internal/trace/v2: TestTraceStressStartStop/Default failures

[gopherbot]

#!watchflakes
default <- pkg == "internal/trace/v2" && test == "TestTraceStressStartStop/Default"

Issue created automatically to collect these failures.

Example (log):

=== RUN   TestTraceStressStartStop/Default
    reader_test.go:112: unexpected error while reading the trace: broken trace: failed to advance: frontier:
        [gen=5]
        
        M 1098478 [ProcStart time=61707444740480 p=3 p_seq=15]
        
        G 12 [status=Waiting seq=0 (gen=5)]
        G 13 [status=Waiting seq=0 (gen=5)]
        G 11 [status=Waiting seq=0 (gen=5)]
        G 43 [status=Waiting seq=1 (gen=5)]
...
        String id=129
        	data="net.(*conn).Write"
        String id=130
        	data="sync.(*WaitGroup).Wait"
        String id=131
        	data="runtime.block"
        String id=132
        	data="/home/swarming/.swarming/w/ir/x/w/goroot/src/runtime/select.go"
        
--- FAIL: TestTraceStressStartStop/Default (0.89s)

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "internal/trace/v2" && test == "TestTraceStressStartStop/Default"

2024-01-26 17:53 gotip-linux-386-longtest go@8081c084 internal/trace/v2.TestTraceStressStartStop/Default (log)

=== RUN   TestTraceStressStartStop/Default
    reader_test.go:112: unexpected error while reading the trace: broken trace: failed to advance: frontier:
        [gen=5]
        
        M 1098478 [ProcStart time=61707444740480 p=3 p_seq=15]
        
        G 12 [status=Waiting seq=0 (gen=5)]
        G 13 [status=Waiting seq=0 (gen=5)]
        G 11 [status=Waiting seq=0 (gen=5)]
        G 43 [status=Waiting seq=1 (gen=5)]
...
        String id=129
        	data="net.(*conn).Write"
        String id=130
        	data="sync.(*WaitGroup).Wait"
        String id=131
        	data="runtime.block"
        String id=132
        	data="/home/swarming/.swarming/w/ir/x/w/goroot/src/runtime/select.go"
        
--- FAIL: TestTraceStressStartStop/Default (0.89s)

— watchflakes

[bcmills]

(Note that linux/386 is currently a first class port.)

[mknyszek]

I see the problem here. P states aren't always correctly cleaned up for future generations when the trace stops. I'll send a CL for this.

[gopherbot]

Change https://go.dev/cl/567077 mentions this issue: runtime: clean up dead P trace state when disabling tracing too