There seems to be some sort of issue with the “Backend keepalive” setting on the DO load balancers.
Whenever I deploy new versions of pods to my cluster, there are a few seconds when 503 errors are returned for most web request. This does not happen when “Backend keepalive” is disabled. These 503 errors only appear for a second or two. And it happens whether or not my backend server has its keepalive functionality.
Has anyone seen something like this?