| danfai | Did anyone face an issue before with UDP, where one member going down was not reflected since a client was constantly sending from the same IP:port combination? I'll have to dig more, but maybe someone saw it before | 13:27 |
|---|---|---|
| opendevreview | Alexander Shishebarov proposed openstack/octavia master: Add set error to LB on memever revert https://review.opendev.org/c/openstack/octavia/+/951956 | 13:32 |
| opendevreview | Alexander Shishebarov proposed openstack/octavia master: Add set error to LB on memever revert https://review.opendev.org/c/openstack/octavia/+/951956 | 13:35 |
| opendevreview | Alexander Shishebarov proposed openstack/octavia master: Add set error to LB on memever revert https://review.opendev.org/c/openstack/octavia/+/951956 | 13:37 |
| opendevreview | Alexander Shishebarov proposed openstack/octavia master: Add set error to LB when amphora error. https://review.opendev.org/c/openstack/octavia/+/951956 | 13:45 |
| johnsom | danfai I think you are asking about members not going into ERROR state on a UDP load balancer when the backend server goes down? | 16:22 |
| johnsom | If so, it's likely a security group configuration error. Since UDP is stateless, the UDP health checks depend on the backend components being able to send the load balancer back ICMP indicating that the backend server is unreachable. If the router/host can't send the ICMP back to the load balancer (such as blocked/dropped by a security rule) the load balancer has no way to know that a UDP service is down. | 16:24 |
| johnsom | The issue is discussed here in the docs: https://docs.openstack.org/octavia/latest/user/guides/basic-cookbook.html#other-health-monitors | 16:26 |
| opendevreview | Michael Johnson proposed openstack/octavia master: Fill out the pyproject.toml for Octavia https://review.opendev.org/c/openstack/octavia/+/951994 | 20:32 |
| rm_work | Ugh ping | 21:22 |
| rm_work[m] | forgot that UDP-CONNECT would technically also be ICMP | 21:23 |
| rm_work[m] | had to deal with that the other day where wireshark told me I was getting an ICMP back from a UDP delivery failure and was like wut | 21:23 |
| rm_work[m] | didn't realize that stuff was delivered by ICMP, but the more you know | 21:24 |
| rm_work[m] | I wonder if we need to extend that note to the UDP-CONNECT entry too | 21:24 |
| rm_work[m] | oh neat we're moving to pyproject.toml | 21:25 |
| rm_work[m] | wait isn't that the old news now and there's some other thing that's the new hotness XD | 21:25 |
| rm_work[m] | <opendevreview> "Merged openstack/octavia master:..." <- oops, did this always work this way or is it new? I guess I did remove-ssh the dumb way | 21:27 |
| johnsom | Probably, but we are just entering the land of pyproject.toml (with pbr bolted on the side) | 22:43 |
| johnsom | The SSH thing came in when we fixed an issue with DIB and the cloud images.... We need to revisit that workaround and see if we can remove it frankly | 22:44 |
| johnsom | https://github.com/openstack/octavia/blob/master/elements/amphora-agent/package-installs.yaml#L40 | 22:44 |
| johnsom | We really should not need/be including all of those base packages IMO | 22:45 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!