15:00:56 <ad_rien_> #startmeeting massively-distributed-clouds
15:00:57 <openstack> Meeting started Wed Dec 20 15:00:56 2017 UTC and is due to finish in 60 minutes.  The chair is ad_rien_. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:58 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:01:00 <openstack> The meeting name has been set to 'massively_distributed_clouds'
15:01:04 <ad_rien_> #chair ad_rien_
15:01:05 <openstack> Current chairs: ad_rien_
15:01:12 <ad_rien_> #topic roll call
15:01:14 <ad_rien_> Hi guys
15:01:37 <kgiusti> o/
15:02:05 <ad_rien_> I don't know how many folks will be there today
15:02:22 <ad_rien_> rcherrueau and dpertin are giving a lecture regarding enos
15:02:27 <ad_rien_> so anyone else ?
15:02:29 <msimonin> o/
15:02:31 <msimonin> :)
15:02:35 <avankemp> Hi everyone
15:02:50 <kgiusti> ansmith is on holiday until 1/2
15:03:07 <ad_rien_> yes… he is right ;)
15:03:40 <ad_rien_> so let's wait two or three minutes,meanwhile you can give a look to the different links I put on the etherpad
15:04:02 <ad_rien_> #link https://etherpad.openstack.org/p/massively_distributed_ircmeetings_2017 Agenda line 1638
15:05:28 <ad_rien_> anyone from the dragonflow team?
15:05:51 <ad_rien_> ok
15:06:20 <ad_rien_> so let's start
15:06:26 <ad_rien_> #topic news
15:06:52 <oanson> Hi
15:06:57 <oanson> Sorry I'm late
15:07:03 <msimonin> Hi oanson
15:07:27 <ad_rien_> 1./ regarding the telcos organized by the foundation, I think we have an interesting session where we discuss some scenarios regarding possible edge use-cases
15:07:57 <ad_rien_> I believe we will consolidate this list by begining of January
15:08:17 <ad_rien_> You can give it a look (amend and complete it I guess)
15:08:25 <msimonin> which link is it ?
15:08:42 <ad_rien_> #link  https://etherpad.openstack.org/p/2017_edge_computing_working_sessions_existing_gaps
15:09:09 <ad_rien_> the other link is related to a template related to the use-case. Since parus is not there today, I propose to skip that point
15:10:21 <ad_rien_> Then there is another interesting post that has been made by Matt Riedmann
15:10:38 <ad_rien_> regarding the use of CellV2 concepts to deal with edge infrastructures
15:11:08 <ad_rien_> I would have appreciated to reply but unfortunately I cannot see the post anymore (stupid click on the report instead of forward :()
15:11:26 <ad_rien_> If someone knows how I can rollback such an action, please let me know
15:11:39 <ad_rien_> Meanwhile, could you please tell me whether you can access it?
15:12:14 <kgiusti> +1
15:12:24 <ad_rien_> thanks,
15:12:24 <msimonin> I can't but maybe other can, I'm not linkedin friendly
15:13:36 <ad_rien_> ok
15:13:49 <ad_rien_> if others can see it, then I will manage from my side sorry ;)
15:13:53 <ad_rien_> so let's go ahead
15:14:46 <ad_rien_> Last but not the least, I should send informations regarding edge sessions during the next PTG.
15:15:01 <ad_rien_> Could you please confirm that you will attend/by interested to take part to some discussions in Dublin
15:15:13 <ad_rien_> if yes, please update the etherpad
15:15:23 <ad_rien_> #link https://etherpad.openstack.org/p/2017_edge_computing_working_sessions
15:16:40 <msimonin> Is this epad related to the PTG ?
15:16:54 <ad_rien_> kgiusti:  you told me that you should be in Dublin with Andy ?
15:17:15 <kgiusti> ad_rien_: yes - we're confirmed and on that epad
15:17:18 <ad_rien_> #link https://etherpad.openstack.org/p/2017_edge_computing_working_sessions line 21, PTG attendees
15:17:40 <kgiusti> line 24 :)
15:18:43 <ad_rien_> ok
15:18:52 <ad_rien_> so that's all from my side regarding the news
15:19:34 <ad_rien_> anythings else from your side?
15:20:16 <ad_rien_> @msimonin can you please share some information regarding the performance WG?
15:20:22 <ad_rien_> before moving to the next topic?
15:21:02 <ad_rien_> msimonin: ?
