runtime: TestSelectFairness failures with "unfair select"

[bcmills]

#!watchflakes
post <- pkg == "runtime" && test == "TestSelectFairness"

greplogs --dashboard -md -l -e 'FAIL: TestSelectFairness .*(?:\n .*)*unfair select'

2022-04-20T16:05:16-338a817/illumos-amd64
2022-03-23T19:31:43-2792a41/netbsd-386-9_0
2020-08-19T00:10:22-98a0071/js-wasm

@golang/runtime: the comment for this test suggests that it has a confidence level of “six nines”, which would be one spurious failure per ~1M runs. Is it still in line with that target?

See previously #34808, #22047.

[mengzhuo]

The "six nines" requires less than 244 out of 10000 trials however the fail logs shows 5245,5253,5246.

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "runtime" && test == "TestSelectFairness"

2022-11-18 17:59 linux-386-buster go@b2faff18 runtime.TestSelectFairness (log)

--- FAIL: TestSelectFairness (0.01s)
    chan_test.go:488: 5256 4744 0.5256 0.025599999999999956
    chan_test.go:490: unfair select: in 10000 trials, results were 5256, 4744

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "runtime" && test == "TestSelectFairness"

2023-07-31 17:13 netbsd-386-9_3 go@4a14d9c9 runtime.TestSelectFairness (log)

--- FAIL: TestSelectFairness (0.02s)
    chan_test.go:488: 4742 5258 0.4742 0.02579999999999999
    chan_test.go:490: unfair select: in 10000 trials, results were 4742, 5258

— watchflakes

[randall77]

This last one looks to be about a 1 in a million chance of being that unbalanced (258 extra heads above the expected 5000 out of 10000 trials).
Maybe there is an issue here? But I'm inclined to just soften the test. I don't think the world would end if selects weren't perfectly random (60/40, say).

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "runtime" && test == "TestSelectFairness"

2023-10-12 20:11 darwin-amd64-10_15 go@15a274b6 runtime.TestSelectFairness (log)

--- FAIL: TestSelectFairness (0.01s)
    chan_test.go:488: 5247 4753 0.5247 0.024700000000000055
    chan_test.go:490: unfair select: in 10000 trials, results were 5247, 4753

— watchflakes

[gopherbot]

Change https://go.dev/cl/538815 mentions this issue: runtime: make select fairness test less picky