net/http: apparent deadlocks in TestHandlerAbortRacesBodyRead since 2022-08-09

[bcmills]

#!watchflakes
post <- pkg == "net/http" && test == "TestHandlerAbortRacesBodyRead"

---

greplogs -l -e 'panic: test timed out .*(?:\n.*)*net/http\.\(\*Transport\)\.RoundTrip.*\n\t.+\nnet/http_test\.TestHandlerAbortRacesBodyRead'
2022-08-16T20:39:36-6a9c674/plan9-amd64-0intro
2022-08-09T16:41:32-25d74f3/darwin-amd64-12_0

Marking as release-blocker for Go 1.20 because darwin/amd64 is a first-class port and this failure mode seems to be new as of this cycle.

[mknyszek]

CC @neild

[neild]

Added watchflakes stanza to issue description.

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net/http" && test == "TestHandlerAbortRacesBodyRead"

2022-08-09 16:41 darwin-amd64-12_0 go@25d74f32 net/http.TestHandlerAbortRacesBodyRead (log)

panic: test timed out after 3m0s

runtime.gopark(0xc000269c20?, 0x0?, 0x20?, 0x47?, 0x100e77f?)
	/Users/gopher/workdir/go/src/runtime/proc.go:363 +0xd6 fp=0xc0005a8e38 sp=0xc0005a8e18 pc=0x103abb6
runtime.goparkunlock(...)
	/Users/gopher/workdir/go/src/runtime/proc.go:369
runtime.semacquire1(0xc000269c38, 0x0?, 0x1, 0x0)
	/Users/gopher/workdir/go/src/runtime/sema.go:150 +0x1fe fp=0xc0005a8ea0 sp=0xc0005a8e38 pc=0x104b99e
sync.runtime_Semacquire(0xc000240900?)
	/Users/gopher/workdir/go/src/runtime/sema.go:62 +0x25 fp=0xc0005a8ed0 sp=0xc0005a8ea0 pc=0x1067485
sync.(*WaitGroup).Wait(0x1595640?)
	/Users/gopher/workdir/go/src/sync/waitgroup.go:139 +0x52 fp=0xc0005a8ef8 sp=0xc0005a8ed0 pc=0x10743b2
net/http_test.TestHandlerAbortRacesBodyRead(0xc00057a680)
	/Users/gopher/workdir/go/src/net/http/transport_test.go:6567 +0x145 fp=0xc0005a8f70 sp=0xc0005a8ef8 pc=0x1402fa5
testing.tRunner(0xc00057a680, 0x150d300)

— watchflakes

[heschi]

@neild Any opinion on whether this represents a release blocker?

[neild]

I think this is occurring rarely enough that it's probably not new as of this cycle, and we just didn't notice any failures previously. So not a release blocker, although it does need fixing.

[heschi]

Thanks!

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net/http" && test == "TestHandlerAbortRacesBodyRead"

2023-02-06 04:29 freebsd-amd64-race go@def0be5e net/http.TestHandlerAbortRacesBodyRead (log)

--- FAIL: TestHandlerAbortRacesBodyRead (0.00s)
    --- FAIL: TestHandlerAbortRacesBodyRead/h1 (0.34s)
        testing.go:1446: race detected during execution of test
    --- FAIL: TestHandlerAbortRacesBodyRead/h2 (0.38s)
        testing.go:1446: race detected during execution of test

— watchflakes