15:21:05 <ad_rien_> still there :-P
15:21:07 <msimonin> Do you have somethiong in mind ?
15:21:22 <ad_rien_> are there new concrete actions ?
15:21:40 <ad_rien_> (a part from the AMQP evaluations you are doing guys)
15:21:56 <ad_rien_> for instance, around the cell V2?
15:22:07 <msimonin> Actually, well, no
15:22:10 <msimonin> :)
15:22:14 <ad_rien_> ok thanks :D
15:23:19 <ad_rien_> #topic ongoing-actions: use-cases
15:23:35 <ad_rien_> ok I think we can  skip that one
15:23:50 <ad_rien_> parus is not there and I didn't see any folks from FBK
15:24:00 <ad_rien_> Can I move to the next topic?
15:24:28 <msimonin> +1
15:24:32 <ad_rien_> #topic ongoing-actions: amqp
15:24:39 <ad_rien_> So msimonin kgiusti please
15:25:13 <kgiusti> So we've been meeting weekly to discuss the oslo.messaging testing
15:25:57 <kgiusti> Last we talked msimonin and avankemp were still testing small distributions under rabbit
15:26:46 <kgiusti> We're getting together tomorrow for an update.
15:27:01 <ad_rien_> ok thanks
15:27:11 <kgiusti> msimonin: avankemp anything I missed?
15:27:23 <msimonin> That's perfect :)
15:27:37 <avankemp> nop, maybe I can just add that we're scaling the experiments now to thousands of agents
15:27:55 <kgiusti> excellent!
15:27:56 <msimonin> +1
15:28:20 <ad_rien_> +1
15:28:38 <ad_rien_> msimonin:  you put a couple of questions on the etherpad. Do you want to go through please?
15:28:47 <msimonin> oh yes sure
15:28:58 <msimonin> especially if we have kgiusti and oanson
15:29:09 <msimonin> it's related to oslo.messaging and dragonflow
15:29:13 <kgiusti> \o/
15:29:15 <ad_rien_> please
15:29:16 <ad_rien_> :D
15:29:50 <msimonin> last time oanson told us that the pub/sub implementation of dragonflow was plugable
15:29:56 <oanson> Yes
15:30:04 <msimonin> but there was no plugin for oslo.messaging
15:30:04 <oanson> I was just reading through the question in the agenda.
15:30:08 <msimonin> ok
15:30:08 <msimonin> :)
15:30:11 <oanson> I don't know oslo.messaging too well
15:30:20 <kgiusti> I don't believe I've worked with dragonflow - is there a good website describing it?
15:30:29 <kgiusti> (save me the google fu)
15:30:36 <oanson> We have a blog, but mostly it's just us for now :)
15:30:41 <msimonin> (you should use qwant :) )
15:30:43 <oanson> Especially for technical details
15:30:49 <kgiusti> oanson: I recommend you keep it that way.
15:30:59 <kgiusti> oanson: oslo.messaging that is ;)
15:31:06 <oanson> xD
15:31:28 <kgiusti> So from the epad is the Q if an oslo.messaging plug in to dragonflow available?
15:31:45 <kgiusti> Not that I'm aware of.
15:31:47 <oanson> Currently there's no oslo.messaging driver.
15:31:52 <msimonin> yes
15:32:12 <oanson> It would be a great contribution, but we don't have the resources to do it. Especially since we have 2-3 other drivers already working well
15:32:14 <msimonin> the point is what would be missing in oslo messaging (or not)
15:32:43 <kgiusti> msimonin: in terms of dragonflow messaging requirements?
15:32:47 <oanson> I doubt there's anything missing. Dragonflow only needs a way to send a message, be notified when a message is received
15:32:51 <kgiusti> msimonin: like topic subscriptions?
15:32:51 <msimonin> I don't know if line 1665 to l1667 are clear enough for example
15:32:55 <oanson> There is also the ability to subscribe to topics
15:33:11 <msimonin> are those topics dynamics ?
15:33:17 <oanson> Yes
15:33:27 <oanson> Currently we have a global topic that every one registers to. And a topic per tenant/project
15:33:42 <msimonin> do you have a topic per sec-group for example ?
15:33:59 <oanson> Not currently.
15:34:00 <msimonin> that an agent would subscribe to if needed
15:34:16 <oanson> But our agent is an all-or-nothing
15:34:24 <msimonin> ah ?
15:34:30 <msimonin> what do you mean ?
