You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Some context: a few days ago I asked on Discord if hyper ever saw integration::http2_parallel_10 test failing in ci on Mac. But you mention you don't run that test on Mac because of weirdness there. I did tried to figure out what went wrong (as I develop on mac and was worried this might be part of a bigger problem), and found two issues. Solving one is already enough, but both solving both is probably ideal.
Problem one: calling shutdown twice on socket on mac/bsd results in ENOTCONN. Calling shutdown here happens on automatically and this is not really a Hyper problem, but more of a Tokio problem. For this I responded here and will try to work out a solution with the Tokio team
Problem two: a Hyper H2 problem. Depending on the exact order of shutdown this happens:
Receive GoAway from client
Client starts to close tcp
Server reads from frame but receives pending (vs ready(None) on successful shutdown)
Server sends GoAway to client instead of just closing
Client responds with reset to this unexpected packet
Macos closes socket completely now
Server does shutdown on socket resulting in problem 1
Biggest issue here is problem 1 and if that one is fixed this is honestly not really a problem anymore. But still wanted to open this issue to track this problem and to ask the question: do we need/want to solve problem 2?
Some context: a few days ago I asked on Discord if hyper ever saw
integration::http2_parallel_10
test failing in ci on Mac. But you mention you don't run that test on Mac because of weirdness there. I did tried to figure out what went wrong (as I develop on mac and was worried this might be part of a bigger problem), and found two issues. Solving one is already enough, but both solving both is probably ideal.Problem one: calling shutdown twice on socket on mac/bsd results in ENOTCONN. Calling shutdown here happens on automatically and this is not really a Hyper problem, but more of a Tokio problem. For this I responded here and will try to work out a solution with the Tokio team
Problem two: a Hyper H2 problem. Depending on the exact order of shutdown this happens:
Biggest issue here is problem 1 and if that one is fixed this is honestly not really a problem anymore. But still wanted to open this issue to track this problem and to ask the question: do we need/want to solve problem 2?
Some relevant log lines of this failing:
Related issues:
The text was updated successfully, but these errors were encountered: