| openstackgerrit | Merged openstack/octavia-tempest-plugin master: Add stable/victoria jobs to check and gate queues https://review.opendev.org/c/openstack/octavia-tempest-plugin/+/755210 | 00:30 |
|---|---|---|
| *** sapd1 has joined #openstack-lbaas | 01:49 | |
| *** irclogbot_2 has quit IRC | 02:06 | |
| *** irclogbot_1 has joined #openstack-lbaas | 02:09 | |
| *** rcernin has quit IRC | 02:18 | |
| *** rcernin has joined #openstack-lbaas | 02:21 | |
| *** rcernin has quit IRC | 02:35 | |
| *** rcernin has joined #openstack-lbaas | 02:49 | |
| *** sapd1 has quit IRC | 03:23 | |
| *** sapd1 has joined #openstack-lbaas | 03:31 | |
| *** sapd1 has quit IRC | 03:36 | |
| *** rcernin has quit IRC | 03:38 | |
| *** rcernin has joined #openstack-lbaas | 04:19 | |
| *** rcernin has quit IRC | 04:43 | |
| *** rcernin has joined #openstack-lbaas | 07:00 | |
| *** rcernin has quit IRC | 07:01 | |
| *** rcernin has joined #openstack-lbaas | 07:01 | |
| *** rcernin has quit IRC | 07:48 | |
| *** sapd1 has joined #openstack-lbaas | 07:58 | |
| *** icey has quit IRC | 08:03 | |
| *** icey has joined #openstack-lbaas | 08:05 | |
| *** vesper11 has joined #openstack-lbaas | 09:25 | |
| *** vesper11 has quit IRC | 09:26 | |
| *** eandersson has quit IRC | 09:27 | |
| *** eandersson has joined #openstack-lbaas | 09:27 | |
| *** vesper11 has joined #openstack-lbaas | 09:28 | |
| *** rcernin has joined #openstack-lbaas | 09:59 | |
| *** rcernin has quit IRC | 10:01 | |
| *** rcernin has joined #openstack-lbaas | 10:02 | |
| *** vesper11 has quit IRC | 10:21 | |
| *** rcernin has quit IRC | 12:08 | |
| *** rcernin has joined #openstack-lbaas | 12:18 | |
| *** rcernin has quit IRC | 12:40 | |
| *** jamesdenton has quit IRC | 12:42 | |
| *** jamesdenton has joined #openstack-lbaas | 12:43 | |
| *** sapd1 has quit IRC | 14:46 | |
| *** sapd1 has joined #openstack-lbaas | 15:39 | |
| *** sapd1 has quit IRC | 15:51 | |
| *** sapd1 has joined #openstack-lbaas | 16:03 | |
| *** jamesdenton has quit IRC | 16:40 | |
| *** jamesdenton has joined #openstack-lbaas | 16:41 | |
| *** sapd1 has quit IRC | 16:43 | |
| *** __ministry1 has joined #openstack-lbaas | 17:00 | |
| *** __ministry1 has quit IRC | 17:32 | |
| *** rcernin has joined #openstack-lbaas | 18:37 | |
| *** rcernin has quit IRC | 18:41 | |
| *** fnpanic has joined #openstack-lbaas | 20:41 | |
| fnpanic | hi guys | 20:41 |
| fnpanic | is it possible to spawn the two active/standy amphoras in two AZs? | 20:41 |
| fnpanic | i see this patch here but it looks like it is not merged yet | 20:41 |
| fnpanic | https://review.opendev.org/c/openstack/octavia/+/558962 | 20:41 |
| johnsom | No, not at this time | 20:42 |
| fnpanic | ok | 20:42 |
| johnsom | Question for you however, do you have IP visibility across the AZs? | 20:43 |
| fnpanic | yes | 20:43 |
| johnsom | I.e. can the VIP move across your AZs? | 20:43 |
| fnpanic | l2 for the managemnt and the VIP network aswell | 20:43 |
| johnsom | Ok, thanks | 20:43 |
| fnpanic | the managment can be routed right? | 20:43 |
| johnsom | Yes | 20:43 |
| fnpanic | k | 20:43 |
| johnsom | It's just the VIP in an scenario where the amphora a spread across AZs. | 20:44 |
| fnpanic | so when i loose one az today, an the octavia control plane is stretched | 20:44 |
| johnsom | We are curious how many people have the capability to allow that VIP to migrate between the AZs | 20:44 |
| fnpanic | will the healthmanager fix the amphoras automatically? | 20:44 |
| fnpanic | not so many i guess but it is doable with vxlan in the provider network | 20:45 |
| fnpanic | maybe this does not work because of the az profile right? | 20:46 |
| johnsom | If an amphora in a load balancer fails, for whatever reason (meteor hits DC for AZ 1), a health manager will start a failover process for it to rebuild it. | 20:46 |
| fnpanic | so how does this play with az profiles introduced in ussuri? | 20:47 |
