| openstackgerrit | Hongbin Lu proposed openstack/kuryr-libnetwork: Add prefix to specified name of subnetpool https://review.openstack.org/426623 | 00:49 |
|---|---|---|
| openstackgerrit | Hongbin Lu proposed openstack/kuryr-libnetwork: Add prefix to the specified name of subnetpool https://review.openstack.org/426623 | 00:56 |
| *** v1k0d3n has quit IRC | 02:00 | |
| *** v1k0d3n has joined #openstack-kuryr | 02:02 | |
| *** hongbin has joined #openstack-kuryr | 02:44 | |
| *** yuanying has joined #openstack-kuryr | 04:48 | |
| openstackgerrit | Hongbin Lu proposed openstack/kuryr: Update the libnetwork devref https://review.openstack.org/426644 | 05:08 |
| *** hongbin has quit IRC | 05:10 | |
| *** saneax-_-|AFK is now known as saneax | 05:29 | |
| *** yedongcan has joined #openstack-kuryr | 06:09 | |
| *** yamamoto has quit IRC | 07:27 | |
| *** pcaruana has joined #openstack-kuryr | 08:15 | |
| *** devvesa has joined #openstack-kuryr | 08:19 | |
| *** saneax is now known as saneax-_-|AFK | 08:27 | |
| *** saneax-_-|AFK is now known as saneax | 08:36 | |
| apuimedo | irenab: pong | 08:48 |
| *** apuimedo has quit IRC | 08:50 | |
| *** apuimedo has joined #openstack-kuryr | 08:51 | |
| *** apuimedo has quit IRC | 08:59 | |
| *** apuimedo has joined #openstack-kuryr | 09:00 | |
| *** saneax is now known as saneax-_-|AFK | 09:01 | |
| irenab | apuimedo, good morning | 09:03 |
| apuimedo | irenab: good morning | 09:06 |
| apuimedo | :-) | 09:06 |
| *** saneax-_-|AFK is now known as saneax | 09:11 | |
| *** saneax is now known as saneax-_-|AFK | 09:15 | |
| *** janki has joined #openstack-kuryr | 09:31 | |
| openstackgerrit | Luis Tomas Bolivar proposed openstack/kuryr-kubernetes: [WIP] Adding pool of ports to speed up containers booting/deletion https://review.openstack.org/426687 | 09:32 |
| *** jchhatbar_ has joined #openstack-kuryr | 09:36 | |
| apuimedo | ltomasbo: starting the week in a good note! | 09:38 |
| *** janki has quit IRC | 09:38 | |
| ltomasbo | :D | 09:38 |
| ltomasbo | this is just a push of what I did last Friday | 09:39 |
| ltomasbo | I will work on that later on! It is just a quick PoC (as you saw it) | 09:39 |
| apuimedo | :-) | 09:39 |
| *** garyloug has joined #openstack-kuryr | 09:41 | |
| *** jchhatbar_ has quit IRC | 09:41 | |
| *** janki has joined #openstack-kuryr | 09:41 | |
| *** saneax-_-|AFK is now known as saneax | 09:58 | |
| *** janki has quit IRC | 09:58 | |
| openstackgerrit | Ilya Chukhnakov proposed openstack/kuryr-kubernetes: Improve pipeline/Async logging https://review.openstack.org/423903 | 10:13 |
| openstackgerrit | Dongcan Ye proposed openstack/kuryr-libnetwork: Modify fake v4 subnet result https://review.openstack.org/426698 | 10:22 |
| *** yedongcan has left #openstack-kuryr | 10:23 | |
| *** neiljerram has joined #openstack-kuryr | 10:30 | |
| *** yuanying_ has joined #openstack-kuryr | 10:48 | |
| *** yuanying has quit IRC | 10:51 | |
| *** janki has joined #openstack-kuryr | 11:12 | |
| openstackgerrit | Luis Tomas Bolivar proposed openstack/kuryr-kubernetes: [WIP] Adding pool of ports to speed up containers booting/deletion https://review.openstack.org/426687 | 11:16 |
| *** yuanying_ has quit IRC | 11:25 | |
| *** dims has joined #openstack-kuryr | 11:26 | |
| irenab | ltomasbo: hi | 11:38 |
