os/signal: TestNohup flaky

[bcmills]

Seen on the solaris-amd64-oraclerel builder in https://build.golang.org/log/d5185d101eddc0f5a05103f11b74caf24421bf46:

--- FAIL: TestNohup (1.25s)
    signal_test.go:329: ran test with -send_uncaught_sighup=1 and it succeeded: expected failure.
        Output:
        PASS
FAIL
FAIL	os/signal	3.575s

See previously #8682, #5526.

CC @ianlancetaylor

[bcmills]

Same symptom on linux-ppc64le-power9osu: https://build.golang.org/log/ae981484d0f765e824066be33e4fb24537b07db5

[ianlancetaylor]

Note that the case in #33174 (comment) is not the same symptom, but rather the reverse:

--- FAIL: TestNohup (1.62s)
    signal_test.go:354: ran test with -send_uncaught_sighup=1 under nohup and it failed: expected success.
        Error: signal: user defined signal 1
        Output:

[ianlancetaylor]

2017-11-29T16:27:42-be67e26/linux-ppc64le-buildlet
2018-11-16T21:11:27-6f5be37/dragonfly-amd64
2019-07-17T22:57:03-4814e7d/solaris-amd64-oraclerel
2019-08-29T09:26:02-cb325fe/solaris-amd64-oraclerel
2019-09-06T21:22:26-8a8cf5b/solaris-amd64-oraclerel
2019-09-09T23:04:30-a3a1bdf/solaris-amd64-oraclerel
2019-09-11T03:32:35-904f046/linux-ppc64le-power9osu
2019-09-11T23:01:45-e7e2b1c/linux-arm64-packet

[ianlancetaylor]

I think all these cases could be explained if there is an unusually long time, more than 100 milliseconds, between calling syscall.Kill and the time the program actually receives the signal.

[gopherbot]

Change https://golang.org/cl/194879 mentions this issue: os/signal: split up sleeps waiting for signal

[bcmills]

Still flaky, unfortunately:
2020-02-20T19:12:11-75ea964/solaris-amd64-oraclerel

[bcmills]

2020-03-25T01:11:15-f4fe891/solaris-amd64-oraclerel