| *** thorst has joined #openstack-watcher | 01:13 | |
| *** thorst has quit IRC | 01:22 | |
| *** licanwei has joined #openstack-watcher | 01:26 | |
| *** thorst has joined #openstack-watcher | 01:31 | |
| *** thorst has quit IRC | 01:32 | |
| *** thorst has joined #openstack-watcher | 02:27 | |
| *** thorst has quit IRC | 02:27 | |
| *** thorst has joined #openstack-watcher | 03:36 | |
| *** thorst has quit IRC | 03:41 | |
| *** thorst has joined #openstack-watcher | 04:38 | |
| *** thorst has quit IRC | 04:46 | |
| *** kuntumtarm has joined #openstack-watcher | 06:07 | |
| *** diga has joined #openstack-watcher | 06:32 | |
| *** kuntumtarm has quit IRC | 06:34 | |
| *** thorst has joined #openstack-watcher | 06:48 | |
| *** thorst has quit IRC | 06:56 | |
| *** alexchadin has joined #openstack-watcher | 07:05 | |
| *** dtardivel has joined #openstack-watcher | 07:14 | |
| *** jed56 has joined #openstack-watcher | 07:35 | |
| *** alexchadin has quit IRC | 07:37 | |
| *** alexchadin has joined #openstack-watcher | 07:37 | |
| *** vincentfrancoise has joined #openstack-watcher | 07:39 | |
| *** thorst has joined #openstack-watcher | 07:54 | |
| *** Kevin_Zheng has joined #openstack-watcher | 07:59 | |
| *** thorst has quit IRC | 08:02 | |
| *** thorst has joined #openstack-watcher | 08:59 | |
| openstackgerrit | Vincent Françoise proposed openstack/watcher-specs: Specs for watcher-notifications-ovo https://review.openstack.org/384370 | 09:03 |
|---|---|---|
| *** thorst has quit IRC | 09:06 | |
| *** Guest10079 is now known as zigo | 09:23 | |
| *** alexchadin has quit IRC | 09:26 | |
| *** alexchadin has joined #openstack-watcher | 09:26 | |
| diga | acabot: Hi | 09:28 |
| diga | alexchadin: vincentfrancoise : Hi | 09:28 |
| diga | all given comments are resolved & submitted updated PS on friday | 09:29 |
| *** alexchadin has quit IRC | 09:31 | |
| *** alexchadin has joined #openstack-watcher | 10:39 | |
| sballe_ | Morning | 10:43 |
| *** thorst has joined #openstack-watcher | 11:08 | |
| *** thorst has quit IRC | 11:16 | |
| alexchadin | hello jed56 | 11:39 |
| jed56 | Hello alexchadin | 11:39 |
| jed56 | как вы? | 11:39 |
| jed56 | :) | 11:39 |
| alexchadin | Я отлично. А вы?:) | 11:39 |
| alexchadin | jed56: are you learning Russian language?:) | 11:40 |
| jed56 | no but a good friend of me is russian | 11:41 |
| alexchadin | jed56: oh, that's great! It is pretty hard to learn language even for native. | 11:42 |
| jed56 | okay ;-) | 11:42 |
| alexchadin | jed56: do you have about 10 minutes to discuss improvements of action plan storage? I'm working on the spec for now. | 11:43 |
| jed56 | yes of course | 11:43 |
| jed56 | did you push a review ? | 11:44 |
| alexchadin | review on what? | 11:44 |
| jed56 | on the action-plan-storage blueprint ? | 11:44 |
| alexchadin | Not yet:) | 11:44 |
| alexchadin | Look | 11:45 |
| alexchadin | I've discussed with our architecture about action weights, he said that all the migrations can be launched at the same time for specified action plan. | 11:47 |
| alexchadin | so, we can leave next_action field | 11:47 |
| alexchadin | if we want to specify special subgraph | 11:48 |
| jed56 | alexchadin: IMHO, this is not a good idea because if you have 20 migrations at the same time you can saturate the network | 11:48 |
| alexchadin | We will limit them | 11:48 |
| jed56 | Action weight is a simple implementation of the planner, it has many drawback | 11:49 |
| alexchadin | Every action may have priority | 11:51 |
| *** thorst has joined #openstack-watcher | 11:51 | |
| alexchadin | we can run actions in parallel in order of these priorities | 11:51 |
| jed56 | yes | 11:52 |