| openstackgerrit | Luis Tomas Bolivar proposed openstack/kuryr-kubernetes: [WIP] Adding pool of ports to speed up containers booting/deletion https://review.openstack.org/426687 | 11:39 |
| irenab | may I ask you to add some notes regarding what you want to achive and how you want to address it in the blueprint description | 11:39 |
| ltomasbo | irenab, yes you can! | 11:41 |
| irenab | ltomasbo: :-) | 11:41 |
| ltomasbo | :D | 11:41 |
| ltomasbo | I will update the description with more detailed information | 11:42 |
| irenab | ltomasbo: thanks! | 11:44 |
| *** janki has quit IRC | 12:30 | |
| *** janki has joined #openstack-kuryr | 12:51 | |
| *** devvesa has quit IRC | 12:53 | |
| *** gsagie has joined #openstack-kuryr | 12:54 | |
| irenab | ivc_: apuimedo : check this email tread: https://groups.google.com/forum/#!msg/kubernetes-dev/O_u9otoPRkE/tjHlpV3SBwAJ | 12:55 |
| *** garyloug has quit IRC | 13:05 | |
| *** devvesa has joined #openstack-kuryr | 13:08 | |
| *** garyloug has joined #openstack-kuryr | 13:20 | |
| *** saneax is now known as saneax-_-|AFK | 13:29 | |
| *** saneax-_-|AFK is now known as saneax | 13:37 | |
| ivc_ | irenab thats one weird discussion. i'd say 'watch' is all we need, not even that 'getall' as OP suggested | 13:41 |
| irenab | ivc_: I was also suprised by the approach, but looks the k8s team has some history behind this | 13:42 |
| ivc_ | irenab i'm not sure if it actually is 'k8s team' and not just some random guys who got it wrong | 13:44 |
| ivc_ | anyway they are talking about 'cache' and we try to be stateless to prevent such sync issues | 13:46 |
| ivc_ | so far the only stateful thing we need (to be implemented) is the cleanup code (in case k8s resource was deleted during kuryr downtime) - where we can't rely on watch history | 13:47 |
| ivc_ | and thats pretty close to the neutron port reuse (and we already got ltomasbo working on it https://review.openstack.org/#/c/426687/ !) | 13:48 |
| irenab | ivc_: just wanted you to be aware of the reasoning for what is done currently | 13:50 |
| apuimedo | ivc_: and I saw ltomas work functioning :-) | 13:53 |
| irenab | apuimedo: I think ivc_ would balme me on being too pedantic again, but I asked ltomasbo to summarize his direction on blueprint :-) | 13:53 |
| apuimedo | irenab: good point! | 13:54 |
| ivc_ | apuimedo, ltomasbo its quite close to what i had in mind, just instead of inheriting i'd rather make a composable decorator so we can reuse the 'pool' with any backend | 13:54 |
| apuimedo | the process is. We get reuse -> We do pre-allocation and pool management | 13:54 |
| irenab | where pool is stored? | 13:55 |
| apuimedo | ivc_: I didn't look at the patch yet, only at the hack we were doing in between | 13:55 |
| apuimedo | irenab: I was considering Thirdparty objects | 13:55 |
| irenab | aouso k8s store | 13:55 |
| apuimedo | k8s store, yes | 13:55 |
| apuimedo | otherwise HA will be a pita | 13:56 |
| irenab | yea | 13:56 |
| irenab | apuimedo: what about multi tenancy? | 13:56 |
| ivc_ | apuimedo i'd say we don't store it anywhere but memory and rely on neutron device_owner/tags | 13:56 |
| irenab | will do since tenant for now? | 13:56 |
| irenab | single | 13:56 |
| apuimedo | ivc_: that's what ltomasbo proposes | 13:56 |
| apuimedo | :-) | 13:56 |
| irenab | ivc_: we may need sheduler to be involved later on, maybe k8s is better | 13:57 |
| apuimedo | irenab: I'm ashamed to say | 13:57 |
