Oliver Gould

FWIW, I'm not seeing this behavior on my local k3d cluster with emojivoto...

@johnswarbrick Sorry, I may have given you a slightly incorrect command. These logs are from the proxy, but I'm curious to see the logs from the destination controller. I think...

@jensoncs Richer client-side policies are planned for stable-2.12.0

Thanks for the writeup @dghubble!

> error shutting down IO: client: Transport endpoint is not connected (os error 107) The proxy is getting a socket error. When the proxy is attempting to read or write...

I can repro something similar locally (in k3d) with a recent edge: ``` :; linkerd version Client version: edge-22.8.3 Server version: edge-22.8.3 ``` ``` :; k logs -n ingress-nginx ingress-nginx-defaultbackend-78dd86b9f5-86pqr...

``` :; kubectl sniff -n ingress-nginx ingress-nginx-defaultbackend-78dd86b9f5-86pqr -o nginx-backend.pcap --privileged --socket /run/k3s/containerd/containerd.sock ``` [nginx-backend.pcap.gz](https://github.com/linkerd/linkerd2/files/9508385/nginx-backend.pcap.gz) --- Unfortunately this misses proxybackend traffic and only covers kubeletproxy traffic :(

We see two connections firing probe requests ~simultaneously, with only one failing: ```log [ 138.917264s] DEBUG ThreadId(01) inbound:accept{client.addr=10.42.0.1:57960}:server{port=8080}: linkerd_app_inbound::accept: Accepted policy=ServerPolicy { protocol: Detect { http: [Route { hosts: [],...

https://github.com/jmreid/ingress-nginx/blob/d9f58144eb153dadab79f50038dfbd4b7db4fbe3/images/404-server/server.go appears to be a copy of the server code -- this image is no longer present in the ingress-nginx repo, though. Its HTTP server has the settings: ```go return...

The default idle timeout was reduced in stable-2.11.5 and stable-2.12.1, which should help fix the problem for the default nginx backend. Let us know if you see this problem on...