15:34:33 <oanson> We have a single agent that does everything:
15:34:49 <oanson> The Dragonflow controller runs on each compute node.
15:35:02 <oanson> It connects to the OVS switch, and implements everything it can through that
15:35:11 <oanson> What it does is defined by the applications that were loaded
15:35:57 <oanson> So the Dragonflow controller (our single agent) subscribes (using the pub/sub mechanism) to events according to specific tenants/projects
15:36:28 <oanson> If selective proactive is enabled, only the tenants installed on it (i.e. ports bound to this node). Otherwise, it subscribes to everything.
15:36:39 <oanson> msimonin, this was a bit long. Does it answer your question?
15:37:01 <msimonin> If a sec-group is modified, are all the agents notified ?
15:37:15 <oanson> In general, yes
15:37:34 <msimonin> even if they aren't interrested in this secgroup
15:37:36 <msimonin> ?
15:37:39 <oanson> Yes
15:37:58 <oanson> Basically, we can filter this on the tenant level - if the node doesn't have any tenant ports installed, it won't get the event.
15:38:13 <msimonin> so in a massively distributed clouds this would make a big multicast ?
15:38:13 <oanson> But all nodes with ports belonging to the tenant (or project) will get the event
15:38:35 <oanson> Yes.
15:38:47 <oanson> We trust the publish subscribe mechanism to be light enough.
15:39:43 <msimonin> and reliable enough ?
15:40:08 <oanson> No. We verify we're up-to-date by syncing with the database
15:40:30 <oanson> I think the sync time is configurable
15:40:39 <msimonin> if an agent miss some messages, it detects it and resync with the db ?
15:40:46 <oanson> Yes
15:41:48 <msimonin> I think this would be a good exercice to check if oslo messaging could be a plugin.
15:41:58 <kgiusti> oanson: msimonin: I can do that
15:42:18 <oanson> That would be great!
15:42:23 <kgiusti> hopefully by next meeting, but no promises (will *try* to limit the egg nog)
15:42:23 <ad_rien_> sorry
15:42:33 <msimonin> kgiusti: if there anything I can do to help ?
15:42:35 <oanson> kgiusti, I am available for any help you need
15:42:40 <ad_rien_> I've been preempted
15:43:19 <msimonin> I'm interesting in the topic and this sounds related to the experiments we are planning to do
15:43:22 <kgiusti> msimonin: did you want to do the evaluation?  I didn't mean to rush in
15:43:34 <msimonin> regarding the massively distributed RPCs
15:43:43 <ad_rien_> Guys may I ask you to take some notes on the pad just to have an overview of what you discuss? races for the other
15:43:53 <ad_rien_> s/race for the other/ /
15:44:03 <kgiusti> ad_rien_: yep - happy holidays + new year to you!
15:44:09 <msimonin> kgiusti: not I'm interested but since I don't know at all the oslo_messaging internals I don't think I am the best candidate
15:44:43 <kgiusti> msimonin: let me take a quick pass through then we can plan work if that is the next logical step.
15:44:58 <msimonin> kgiusti: ack :)
15:45:10 <ad_rien_> (kgiusti I'm back now, have been preempted for the last couple of minutes)
15:45:32 <ad_rien_> Can we move to the other topic
15:45:33 <ad_rien_> ?
15:45:40 <msimonin> should we put an action ?
15:45:44 <kgiusti> oanson: if you can send me some pointers where I can start ramp up on dragonflow + drivers (kgiusti@gmail.com) that'd be great
15:45:51 <kgiusti> msimonin: I'll do that
15:45:57 <oanson> kgiusti, sure. I'll do that!
15:46:11 <ad_rien_> Actually, I would like to start preparing the PGT meeting as I think it is important to come with foods for core-devs discussions
15:46:12 <ad_rien_> ?
15:46:23 <kgiusti> go for it - I'm all set!
15:46:27 <ad_rien_> can we switch or you have a last point to discuss ?
15:47:15 <kgiusti> switch ok!
15:47:18 <kgiusti> :)
15:47:23 <msimonin> (oanson, kgiusti could you cc me at some point, I'll try to learn)
15:47:28 <ad_rien_> oanson:  msimonin ok for you too?
