MarcoPolo
commented
3 weeks ago Thanks for the detailed report @elecbug! Have you noticed this issue in real networks, or just localhost/low-latency home networks?
My first suspicion is that this is related to TCP simultaneous open feature, since that causes confusion as to who is the server or client. Could you try running your repro with a small change for me to disprove this theory?
Please add these lines before your host.Connect
order := (bytes.Compare([]byte(host.ID()), []byte(found)))
if order == -1 {
time.Sleep(100*time.Millisecond)
}
// existing code below
err = host.Connect(ctx, found)This basically has one side wait a bit if its peer id is lexicographically less. This should remove any chance for simultaneous open.
If the issue still happens after that change, please report back and I'll dive deeper.
elecbug
Stebalien
cpeliciari
marten-seemann
While working on my personal project, I noticed that when using the noise protocol, both peers are intermittently getting error messages when they try to connect to each other at the same time.
Link's project is a reproduction of that part. and I solved the problem by storing each other in the peer store and connecting streams without trying to connect directly at the moment.
The file below is a log.txt file included in the link, which shows that noise errors occur intermittently when a docker is used to create an isolated environment and try to connect by repeatedly creating/removing another peer with one peer.
If you want to reproduce the situation, you can reproduce it if you follow the sequence of test.sh .
thanks for reading and efforting.
Version Information