runtime: TestSpuriousWakeupsNeverHangSemasleep failure with file already closed

[bcmills]

greplogs -l -e 'FAIL: TestSpuriousWakeupsNeverHangSemasleep .*(?:\n .*)* file already closed' --since=2022-01-01
2022-05-04T20:06:32-fd6ef06/solaris-amd64-oraclerel

It isn't clear to me whether this is a bug in the Go runtime or standard library, a bug in the test, or a bug in the Solaris platform. Leaving on the backlog to collect more information.

(CC @golang/runtime @golang/solaris @rorth)

[bcmills]

This builder previously had a high failure rate for this test with a different failure mode; see #49727.

[aclements]

I just got this failure on a linux-amd64-nonunified trybot: https://storage.googleapis.com/go-build-log/00af0613/linux-amd64-nounified_f492a55d.log

Re-running greplogs, it looks like it also happened on FreeBSD: 2022-06-02T17:17:19-aae0bef/freebsd-arm-paulzhol

[prattmic]

2022-07-18T15:58:37-c0c1bbd/solaris-amd64-oraclerel:

[prattmic]

This is a bug in the test. Arguably it could also be considered a weakness in either io.ReadAll or os/exec.Cmd.StdoutPipe.

The problem is that the test calls cmd.Wait in a goroutine with no synchronization with io.ReadAll. This sets up a race condition, with two possible outcomes.

Good case:

1. Child closes write end of pipe.
2. io.ReadAll gets EOF from read() due to closed write end, returns no error.
3. cmd.Wait closes read end.

Bad case:

1. Child closes write end of pipe.
2. Child exits.
3. cmd.Wait closes read end.
4. io.ReadAll gets fs.ErrClosed from read() because the read end is closed, returns fs.ErrClosed.

FWIW, note that the child sleeps for 1s between (1) and (2), so (4) must be significantly delayed to lose the race.

This is rather annoying to work around in the context of the test. The best approach may just be to ignore fs.ErrClosed from io.ReadAll, or use a manually-managed os.Pipe instead of cmd.StdoutPipe.

[gopherbot]

Change https://go.dev/cl/420597 mentions this issue: runtime: don't race cmd.Wait and cmd.StdoutPipe read

[bcmills]

greplogs -l -e 'FAIL: TestSpuriousWakeupsNeverHangSemasleep .*(?:\n .*)* file already closed' --since=2022-08-01
2022-08-11T20:16:32-b648591/linux-amd64-race
2022-08-08T17:29:13-9903ab5/linux-amd64-nounified

Just noticed that the fix CL was waiting on a review from me, so +2'd. 😅