| johnsom | That AZ/AZ profile support we have today is for building the load balancer at an edge site. In that case both amphora will be built there if it is an active/standby load balancer | 20:47 |
| fnpanic | yes | 20:48 |
| johnsom | Splitting amphora across AZs is just not implemented yet. | 20:48 |
| fnpanic | yes got this :-) | 20:48 |
| fnpanic | asking if a self heal is possible in case of loosing one AZ | 20:49 |
| johnsom | So, if the LB has a defined AZ, and the amphora fails, we will work with nova to rebuild a new amphora inside that AZ definition (edge site). | 20:49 |
| fnpanic | ok | 20:49 |
| fnpanic | but if no AZ is definded? | 20:49 |
| johnsom | Yeah, today, if the whole AZ goes down, the LBs are offline until that is resolved. | 20:49 |
| fnpanic | ohh | 20:50 |
| fnpanic | so there is no way to failover? | 20:50 |
| johnsom | If no AZ is defined, it is a normal nova allocation. | 20:50 |
| fnpanic | we depend on the LBs | 20:50 |
| johnsom | We do implement nova anti-affinity, but today that is only at the host level, not the AZ | 20:50 |
| fnpanic | but when i do not define one? | 20:50 |
| *** ccamposr has joined #openstack-lbaas | 20:50 | |
| fnpanic | yes we use anti-affinity | 20:51 |
| fnpanic | placing one gigant host in each DC and making one AZ auto of it is not such a great design | 20:51 |
| johnsom | Really you only need to create an Octavia AZ definition if you want to build load balancers in a different nova AZ with different networks. | 20:51 |
| fnpanic | with soft-anti affinity it could recover then. | 20:52 |
| johnsom | I don't think I'm quite following you. Why would you have one host? | 20:52 |
| fnpanic | no, talking about the anti-affinity host feature | 20:53 |
| fnpanic | absuing this and creating one AZ with two hosts, one host in each az. Then nova is forced to spread the amphoras. | 20:53 |
| fnpanic | but this is a bad design | 20:53 |
| fnpanic | does this make sense? | 20:53 |
| johnsom | Yeah, so disabled, they could land on the same host, soft-antiaffinity means nova will try to land them on different hosts, hard anti-affinity means nova must be able to schedule them on different hosts or the create will fail. | 20:54 |
| fnpanic | yes but with hard there is no ha at the amphora level when we lost a DC | 20:54 |
| johnsom | If you can make nova work that way, and you have the right network visibility, sure, go for it. | 20:54 |
| fnpanic | no :-) | 20:55 |
| fnpanic | i do not like this. it would work but it is a bad design from my point of view | 20:55 |
| johnsom | It's better than all on one host. grin | 20:55 |
| johnsom | Yeah, we wish nova supported better scheduling really. | 20:55 |
| fnpanic | yeah that is right ;-) | 20:55 |
| johnsom | To implement cross-AZ amphora, we are basically going to have to become a compute scheduler, which makes us a bit sad. | 20:56 |
| fnpanic | so one last question. when we just let nova do it's job and have visibility of the vip network will it then heal all failed amphoras in the az which survived the meteor? | 20:57 |
| fnpanic | lbass should not. | 20:57 |
| johnsom | Yeah, we will detect it and start a rebuild process, we will just go back to nova and request an instance. So as long as nova can create that instance, you are good to go. | 20:58 |
| fnpanic | ok great | 20:58 |
| fnpanic | we will test this. | 20:58 |
| johnsom | Ok | 20:58 |
| fnpanic | btw octavia is awesome (i guess i said this before or at least someone else ) | 20:59 |
| johnsom | Let us know how it goes | 20:59 |
| fnpanic | will do | 20:59 |
| johnsom | We would like cross AZ load balancers, but we don't have a developer actively working on it at this time. If you are interested, we can connect you with the person that started that patch. It may need a spec written up just to make sure everything is covered, etc. | 21:00 |
| fnpanic | thanks! i will definitely come back. | 21:02 |
| johnsom | o/ | 21:03 |
| *** tkajinam has joined #openstack-lbaas | 22:59 | |
| *** ccamposr has quit IRC | 23:35 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!