sync: TestWaitGroupMisuse2 is flaky

[bcmills]

--- FAIL: TestWaitGroupMisuse2 (4.31s)
    waitgroup_test.go:133: Should panic
    waitgroup_test.go:96: Unexpected panic: <nil>
FAIL
FAIL	sync	4.603s

2020-03-30T17:35:19-3ee782b/linux-s390x-ibm
2020-03-22T08:44:07-683c266/aix-ppc64
2020-03-22T00:10:27-36b815e/linux-s390x-ibm
2019-10-24T12:09:29-a42a396/solaris-amd64-oraclerel
2019-10-03T20:43:01-3ad3508/aix-ppc64
2019-09-24T04:16:35-ced2454/aix-ppc64
2019-09-16T11:56:15-49e7c76/solaris-amd64-oraclerel
2019-09-12T14:09:30-c20eb96/solaris-amd64-oraclerel
2019-09-10T14:25:46-f668573/solaris-amd64-oraclerel
2019-08-28T10:43:11-5fb74fc/solaris-amd64-oraclerel
2019-08-25T16:47:03-89fb80f/solaris-amd64-oraclerel
2019-06-06T19:56:56-d32ec38/aix-ppc64
2019-06-05T23:57:39-e9073ef/aix-ppc64
2019-05-30T22:39:04-3cee550/aix-ppc64
2019-05-30T17:59:57-220552f/aix-ppc64
2019-05-30T15:38:43-16796af/aix-ppc64
2019-05-30T12:00:59-a0c96a9/aix-ppc64
2019-05-29T21:31:37-913d290/aix-ppc64
2019-05-29T01:10:13-200176e/aix-ppc64
2019-04-30T16:59:13-f686a28/aix-ppc64

See previously #22944, #20072, #11443.

(CC @dvyukov @bradfitz @aclements @ianlancetaylor)

[bcmills]

2020-03-31T02:37:05-71d4774/linux-s390x-ibm

[bcmills]

2020-04-16T05:29:18-b89f4c6/linux-s390x-ibm
2020-04-03T13:42:33-6117275/aix-ppc64

[mundaym]

I propose that we just delete this test. It has become a time sink and I don't think there is any way to make it truly reliable. It is testing a race condition between two pieces of code that is fundamentally impossible to guarantee, especially on a VM where the underlying hardware is oversubscribed. The error condition detection that it is testing for is optional - losing it would not affect correct programs - and overlaps with the checks that -race mode does.

[bcmills]

Seems to have gotten less flaky in 2021 — it had been almost a year since the last flake on record!
(But still flaky.)

2021-08-04T15:26:45-6e73886/windows-arm64-10
2020-10-02T08:52:16-15bf061/linux-s390x-ibm
2020-09-12T19:42:29-b3ef90e/linux-s390x-ibm
2020-07-30T15:46:39-6b4dcf1/linux-s390x-ibm
2020-07-27T16:04:55-19a932c/linux-s390x-ibm
2020-07-22T21:31:19-71218db/linux-s390x-ibm
2020-07-10T11:32:36-3a43226/linux-s390x-ibm
2020-07-08T22:16:24-4f2a2d7/linux-s390x-ibm
2020-07-02T22:08:11-5de90d3/linux-s390x-ibm
2020-07-01T15:29:29-a4ba411/linux-s390x-ibm
2020-07-01T14:41:56-7799756/linux-s390x-ibm
2020-06-30T20:14:48-96e8366/linux-s390x-ibm
2020-06-30T18:40:39-5779bb4/linux-s390x-ibm
2020-06-29T21:33:23-8c52173/linux-s390x-ibm
2020-06-24T22:40:11-b465202/linux-s390x-ibm
2020-06-24T20:16:49-72019cf/linux-s390x-ibm
2020-06-19T16:41:13-27a0012/linux-s390x-ibm
2020-06-19T16:14:19-b424a31/linux-s390x-ibm
2020-06-19T02:11:35-3dec253/linux-s390x-ibm
2020-06-18T01:12:21-84baf41/linux-s390x-ibm
2020-06-17T19:48:45-8b98498/linux-s390x-ibm
2020-06-17T00:18:55-e25aa4f/linux-s390x-ibm
2020-06-10T21:39:03-6aadfcd/linux-s390x-ibm
2020-05-13T22:15:39-a88c26e/linux-s390x-ibm
2020-05-07T20:18:02-3321303/linux-s390x-ibm
2020-05-06T13:55:00-4daf871/aix-ppc64
2020-04-29T19:02:32-5a550b6/linux-s390x-ibm
2020-04-24T18:49:16-512277d/linux-s390x-ibm
2020-04-21T20:49:34-2a2423b/linux-s390x-ibm
2020-04-20T22:38:59-0cffc95/linux-s390x-ibm
2020-04-20T20:33:36-40a144b/linux-s390x-ibm
2020-04-17T22:12:02-4d9ecde/linux-s390x-ibm

[gopherbot]

Change https://golang.org/cl/340249 mentions this issue: sync: remove TestWaitGroupMisuse2 and TestWaitGroupMisuse3

[bcmills]

Looks like the flakes are for some reason quite bad on windows-arm64:

greplogs --dashboard -md -l -e 'FAIL: TestWaitGroupMisuse2' --since=2021-08-05

2021-10-12T20:20:41-d032b2b/windows-arm64-10
2021-10-08T00:18:29-7cef831/windows-arm64-10
2021-09-27T20:42:34-301f6c8/windows-arm64-10
2021-09-24T01:18:01-0f1159b/windows-arm64-10
2021-09-15T17:32:52-6196979/windows-arm64-10
2021-09-10T17:13:40-1bf2cd1/windows-arm64-10
2021-09-01T17:25:30-8f397bc/windows-arm64-10
2021-08-24T18:31:34-c2f96e6/windows-arm64-10
2021-08-23T13:09:11-7a6d64f/windows-arm64-10
2021-08-16T14:50:20-ea8298e/windows-arm64-10
2021-08-16T13:38:29-0b61dc4/windows-arm64-10

[dmitshur]

@gopherbot Please backport to Go 1.17 as a test-only fix. The TestWaitGroupMisuse2 test had a flaky failure during the 1.17.9 release, which requires tests to be re-run and slows the process down.