I’ve been experiencing some long calls when I call my exposed port on a worker node.
My configuration is very simple: 3 manager nodes and 2 worker nodes, a service with two replicas running only on workers and with port 8080 published.
Some calls to http://worker1:8080 can take up to 30 seconds.
My understanding is that when I call http://worker1:8080 iptables redirects my request to ingress-sbox:8080 (172.18.0.2:8080). So from worker1 I called http://172.18.0.2:8080 and I could still see my 15 seconds calls. However if I call http://172.18.0.2:8080 from manager1 all calls take a maximum of 1 ms (the service running on port 8080 is just a ping).
I don’t know what happens when I call ingress-sbox:8080. Does it call my service’s VIP on the internal port?
Anyone has any ideias what can be the cause of this issue?
My guess it that ingress-sbox running on worker1 needs to communicate with a manager to get the service’s VIP and the communication is hanging for some reason.