15:47:31 <msimonin> yes
15:47:38 <oanson> Sure
15:47:40 <ad_rien_> thanks
15:48:13 <ad_rien_> So I propose to jump over the two last items regarding the on-going actions as we did not make significant progress and I propose to discuss the PTG topic ;)
15:48:31 <ad_rien_> #topic PTG-Preparation
15:49:17 <ad_rien_> so my goal is to try to prepare as much as possible the PTG
15:49:29 <ad_rien_> I would like to propose to work in subgroups
15:49:32 <msimonin> What will be the format ?
15:49:46 <ad_rien_> in order to collect as much as possible informations regarding the code base
15:49:53 <ad_rien_> format will be the one we want to have
15:50:07 <ad_rien_> I think there will be at least one sessions regarding the use-cases
15:50:07 <msimonin> Oh :)
15:50:29 <ad_rien_> but because I know that we can discuss about use-cases during three days without diving into details
15:50:38 <ad_rien_> I think it is important to come with more concrete stuffs
15:50:51 <ad_rien_> for instance an etherpad with three identified use-cases
15:51:07 <ad_rien_> and then conduct a gap analysis of the current code-base w.r.t. these use-cases
15:51:39 <ad_rien_> So If we can define some milestones just to help us to move forward and be sure we will have enough materials before the PTG
15:51:43 <ad_rien_> I think it would be great
15:51:48 <ad_rien_> right now we have one use-case
15:52:01 <ad_rien_> i.e. control plane is centralized and we have remote nodes
15:52:11 <ad_rien_> vCPE/SDN/...
15:52:15 <ad_rien_> use-cases
15:52:21 <ad_rien_> we can also think about using the cells
15:52:27 <ad_rien_> what are the pros/cons of such an approcha?
15:52:35 <ad_rien_> using the regions what are the pros/cons
15:52:39 <ad_rien_> etc...
15:52:54 <ad_rien_> for each try to see whether we need additional piece of code
15:53:02 <ad_rien_> of if we can just revise the current code bas
15:53:03 <ad_rien_> e
15:53:33 <ad_rien_> or if we should really fork the code in order to propose another implementation/alternative of a particular service
15:53:46 <ad_rien_> (e.g. DragonFlow in comparison to the vanilla neutron implementation).
15:53:51 <ad_rien_> so far so good?
15:53:59 <ad_rien_> Any remarks/other ideas ?
15:55:05 <ad_rien_> no remark :-P ?
15:55:08 <msimonin> Usage of pub/sub in the context of massively distributed deployment ?
15:55:09 <msimonin> :)
15:55:22 <msimonin> (I'm biased)
15:55:23 <ad_rien_> from my view point this is interesting but probably a bit too specific
15:55:38 <ad_rien_> The goal from my viewpoint is to be able to discuss with the core developpers
15:55:46 <ad_rien_> core developpers will be focused on the code base
15:56:20 <ad_rien_> if we arrive with ''just'' another solution without any strong arguments against the current code base, I think our viewpoints will not be considered?
15:56:26 <ad_rien_> but it can be a topic indeed
15:56:47 <ad_rien_> How can core-services can communicate WANWide with intermittent network connectivity
15:56:53 <ad_rien_> Can you add it into the pad
15:56:56 <msimonin> yep
15:56:57 <msimonin> :)
15:56:59 <ad_rien_> thanks
15:57:02 <ad_rien_> anything else ?
15:57:08 <ad_rien_> Two minutes left ?
15:57:35 <msimonin> IMHO the dragonflow implementations are interresting
15:57:39 <ad_rien_> oanson:  maybe you would like to discuss network aspects
15:57:41 <ad_rien_> exactly ;)
15:57:43 <ad_rien_> thanks msimonin
15:57:49 <ad_rien_> but the same we need to come with some materials
15:58:00 <oanson> ad_rien_, sure.
15:58:01 <ad_rien_> so if dragonflow folks are interested and have time to prepare a bit
15:58:04 <ad_rien_> it would be nice
15:58:25 <oanson> Definitely. We'll see what we can come up with.
15:58:27 <ad_rien_> I know that the neutron team is planning to impement a cell like concept within the neutron code
15:58:31 <ad_rien_> great
15:58:33 <ad_rien_> ok thanks guy
15:58:41 <ad_rien_> please put some notes into the pad
15:58:48 <ad_rien_> nothing else from my side
15:58:56 <ad_rien_> I wish you a merry christmas
15:59:06 <msimonin> Thanks ad_rien_ ! You too
15:59:06 <ad_rien_> and all the best for the two next weeks !
15:59:22 <oanson> Thanks! You too.
15:59:27 <ad_rien_> #endmeeting