stressTestWithLittleOutput fails nondeterministically with EAGAIN

[jakepetroules]

✘ Test stressTestWithLittleOutput() recorded an issue at IntegrationTests.swift:1475:6: Caught error: Failed to spawn the new process with underlying error: UnderlyingError(rawValue: 11)
✘ Test stressTestWithLittleOutput() failed after 0.031 seconds with 1 issue.

This might be from execve returning EAGAIN due to exceeding system resource limits, but ulimit claims the max user process count in the GitHub Actions CI is unlimited, so...

[iCharlesHu]

What should we do with EAGAIN? Should subprocess simply try again?

[jakepetroules]

Not sure, as I didn't confirm exactly which syscall the error came from. I don't think EAGAIN is necessarily the sort of thing you retry in a loop like EINTR, though.

[weissi]

This is a bug in the async I/O implementation. EAGAIN means you need to (re-)subscribe to notifications from epoll/kqueue/... as there are currently no bytes available to be read/written. The SwiftNIO codebase (which we'd ideally just use for a robust I/O manager) contains all the necessary code, in the NIO codebase it's typically under .wouldBlock (because EWOULDBLOCK == EAGAIN and "would block" is a better characterisation than "again" IMHO)