net/http: TestServerConnState is flaky

[bcmills]

dragonfly-amd64: https://build.golang.org/log/c79adae2958dc28dd62ca163328fd6dc526ea37f

2019/05/27 18:04:00 http: TLS handshake error from 127.0.0.1:53789: read tcp 127.0.0.1:59376->127.0.0.1:53789: use of closed network connection
2019/05/27 18:04:00 http: TLS handshake error from 127.0.0.1:50713: read tcp 127.0.0.1:59376->127.0.0.1:50713: use of closed network connection
2019/05/27 18:04:00 http: TLS handshake error from 127.0.0.1:56210: read tcp 127.0.0.1:59376->127.0.0.1:56210: use of closed network connection
2019/05/27 18:04:00 http: TLS handshake error from 127.0.0.1:53175: read tcp 127.0.0.1:59376->127.0.0.1:53175: use of closed network connection
2019/05/27 18:04:00 http: TLS handshake error from 127.0.0.1:59361: write tcp 127.0.0.1:59376->127.0.0.1:59361: use of closed network connection
--- FAIL: TestServerConnState (1.00s)
    serve_test.go:4256: Unexpected events.
        Got log:
        Conn 1: new active idle active 
        Conn 2: new active idle active closed 
        Conn 3: new active hijacked 
        Conn 4: new active hijacked 
        Conn 5: new closed 
        Conn 6: new active closed 
        Conn 7: new active idle closed 
        
           Want:
        Conn 1: new active idle active closed 
        Conn 2: new active idle active closed 
        Conn 3: new active hijacked 
        Conn 4: new active hijacked 
        Conn 5: new closed 
        Conn 6: new active closed 
        Conn 7: new active idle closed 
        
FAIL
FAIL	net/http	12.507s

CC @bradfitz

[bradfitz]

Maybe the test is wrong, but Dragonfly network tests in general are flaky (#29583 (comment)) so I'm not super eager to debug this until I've seen it fail on another OS.

[bcmills]

until I've seen it fail on another OS.

solaris-amd64-smartosbuildlet, with the same failure mode: https://build.golang.org/log/44bf73c2404dd6260a9d545b0e5f3a291e399b00

[bcmills]

dragonfly-amd64 again: https://build.golang.org/log/8f5ef2ad0de269bcefc4ac0a0bd47d19ec53c70c

[hbhasker]

just as an FYI. This also flakes under gVisor (https://github.com/google/gvisor). But the failure is pretty consistent. The states for connection 5 and 6 are swapped . This might totally be due to difference in implementation of gvisor network stack vs linux but I think given that this test is flaky on other OS as well it probably is depending on some implementation detail of linux.

I will try and debug to see if I can understand why this happens under gVisor as we do want to be linux compliant.

--- FAIL: TestServerConnState (0.52s)
serve_test.go:4250: Unexpected events.
Got log:
Conn 1: new active idle active closed
Conn 2: new active idle active closed
Conn 3: new active hijacked
Conn 4: new active hijacked
Conn 5: new active closed
Conn 6: new closed
Conn 7: new active idle closed

       Want:
    Conn 1: new active idle active closed
    Conn 2: new active idle active closed
    Conn 3: new active hijacked
    Conn 4: new active hijacked
    Conn 5: new closed
    Conn 6: new active closed
    Conn 7: new active idle closed