Yet another flaky test (or a flaky function!) causing rare failures.

[johnsonjh]

peer: TestPeerConnection test:

--- FAIL: TestPeerConnection (0.00s)
    peer_test.go:342: Running 2 tests
    peer_test.go:211: testPeer: wrong LastSend - got 2020-10-31 20:16:07 +0000 GMT, want 1970-01-01 00:00:00 +0000 GMT

cd peer && go test -count=10000 . | grep 'FAIL: TestPeerConnection' | wc -l`
90

So, the test passes 99.1% of the time.

Will look into it and see what the source of this failure may be.

Originally posted by @johnsonjh in #127 (comment)

[johnsonjh]

Test improvements in some of these 42 PR's I currently have pending (and I'm not even sure which ones!) have helped, but not eliminated this particular error.

It was always about ~90 [+/- 15] failures with develop branch as of 75e4cd861bc3f89dd3dce1aa8d8529120a1c3166. After applying all outstanding PR's, it's now ~25 [+/- 5] out of 10,000 test runs, consistently. The results are the same with Go 1.16-pre and Go 1.15.3.

[johnsonjh]

e99926c#diff-fe29af682b09b1e9bf8169d7c1d2d1f971e70f8f48e7ad4dbb1f624ded93d2f3 is the commit that solved most of the issues - this whole thing is done in a way that's inherently race-prone, but that's seemingly not only the issue here - I'll see if I can find a solution.

[johnsonjh]

It's a hackjob if workaround, but so is a lot of this code:

» go test -count=25000 .
ok  	github.com/pkt-cash/pktd/peer	41.565s

I'll push the commit into PR #161 to fix it.

[johnsonjh]

Added the commit to PR #161, so we can close out #128 if it's merged.

[johnsonjh]

Pretty sure the actual root cause is the race at

github.com/pkt-cash/pktd/addrmgr.(*KnownAddress).LastAttempt()
github.com/pkt-cash/pktd/addrmgr/knownaddress.go:34 +0x5d0

as noted at #170

[johnsonjh]

@cjdelisle

A proper fix for this test is now pending as PR #212 - if that is accepted and merged, then this issue can be closed.