First of all I really like the ease of use of the Rancher loadbalancer for simple tasks.
But there is one thing that keeps me puzzled, after I upgrade a webserver that is behind a loadbalancer I sometimes get 503 notice from the loadbalancer after the webserver is upgraded and running.
Is the loadbalancer doing some health check on the linked containers?
P.S. any tips on how to get zero downtime upgrades would be awesome!
What version of Rancher are you using and what version of rancher/agent-instance are the load balancer agents?
rancher v0.59
and the rancher-agent-instance is v0.8.0
Some suggestions for zero downtime are:
When doing upgrade, use a batch size less than the scale of your service to ensure that there are never 0 running containers.
And configure a health check for your webserver so that the LB would never forward requests to the service until it’s considered up and healthy.