| apuimedo | that I don't have concrete tenancy plans | 13:57 |
| apuimedo | I want it | 13:57 |
| apuimedo | but I still didn't come up with the mapping | 13:57 |
| ivc_ | irenab problem with storing it in k8s is that you'll need to add tons of code for proper syncing | 13:57 |
| apuimedo | it is about trading operation time and complexity (maintaining the pool on k8s) for fast failover | 13:58 |
| irenab | ivc_: becouse of 3rd party, not a regular watch? | 13:58 |
| apuimedo | which, admittedly, in active/passive HA, which we do | 13:58 |
| apuimedo | makes the in-memory approach win | 13:58 |
| apuimedo | so until we have active/active, the thirdparty is not as strong an option | 13:59 |
| ivc_ | irenab because neutron and k8s 3rd party store have to be synced somehow | 13:59 |
| ivc_ | sync is bad | 13:59 |
| apuimedo | ivc_: the good thing is, if the neutron pool surface is only big batch ops, the sync is easier | 14:00 |
| apuimedo | anyway, let's move over to the meeting channel | 14:00 |
| ivc_ | sure | 14:00 |
| irenab | apuimedo: ivc_ I sujject we discuss it on the ltomasbo bp (or patch) | 14:01 |
| ivc_ | irenab in this case we do indeed need some doc (maybe not bp, but just a devref tho) | 14:03 |
| irenab | ivc_: I meant we can even disuss it on launchpad, but devref is much more convenient | 14:04 |
| ivc_ | irenab btw i've updated that patch and added the link to that 'generic' bp :) | 14:04 |
| irenab | ivc_: saw it, +2 already :-) | 14:04 |
| openstackgerrit | Merged openstack/kuryr-libnetwork: Remove subnetpool_id tag for Neutron existing subnet https://review.openstack.org/419735 | 14:09 |
| mchiappe1o | r | 14:19 |
| *** mchiappe1o is now known as mchiappero | 14:20 | |
| apuimedo | mchiappero: that's a funny way to start the irc day | 14:25 |
| apuimedo | :P | 14:25 |
| *** mattmceuen has joined #openstack-kuryr | 14:28 | |
| openstackgerrit | Merged openstack/kuryr-kubernetes: Bump tox min version to 2.3.1 https://review.openstack.org/424972 | 14:34 |
| openstackgerrit | Merged openstack/kuryr-kubernetes: Improve pipeline/Async logging https://review.openstack.org/423903 | 14:39 |
| *** hongbin has joined #openstack-kuryr | 14:45 | |
| janonymous | irenab, apuimedo: i will add a patch for that in requirements for starting up client migration. | 14:48 |
| apuimedo | janonymous: check the mailing list thread from Magnum to see which conclusions they reached | 14:49 |
| apuimedo | as well | 14:49 |
| janonymous | apuimedo: ohh sure | 14:49 |
| apuimedo | thanks Jaivish | 14:50 |
| janonymous | apuimedo:but only dims mail about client and no reply from magnum team | 14:53 |
| janonymous | apuimedo:should i ping container channel? | 14:53 |
| * apuimedo will check when the meeting ends | 14:54 | |
| apuimedo | sorry to keep you up so late (still in Japan, right?) | 14:54 |
| janonymous | apuimedo: yeah :) | 14:55 |
| janonymous | apuimedo: i think i wake up at night :P | 14:56 |
| apuimedo | :-) | 14:57 |
| dims | janonymous : apuimedo : i talked to strigazi as well on containers, please feel free to ping them again | 14:58 |
| apuimedo | alraddarla: ping | 14:58 |
| alraddarla | apuimedo, pong | 14:58 |
| mchiappero | apuimedo: sorry, it was the wrong terminal :D | 14:58 |
| apuimedo | mchiappero: :-) | 14:58 |
| apuimedo | alraddarla: I see that contrail is not integrated with openstack/os-vif | 14:59 |
