PR #263 provides TCP connections to renderd. However, when there is a load balancer (or Kubernetes service backed by the Linux IPVS, in my case) between render_list and renderd, it silently closes TCP connections due to inactivity (because renderd is busy) after some time (15 minutes, in my case) causing render_list to fail. Enabling TCP keepalives on the socket keeps the connection alive.

I am preparing a PR for this.