| jed56 | I have a meeting can we continue in 30 minutes ? | 11:53 |
| sballe_ | jed56: did you take a look at the slides for the summit? Please send me your edits soon | 11:53 |
| jed56 | Yes a little bit :) | 11:53 |
| sballe_ | Ok | 11:53 |
| alexchadin | jed56: yeah, of course | 11:53 |
| acabot | diga : https://review.openstack.org/#/c/342473/ your last PS fail because you have removed references | 11:54 |
| acabot | diga : please run tox -edocs before submitting your PS for review | 11:55 |
| diga | acabot: you said "remove the references" so I did, but you want me to revert back | 11:56 |
| diga | ok | 11:56 |
| diga | for b-com | 11:56 |
| acabot | diga : I didn't say remove references | 11:57 |
| diga | acabot: I means, b-com link | 11:57 |
| acabot | diga : I said "all links to factory.b-com.com are dead links, please use the official doc now on http://docs.openstack.org/developer/watcher/" | 11:57 |
| acabot | diga : keep the links but replace factory.b-com.com by docs.openstack.org | 11:57 |
| diga | b-com links, so I removed b-com links & added | 11:57 |
| diga | ohhh kk | 11:58 |
| diga | acabot: I will add that change | 11:58 |
| diga | acabot: sorry for misunderstanding | 11:58 |
| acabot | diga : no pb | 12:00 |
| diga | acabot: pushed new PS - https://review.openstack.org/#/c/342473/ | 12:26 |
| diga | tox -edocs - successful | 12:26 |
| alexchadin | jed56: http://pastebin.com/VrH0maHD in this case we need to disable hypervisor 2 only after migration process for 5th VM is complete. I see here two ways: either to complete all the migration action, then complete all the disable action or to manage every subgraph independently. If we choose the second way, here is a question: where will we bind actions to each other? On the strategy layer or somehow on planner layer? | 12:27 |
| jed56 | alexchadin: if I understand well | 12:31 |
| jed56 | your question | 12:32 |
| jed56 | we will bind or schedule the action in the planner | 12:32 |
| jed56 | We can do the same thing | 12:32 |
| jed56 | of your chart | 12:32 |
| jed56 | I agree that the weight can't be set on the action in this case | 12:33 |
| jed56 | The Watcher planner like the strategy are pluggable | 12:33 |
| alexchadin | jed56: how will planner know which actions should be binded?:) | 12:34 |
| diga | acabot: updated new PS - https://review.openstack.org/#/c/342473/ | 12:34 |
| jed56 | What do you mean by "binded" ? | 12:35 |
| jed56 | I'm not sure to understand :) | 12:35 |
| alexchadin | jed56: Look - we have 5 VM and 2 hypervisors | 12:36 |
| alexchadin | We want to disable the first hypervisor once 1-4 VMs are migrated | 12:36 |
| alexchadin | and we want to disable the second one once the 5th VM is migrated | 12:36 |
| alexchadin | so, how will we tell the system that disabling of the second hypervisor should come only after migrating of the 5th VMs? | 12:37 |
| alexchadin | 5th VM* | 12:37 |
| jed56 | Okat | 12:38 |
| jed56 | okay | 12:38 |
| jed56 | So you question: How de we develop a heuristic that can schedule in parallel the migrations of the VM's per each hypervisor ? | 12:39 |
| jed56 | that's it ? | 12:41 |
| alexchadin | Not exactly. | 12:41 |
| jed56 | okay :p | 12:41 |
| jed56 | Your question is regarding the way that we will store the results or the way that we will compute the order ? | 12:42 |
| alexchadin | My question is: How can we define the vertexes of every subgraph? | 12:43 |
| alexchadin | jed56: yeah, how can we compute the order of vertexes! | 12:43 |
| jed56 | okay :) | 12:44 |
| jed56 | IMHO, if you have a parameter on the planner that limit the number of parallel migrations to 4 | 12:45 |
| jed56 | you can select in the list actions the hypervisor with the largest number of VM's | 12:47 |
