dial attempt failed: EOF when go test in macOS

[mhchia]

I run the command in both macOS High Sierra(10.13.6) and ubuntu 16.04 and get this issue only in my macOS. I removed the entire $GOPATH/src, and performed go get again. This issue is still there.

$ gx-go rw
$ go test -v
...
=== RUN   TestSparseGossipsub
--- PASS: TestSparseGossipsub (3.00s)
=== RUN   TestDenseGossipsub
--- FAIL: TestDenseGossipsub (0.91s)
	floodsub_test.go:54: dial attempt failed: <peer.ID eQ7kYF> --> <peer.ID SwqzNr> dial attempt failed: EOF
=== RUN   TestGossipsubFanout
--- FAIL: TestGossipsubFanout (1.01s)
	floodsub_test.go:54: dial attempt failed: <peer.ID dMgtPZ> --> <peer.ID SCdsq4> dial attempt failed: EOF
=== RUN   TestGossipsubFanoutMaintenance
--- FAIL: TestGossipsubFanoutMaintenance (0.95s)
	floodsub_test.go:54: dial attempt failed: <peer.ID f8X5mz> --> <peer.ID edj6pD> dial attempt failed: EOF
=== RUN   TestGossipsubFanoutExpiry
--- PASS: TestGossipsubFanoutExpiry (4.53s)
=== RUN   TestGossipsubGossip
--- FAIL: TestGossipsubGossip (1.02s)
	floodsub_test.go:54: dial attempt failed: <peer.ID Y2LATb> --> <peer.ID cmZNWD> dial attempt failed: EOF
=== RUN   TestGossipsubGossipPiggyback
--- FAIL: TestGossipsubGossipPiggyback (1.06s)
	floodsub_test.go:54: dial attempt failed: <peer.ID TixfJG> --> <peer.ID QkahvD> dial attempt failed: EOF
...

After debugging for a while, I still don't have any idea about this issue. I will really appreciate if there is any direction to solve it.

[vyzo]

These are misdials lower in the stack; it's probably stale peers and port reuse that's causing this.

[mhchia]

@vyzo Thanks a lot for the direction!

it's probably stale peers and port reuse that's causing this.

Sorry, I don't understand why that would happen. IIUC, it seems all of the tests are defer cancel(), so every host should be closed accordingly? It is quite interesting and I would like to dig more into it if it is worth solving:)

[vyzo]

Well, Host.Close() cancels a context, which may take time to take effect.

[mhchia]

According to this, is it possibly helping if we change defer cancel() to something like

defer func() {
    cancel()
    time.Sleep(time.Millisecond * 100)
}()

in every test? I did it but it seems no effect.

[Stebalien]

Could be libp2p/go-libp2p-swarm#79.

[vyzo]

not a pubsub bug; midials lower in the stack.

[Stebalien]

This is a test failure in pubsub and is therefore a bug in the pubsub tests.