New pool members receive traffic without a healthcheck

[kay]

When a pool has a health check, new pool members start receiving traffic right away. This causes an issue when a new TCP pool member is not reachable for some reason.

The expected behavior: when a pool has a health check, new pool members must be marked offline until the health check is successful.

See https://github.com/kubernetes/cloud-provider-openstack/issues/2148

[Gregory Thiemonge (gthiemonge)]

Hi,

Do you have more details about that? (the configuration of the octavia resources: lb, listener, pool, hm, etc...)
What release do you use? What is the distrib (and haproxy version) of the amphora?

I did a quick test in my env: haproxy forwarded all the requests to the "ONLINE" members (it tried to reach the new members but then used the other members after a short delay: 2sec) and I had no failed connections.

[Michael Johnson (johnsom)]

This is probably a health check configuration error. The setting "max_retries_down" defaults to 3, meaning it won't mark a member down until three retry intervals have failed. I think in your use case, you want to set that to 1.
See: https://docs.openstack.org/api-ref/load-balancer/v2/index.html?expanded=create-health-monitor-detail#id106