net: TestPacketConn flaky

[alberts]

TestPacketConn seems to be flaky. It has been for a while.

TestPacketConn typically runs in milliseconds, which makes the fact that the test
sometimes takes up to 8 seconds in some runs quite strange.

29 March:

--- FAIL: TestPacketConn-37 (8.73 seconds)
packetconn_test.go:101: PacketConn.WriteTo failed: write unixgram: i/o timeout

7 May:

go version devel +1638d2adb665 Mon May 06 17:34:17 2013 -0700 linux/386
=== RUN TestTCPClose-4
--- PASS: TestTCPClose-4 (0.01 seconds)
=== RUN TestPacketConn-4
--- FAIL: TestPacketConn-4 (1.57 seconds)
packetconn_test.go:112: PacketConn.ReadFrom failed: read unixgram /tmp/nettest093076375:
i/o timeout
=== RUN TestConnAndPacketConn-4
--- PASS: TestConnAndPacketConn-4 (0.06 seconds)

15 May:

go version devel +f51b5497fa4c Wed May 15 04:19:19 2013 +0800 linux/amd64
--- FAIL: TestPacketConn-23 (4.53 seconds)
packetconn_test.go:101: PacketConn.WriteTo failed: write unixgram: i/o timeout

[alberts]

Comment 1:

28 June:
freebsd/386 building fccd815ed3bb:
http://build.golang.org/log/fcded6bcf63bc341dbee9d955c184c977f4c6e7e
--- FAIL: TestPacketConn (0.64 seconds)
packetconn_test.go:101: PacketConn.WriteTo failed: write unixgram: i/o timeout
The read and write deadlines are 100 milliseconds in the test.
I'll look at adding some debug logging to this test at some point.

[rsc]

Comment 2:

Maybe we should just delete all these tests.

Labels changed: added priority-later, go1.2maybe, removed priority-triage.

Status changed to Accepted.

[rsc]

Comment 3:

Labels changed: added testing.

[rsc]

Comment 4:

Labels changed: added go1.3, removed go1.2maybe.

[rsc]

Comment 5:

Labels changed: added release-go1.3.

[rsc]

Comment 6:

Labels changed: removed go1.3.

[rsc]

Comment 7:

Labels changed: added repo-main.

[mikioh]

Comment 8:

Labels changed: added release-go1.4, removed release-go1.3.

[rsc]

Comment 9:

Labels changed: added release-none, removed release-go1.4.

[bradfitz]

I'm reopening this.

The new OpenBSD builder just hit this error:

http://build.golang.org/log/7b3870c037b6135cc3444895259560b2cd599479

--- FAIL: TestPacketConn (0.20s)
    packetconn_test.go:106: PacketConn.WriteTo failed: write unixgram: i/o timeout
FAIL
FAIL    net 9.254s

And it's 4 CPUs on GCE running a dedicated VM per build revision, so the previous arguments (overloaded, stale processes, too weak) no longer apply.

This is just flaky.

@mikioh, is this something you could look into?

[alexbrainman]

I can see it happens on windows-386 builder:

http://build.golang.org/log/9e1e98900aa5c5b40fefed8cd0c5e15c5f8d4252
http://build.golang.org/log/015b2f0d3d6aafd0eb5a994496495dd1eb254780

--- FAIL: TestPacketConn (0.38s)
packetconn_test.go:99: PacketConn.WriteTo failed: WSASendto udp 127.0.0.1:54017: i/o timeout
FAIL
FAIL net 11.023s

Alex

[mikioh]

With a few changes (not pushed to review process yet), we can see some socket and testing goroutine leaks:

net.TestWriteTimeout.func2(0xc20801e070, 0xc20800f000, 0x1000, 0x1000, 0xc208430300)
    /go/src/net/timeout_test.go:170 +0x98
created by net.TestWriteTimeout
    /go/src/net/timeout_test.go:173 +0x603

net.runDatagramPacketConnServer(0xc208402240, 0x2ad7b0, 0x3, 0x2b1c70, 0xb, 0xc2084303c0, 0xc208430420)
    /go/src/net/server_test.go:392 +0x60c
created by net.TestTimeoutUDP
    /go/src/net/timeout_test.go:247 +0xc9

5: {Cookie:562949953552384 Err:<nil> SocketErr:0}

{Family:1 Type:2 Protocol:0 Opened:17 Accepted:0 Connected:5 Closed:17}
{Family:2 Type:1 Protocol:0 Opened:300 Accepted:144 Connected:150 Closed:433}
{Family:1 Type:1 Protocol:0 Opened:11 Accepted:5 Connected:5 Closed:16}
{Family:30 Type:1 Protocol:0 Opened:13 Accepted:4 Connected:4 Closed:17}
{Family:2 Type:2 Protocol:0 Opened:43 Accepted:0 Connected:10 Closed:43}
{Family:30 Type:2 Protocol:0 Opened:8 Accepted:0 Connected:2 Closed:8}

Looks like we need to fix mock servers and timeout tests for avoiding some conflicts.

[mikioh]

Interference between test cases are removed by recent changes. Also we don't see this failure at tip, especially after aedee30. Closing.