17:56 |
| amrith | the default is read-committed for mysql | 17:56 |
| amrith | yes, if you run in something more, be ready for deadlocks galore | 17:56 |
| amrith | anyway, I'll reply to jay on this aspect | 17:57 |
| DuncanT | amrith: So I think our best way forward is to define a python interface and its semantics and then figure out how to implement it (and stress test it)? | 17:57 |
| nikhil | amrith: in glance, we explicitly implemented test&set for some update operations | 17:57 |
| vilobhmm11 | amrith : but in openstack context for now we are mostly concered about sql db only no *nosql* in openstack right…so sql based i.e. mysql is read-commited by default.. | 17:57 |
| nikhil | not sure how much of that is upstream (gah) | 17:57 |
| *** tyr_ has quit IRC | 17:58 | |
| DuncanT | nikhil: We have something similar, but I think they only operate on one table and use update...where in our case | 17:58 |
| vilobhmm11 | alrite guys we are running out of time | 17:58 |
| harlowja | howdy | 17:58 |
| nikhil | :) | 17:58 |
| vilobhmm11 | hello harlowja…just wrapping up | 17:58 |
| nikhil | I need a few mins pls | 17:59 |
| vilobhmm11 | #topic open discussion | 17:59 |
| *** openstack changes topic to "open discussion (Meeting topic: quotas-wg)" | 17:59 | |
| harlowja | i'm around :-P | 17:59 |
| harlowja | but on phone, ha | 17:59 |
| nikhil | thanks | 17:59 |
| vilobhmm11 | yes nikhil | 17:59 |
| amrith | nikhil gets 1 min | 17:59 |
| nikhil | :) | 17:59 |
| harlowja | ya, i'll lead, but u guys need to drive ;) | 17:59 |
| harlowja | pong | 17:59 |
| vilobhmm11 | harlowja : np | 17:59 |
| nikhil | I was just thinking.. | 17:59 |
| nikhil | glance midcycle is in boston/cambridge | 17:59 |
| nikhil | june 15-17 | 17:59 |
| nikhil | I know amrith is in that area (from twitter) ? | 17:59 |
| nikhil | would it be possible for others to join | 17:59 |
| nikhil | may be the 14th for quotas? | 18:00 |
| amrith | I will be happy to attend but 15-17 is ... I hae a board meeting on 15, I can do 16 and 17 | 18:00 |
| nikhil | seems to me that summit will be too short for long discussions | 18:00 |
| amrith | I can do 14th | 18:00 |
| amrith | but I'd prefer 16th or 17th | 18:00 |
| DuncanT | That's a long flight for me, I'm unlikely to get funding | 18:00 |
| nikhil | umm, let me ask shamail too | 18:00 |
| nikhil | I can arrange for remote attendance | 18:01 |
| nikhil | all I need is your full attention for half a day | 18:01 |
| DuncanT | remote is certainly doable | 18:01 |
| vilobhmm11 | nikhil : need to check about funding..otherwise can join via hangouts | 18:01 |
| DuncanT | It will have to be morning though - I'm at GMT+2/3 | 18:01 |
| amrith | should we move to #openstack-quota? | 18:01 |
| nikhil | something that you can block in your .cal and let mgmt know unavailability for that time | 18:01 |
| nikhil | sure | 18:01 |
| vilobhmm11 | sure | 18:02 |
| nikhil | we can discuss this at the summit | 18:02 |
| nikhil | food for thought | 18:02 |
| vilobhmm11 | lets move to #openstack-quota | 18:02 |
| nikhil | thanks all! | 18:02 |
| vilobhmm11 | thanks! | 18:02 |
| vilobhmm11 | #endmeeting | 18:02 |
| *** openstack changes topic to "OpenStack Meetings || https://wiki.openstack.org/wiki/Meetings" | 18:02 | |
| openstack | Meeting ended Tue Apr 19 18:02:15 2016 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4) | 18:02 |
| openstack | Minutes: http://eavesdrop.openstack.org/meetings/quotas_wg/2016/quotas_wg.2016-04-19-16.59.html | 18:02 |
| openstack | Minutes (text): http://eavesdrop.openstack.org/meetings/quotas_wg/2016/quotas_wg.2016-04-19-16.59.txt | 18:02 |
| openstack | Log: http://eavesdrop.openstack.org/meetings/quotas_wg/2016/quotas_wg.2016-04-19-16.59.log.html | 18:02 |
| DuncanT | Sure, I'll pencil in all three days for now | 18:02 |
