Nonblocking socket initialisation

[szymonsobczak]

This fixes #78.

[coveralls]

Coverage decreased (-0.94%) when pulling 08c85fb on futuresimple:nonblocking_socket into b4dc2be on bpot:master.

[bpot]

This looks great! I would like to have an integration test that reproduces the issue. Have you looked at writing one?

It would probably have to be linux only (b/c of iptables) but I think that's fine.

[szymonsobczak]

Hi @bpot,

I've added the integration test. It requires

• pfctl (BSD/OSX utility for controlling the packet filter)
• sudo permissions to run the pfctl
  It disables the new spec when sudo pfctl is not available. See https://github.com/bpot/poseidon/pull/79/files#diff-e252b1c2252b02985f13ae37a731c3c0R94

The spec fails when connection is using TCPSocket.new() (in my case it takes >30s to timeout) and passes with connect_with_timeout().

Some minor fixes were required to run the suite on OSX, like running OS commands differently and ensuring required dirs exist.

I've also worked a bit with sleeps, as the tests were unnecessarily slow and hard to work with. There is still room for improvement - most of the actions can be detected from watching kafka logs, so there should be no need for sleeps whatsoever.

BTW, we're using your library on production already for a couple months now and it looks stable, so maybe it's time to release v1.0.0?

[coveralls]

Coverage decreased (-0.43%) to 91.51% when pulling b67a106 on futuresimple:nonblocking_socket into b4dc2be on bpot:master.

[coveralls]

Coverage decreased (-0.84%) to 91.1% when pulling a6877d1 on futuresimple:nonblocking_socket into b4dc2be on bpot:master.

[marekstachura]

@szymonsobczak hey, the content of MessageToSend could be better... ;)

[mensfeld]

is this going to be merged?