| jed56 | create 4 vertex from the source node | 12:48 |
| jed56 | then add a new node of level 2 add a new vertex | 12:48 |
| alexchadin | ok, so we will launch actions in parallel by something like "layers" (we start with migrations, then continue with disabling node and so on). Right? | 12:48 |
| alexchadin | diga: ping | 12:49 |
| jed56 | https://www.irccloud.com/pastebin/FeQ9gpo8/ | 12:51 |
| jed56 | You can image many concurrent layer | 12:52 |
| alexchadin | jed56: so here we have at least two types of vertex - node and migration, right? | 12:53 |
| alexchadin | and disable | 12:53 |
| alexchadin | the third:) | 12:53 |
| sballe_ | jed56: it is a Holliday in the US today. Please send me your slides by EOD Wednesday and I will finalize them on Thu and Fri | 12:54 |
| jed56 | The vertex is modeling a transition that can be concurrent if you have several vertex from one none | 12:55 |
| jed56 | s/none/node | 12:55 |
| alexchadin | jed56: we start from level 0 and goes up? | 12:58 |
| alexchadin | go* | 12:58 |
| jed56 | You have to understand that we three notions | 13:00 |
| jed56 | We have the node, the vertex and the task | 13:00 |
| jed56 | In our case the task are the actions | 13:00 |
| jed56 | For example in taskflow | 13:00 |
| jed56 | r = gf.Flow("root") | 13:00 |
| jed56 | n_1 = gf.Flow("NODE_1") | 13:01 |
| jed56 | n_2 = gf.Flow("NODE_2") | 13:01 |
| jed56 | class MigrateVM(task.Task): | 13:01 |
| jed56 | def execute(self): | 13:01 |
| jed56 | print("Running %s" % self.name) | 13:01 |
| jed56 | flow.link(r, n_1) | 13:02 |
| jed56 | flow.kink(r,n_2) | 13:02 |
| jed56 | so NODE_1 and NODE_2 will be executed in concurrent | 13:03 |
| jed56 | mig = MigrateVM(VM1) | 13:03 |
| jed56 | n_1.add(mig) | 13:03 |
| jed56 | In order to solve that we add new parameter | 13:04 |
| jed56 | default planner = we can add a new parameter max 2 migration per hypervisor | 13:05 |
| jed56 | and 10 migrations max in global | 13:05 |
| jed56 | this is just an example but we need to have the topology and the bandwidth available to compute a good solution | 13:05 |
| alexchadin | now it makes more sense | 13:07 |
| *** diga has quit IRC | 13:07 | |
| jed56 | Great ! :) | 13:07 |
| alexchadin | well thank you a lot, now I understand what you want:) | 13:09 |
| *** alexchadin has quit IRC | 13:09 | |
| *** alexchadin has joined #openstack-watcher | 13:09 | |
| jed56 | welcome | 13:26 |
| jed56 | your welcome | 13:27 |
| *** alexchadin has quit IRC | 13:31 | |
| *** hanrong has quit IRC | 13:49 | |
| *** michaelgugino has joined #openstack-watcher | 13:52 | |
| *** diga has joined #openstack-watcher | 14:12 | |
| diga | acabot: vincentfrancoise : Hi | 14:12 |
| diga | need your advise on alex's comment | 14:12 |
| diga | In the previous he has proposed not to use audit/base.py & now he is again asking me to use audit/base.py. | 14:14 |
| diga | acabot: I was also saying better to go with auto/base.py, need your final call on this | 14:20 |
| openstackgerrit | Mykyta Karpin proposed openstack/puppet-watcher: Fix documentation for log_dir parameter https://review.openstack.org/384560 | 14:34 |
| *** openstack has joined #openstack-watcher | 18:22 | |
| *** jed56 has quit IRC | 18:27 | |
| *** openstack has joined #openstack-watcher | 19:04 | |
| *** diga has quit IRC | 19:14 | |
| *** dtardivel has quit IRC | 19:24 | |
| *** michaelgugino has quit IRC | 20:23 | |
| *** thorst has quit IRC | 21:27 | |
| *** thorst has joined #openstack-watcher | 21:56 | |
| *** thorst has quit IRC | 22:01 | |
| *** thorst has joined #openstack-watcher | 23:10 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!