| *** mtreinish has quit IRC | 18:03 | |
| *** vilobhmm11 has quit IRC | 18:04 | |
| *** Qijing has left #openstack-meeting-cp | 18:06 | |
| harlowja | lol | 18:07 |
| harlowja | vilobhmm11 i think my advice for all of this, is just start coding, don't get stuck in <architecture-perfectness> ;) | 18:08 |
| harlowja | u can end up in <architecture-perfectness> forever, lol | 18:08 |
| *** mtreinish has joined #openstack-meeting-cp | 18:08 | |
| *** sdake_ has quit IRC | 18:41 | |
| *** tyr_ has joined #openstack-meeting-cp | 18:46 | |
| *** sdake has joined #openstack-meeting-cp | 18:48 | |
| *** tyr__ has joined #openstack-meeting-cp | 18:50 | |
| *** ebalduf has quit IRC | 18:52 | |
| *** tyr_ has quit IRC | 18:53 | |
| *** amrith is now known as _amrith_ | 18:58 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 19:02 | |
| *** rocky_g has joined #openstack-meeting-cp | 19:03 | |
| *** sheel has quit IRC | 19:05 | |
| *** sigmavirus24 is now known as sigmavirus24_awa | 19:06 | |
| *** mc_nair has joined #openstack-meeting-cp | 19:27 | |
| *** mtreinish has quit IRC | 19:41 | |
| *** mtreinish has joined #openstack-meeting-cp | 19:42 | |
| *** mtreinish has quit IRC | 19:49 | |
| *** sigmavirus24_awa is now known as sigmavirus24 | 20:06 | |
| *** _amrith_ is now known as amrith | 20:07 | |
| *** jungleboyj has quit IRC | 20:36 | |
| *** diablo_rojo has joined #openstack-meeting-cp | 20:43 | |
| *** xarses has joined #openstack-meeting-cp | 20:46 | |
| *** tristanC_ is now known as tristanC | 20:54 | |
| *** dims_ has joined #openstack-meeting-cp | 21:02 | |
| *** dims has quit IRC | 21:02 | |
| *** amrith is now known as _amrith_ | 21:05 | |
| *** vilobhmm11 has quit IRC | 21:07 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 21:08 | |
| *** rocky_g has quit IRC | 21:11 | |
| *** xarses has left #openstack-meeting-cp | 21:14 | |
| *** sdake_ has joined #openstack-meeting-cp | 21:30 | |
| *** diablo_rojo has quit IRC | 21:31 | |
| *** sdake has quit IRC | 21:33 | |
| *** vilobhmm11 has quit IRC | 21:43 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 21:44 | |
| *** vilobhmm11 has quit IRC | 21:45 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 21:45 | |
| *** vilobhmm11 has quit IRC | 21:45 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 21:46 | |
| *** vilobhmm11 has quit IRC | 21:46 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 21:46 | |
| *** _amrith_ is now known as amrith | 21:50 | |
| *** vgridnev has quit IRC | 21:53 | |
| *** vilobhmm11 has quit IRC | 21:59 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 21:59 | |
| *** amrith is now known as _amrith_ | 22:00 | |
| *** xyang1 has quit IRC | 22:01 | |
| *** sigmavirus24 is now known as sigmavirus24_awa | 22:02 | |
| *** mtreinish has joined #openstack-meeting-cp | 22:16 | |
| *** sdake_ is now known as sake | 22:26 | |
| *** vilobhmm11 has quit IRC | 22:32 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 22:32 | |
| *** vilobhmm11 has quit IRC | 22:33 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 22:33 | |
| *** vilobhmm11 has quit IRC | 22:40 | |
| *** persia has quit IRC | 22:46 | |
| *** mageshgv has quit IRC | 22:54 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 22:55 | |
| *** persia has joined #openstack-meeting-cp | 23:04 | |
| *** sdague has quit IRC | 23:07 | |
| *** mageshgv has joined #openstack-meeting-cp | 23:07 | |
| *** david-lyle_ has joined #openstack-meeting-cp | 23:14 | |
| *** david-lyle has quit IRC | 23:14 | |
| *** sake is now known as sdake | 23:22 | |
| *** vilobhmm11 has quit IRC | 23:49 | |
| *** vilobhmm11 has joined #openstack-meeting-cp | 23:49 | |
| *** david-lyle_ is now known as david-lyle | 23:57 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!