| alraddarla | apuimedo, mattmceuen is also here to chat about this :) | 14:59 |
| apuimedo | so I guess that we could have the os-vif plug first in kuryr-kubernetes and if it works, push it to openstack/os-vif and drop from kuryr-kubernetes | 14:59 |
| janonymous | dims,apuimedo: sure | 15:00 |
| * apuimedo checking contrail binding code | 15:01 | |
| mattmceuen | With OpenContrail not falling under OpenStack governance, would that present a challenge with getting the os-vif plug into openstack/os-vif? | 15:01 |
| apuimedo | mattmceuen: nope | 15:01 |
| apuimedo | it only prevents us from putting an upstream gate for it | 15:02 |
| apuimedo | probably | 15:02 |
| apuimedo | (although not sure) | 15:02 |
| apuimedo | alraddarla: mattmceuen: Is this the binding code? https://github.com/openstack/nova/blob/e8cd3acf2d926c493aa4ae809397b2994a6679fc/nova/virt/libvirt/vif.py#L705-L745 | 15:02 |
| mattmceuen | We'll need to check. We don't have hands-on experience developing for contrail yet, but have teammates who have. | 15:04 |
| mattmceuen | Looks promising! | 15:04 |
| apuimedo | mattmceuen: very well, so the first step is to check if the binding is really just calling vrouter-port-control | 15:04 |
| apuimedo | and if it is, make it into something that os-vif.plug can run. @ivc_: os-vif could load drivers that live in kuryr-k8s, right? | 15:05 |
| ivc_ | apuimedo yes | 15:05 |
| mattmceuen | If that's the case, what SDN controller would be a good one to model after for kuryr integration? | 15:06 |
| hongbin | apuimedo: i have a question about the subnet creation in libnetwork, will be your next guest after you finish the current discussion | 15:06 |
| apuimedo | mattmceuen: unfortunately, this would be the first that is not already part of openstack/os-vif | 15:06 |
| apuimedo | mattmceuen: let me get you a couple of links | 15:07 |
| apuimedo | hongbin: good! | 15:07 |
| apuimedo | keep 'em coming everybody | 15:07 |
| apuimedo | mattmceuen: alraddarla: https://github.com/openstack/os-vif/blob/master/vif_plug_ovs/ovs.py | 15:08 |
| apuimedo | this is an example of a class that implements an os-vif plugin | 15:08 |
| apuimedo | we do not need all the methods implemented though | 15:08 |
| apuimedo | afaik we only call plug() | 15:08 |
| apuimedo | then, the other part is | 15:08 |
| apuimedo | https://github.com/openstack/kuryr-kubernetes/blob/794ec706c5fbe0da6e49bf20ba2439d8eb39ae7e/kuryr_kubernetes/os_vif_util.py#L258-L281 | 15:10 |
| apuimedo | In case that you need to serialize more fields for contrail inside teh vif annotation, this is an example of a newer serialization we added | 15:10 |
| apuimedo | and finally | 15:10 |
| apuimedo | https://github.com/openstack/kuryr-kubernetes/blob/794ec706c5fbe0da6e49bf20ba2439d8eb39ae7e/kuryr_kubernetes/cni/binding/bridge.py | 15:11 |
| apuimedo | here's the veth creation code and l3 config | 15:11 |
| apuimedo | s/l3/l2/ | 15:11 |
| apuimedo | if contrail needed anything different, you'd probably have a module like this one | 15:12 |
| apuimedo | mattmceuen: alraddarla: sorry for the extra work, the first extern driver pays the price :/ | 15:12 |
| mattmceuen | Gotcha. Thanks for all this helpful material apuimedo, this will get us off to a great start! | 15:12 |
| alraddarla | apuimedo, no problem! this is great! We are going to start looking into this | 15:12 |
