Running locally found EpollSocket::Accept was just blocking forever and never suspending the fiber - looks like FD_CLOEXEC is being set with F_SETFL instead of F_SETFD which is then breaking O_NONBLOCK
@romange this is part of the cause of the deadlock you mentioned (if Accept never releases the fiber the accept loop never finishes so AcceptServer::Stop blocks in the test tear down) - though theres then another issue where shutdown isn't waking up the accept fiber so it still blocks Stop (will look at that separately)
Running locally found
EpollSocket::Acceptwas just blocking forever and never suspending the fiber - looks likeFD_CLOEXECis being set withF_SETFLinstead ofF_SETFDwhich is then breakingO_NONBLOCK@romange this is part of the cause of the deadlock you mentioned (if
Acceptnever releases the fiber the accept loop never finishes soAcceptServer::Stopblocks in the test tear down) - though theres then another issue whereshutdownisn't waking up theacceptfiber so it still blocksStop(will look at that separately)