| mattmceuen | We'll dig in today/tomorrow to wrap our heads around this and hit you up with questions shortly thereafter :-D | 15:13 |
| alraddarla | ^ guaranteed :) | 15:13 |
| apuimedo | mattmceuen: alraddarla: please ping ivc_, irenab or me if you need help | 15:13 |
| alraddarla | (the questions part) | 15:13 |
| apuimedo | cool | 15:13 |
| apuimedo | alraddarla: also, I'd appreciate if you document the steps you follow, so that we can leave it as a guide for others ;-) | 15:14 |
| alraddarla | No problem apuimedo! | 15:14 |
| alraddarla | Thanks again | 15:14 |
| apuimedo | you're welcome | 15:14 |
| apuimedo | ! | 15:14 |
| apuimedo | btw. Is anybody in the channel going to the PTG? | 15:15 |
| janonymous | apuimedo:just last question for today, about k8s client brief, should i do in vtg or details are required early | 15:20 |
| apuimedo | janonymous: we can talk about it earlier on the mailing list | 15:23 |
| apuimedo | and work it into the vtg as well | 15:23 |
| janonymous | apuimedo: cool! thanks ) | 15:24 |
| hongbin | apuimedo: have a minute? | 15:25 |
| apuimedo | hongbin: go ahead :-) | 15:26 |
| hongbin | apuimedo: i am looking at the code to create subnet in libnetwork | 15:27 |
| hongbin | apuimedo: it seems it creates a new subnet as long as there is not existing subnet with the pool cidr? | 15:27 |
| apuimedo | yes. I think that's what vikasc did when he split into ipam (before it was it all remote driver) | 15:28 |
| hongbin | apuimedo: i wonder if the logic could be modified a bit, kuryr creates a new subnet if there is not exsiting subnet in the network/subnetpool | 15:29 |
| hongbin | apuimedo: because i don't see why kuryr needs to create a subnet if there is already one (even with a smaller cidr) | 15:30 |
| *** saneax is now known as saneax-_-|AFK | 15:30 | |
| apuimedo | hongbin: I'm missing the differnece :P | 15:30 |
| apuimedo | *difference | 15:30 |
| hongbin | apuimedo: right now: create subnet if there is no subnet with the pool cidr | 15:30 |
| hongbin | apuimedo: proposed: create subnet if there is no subnet in the network/subnetpool (no cidr requirement) | 15:31 |
| apuimedo | oh, I see | 15:32 |
| hongbin | apuimedo: this handles the case that subnetpool and subnet are of different cidr (right now, kuryr assumes it is the same) | 15:33 |
| apuimedo | let me look for the code | 15:33 |
| hongbin | sure | 15:33 |
| *** janki has quit IRC | 15:34 | |
| apuimedo | hongbin: so you are arguing for removing the 'cidr=six.text_type(cidr)' from https://github.com/openstack/kuryr-libnetwork/blob/master/kuryr_libnetwork/controllers.py#L638-L639 | 15:40 |
| apuimedo | is that right? | 15:40 |
| hongbin | apuimedo: yes | 15:40 |
| hongbin | apuimedo: this is because there are cases that subnetpool and subnet with a different cidr | 15:41 |
| apuimedo | like? | 15:42 |
| apuimedo | you mean a manually created subnet that is not bound to the pool? | 15:42 |
| hongbin | apuimedo: yes | 15:42 |
| hongbin | apuimedo: for example, devstack create shared-default-subnetpool and private-subnet | 15:42 |
| hongbin | apuimedo: they are under different cidr | 15:43 |
| apuimedo | right | 15:46 |
| hongbin | apuimedo: i think simply pick a subnet under the subnetpool is enough | 15:46 |
| hongbin | apuimedo: it seems there is no need to require picking the one with the same cidr | 15:47 |
| apuimedo | hongbin: won't that break the user assumptions, if they create a docker net with a cidr | 15:48 |
| apuimedo | and then they create containers on that net and the ips are on a different range? | 15:48 |
| hongbin | apuimedo: kuryr should check if the subnet is under the pool | 15:49 |
| hongbin | apuimedo: user request a net with cidr -> find a subnetpool with the cidr -> find a subnet under the pool | 15:50 |
| hongbin | apuimedo: then the subnet must within the requested cidr (just with a smaller range) | 15:50 |
| apuimedo | right | 15:52 |
| hongbin | of course, the range doesn't match exactly | 15:53 |
| hongbin | that is the drawback | 15:53 |
| apuimedo | and that's a bit of problem | 15:53 |
| apuimedo | because we let neutron do the ipam | 15:53 |
| apuimedo | and we risk getting addreses out of the --ip-range the docker user passed | 15:54 |
| apuimedo | which is not acceptable | 15:54 |
| hongbin | i see | 15:54 |
| hongbin | another option is dropping the --ip-range option when users specify the subnetpool | 15:54 |
| apuimedo | hongbin: how? | 15:56 |
| apuimedo | just erroring out saying it is not supported to specify both? | 15:56 |
| hongbin | apuimedo: wait, --ip-range is the pool cidr? | 15:56 |
| apuimedo | mmm, don't remember now :P | 15:57 |
| apuimedo | I'll have to check | 15:57 |
| hongbin | apuimedo: https://review.openstack.org/#/c/426644/1/doc/source/devref/libnetwork_remote_driver_design.rst | 15:57 |
| hongbin | apuimedo: it seems if both --subnet and --ip-range are given, kuryr picked the --ip-range as pool cidr | 15:58 |
| apuimedo | yup | 15:58 |
| apuimedo | which is the intuitive thing with Neutron, but not for Docker afaict | 15:58 |
| hongbin | there is a problem of picking a smaller cidr than --ip-range? | 15:59 |
| apuimedo | you mean if subnet is smaller than ip-range? | 15:59 |
| hongbin | if yes, then we have to disallow --ip-range when using specified subnetpool | 15:59 |
| hongbin | yes | 15:59 |
| apuimedo | I think with docker it's the other way around | 16:00 |
| apuimedo | that subnet can't be smaller than ip-range | 16:00 |
| apuimedo | but I have to check | 16:00 |
| hongbin | ok | 16:00 |
| apuimedo | hongbin: I'll check it later | 16:00 |
| hongbin | apuimedo: sure. thanks for answering the questions | 16:00 |
| apuimedo | maybe we should bring this up to the ML so vikas, yedongcan and limao can comment on it | 16:01 |
| apuimedo | gotta go for a while now | 16:01 |
| hongbin | ok | 16:01 |
| *** devvesa has quit IRC | 16:01 | |
| hongbin | will do that | 16:01 |
| leifmadsen | o/ | 17:15 |
| *** oanson has quit IRC | 17:15 | |
| *** oanson has joined #openstack-kuryr | 17:17 | |
| *** saneax-_-|AFK is now known as saneax | 17:25 | |
| *** garyloug has quit IRC | 18:15 | |
| *** saneax is now known as saneax-_-|AFK | 18:33 | |
| *** tonanhngo has joined #openstack-kuryr | 18:34 | |
| *** neiljerram has quit IRC | 18:49 | |
| *** tonanhngo has quit IRC | 20:09 | |
| *** tonanhngo has joined #openstack-kuryr | 21:53 | |
| *** yamamoto has joined #openstack-kuryr | 22:03 | |
| apuimedo | leifmadsen: hey! | 22:39 |
| apuimedo | still around? | 22:39 |
| leifmadsen | apuimedo: aye | 22:43 |
| leifmadsen | kind of anyways heh | 22:44 |
| apuimedo | :-) | 22:44 |
| *** saneax-_-|AFK is now known as saneax | 23:03 | |
| *** pmannidi has joined #openstack-kuryr | 23:39 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!