Do not flush OutputStream to avoid contamination

[itugs]

Related with #997

BinaryJedisPubSub#process use Client#getObjectMultiBulkReply and it flush output stream before read from socket. This cause contamination of outputbuffer and redis close client connection due to Protocol Error. Using Client#getRawObjectMultiBulkReply will fix this.

JedisPubSub already use it. So if we change that test case with JedisPubSub it do not fail.
Maybe there was a missing.

Here is TCP dump log of test case provided #996

11:16:26.417775 IP localhost.61043 > localhost.6379: Flags [P.], seq 1:34, ack 1, win 12759, options [nop,nop,TS val 596420889 ecr 596420884], length 33
E..Un.@.@............s....o.... ..1..I.....
#...#...*2  <--- first subscribe
$9
SUBSCRIBE
$8
whatever

11:16:26.417800 IP localhost.6379 > localhost.61043: Flags [.], ack 34, win 12758, options [nop,nop,TS val 596420889 ecr 596420889], length 0
E..4(A@.@..............s... ..o...1..(.....
#...#...
11:16:26.417844 IP localhost.6379 > localhost.61043: Flags [P.], seq 1:38, ack 34, win 12758, options [nop,nop,TS val 596420889 ecr 596420889], length 37
E..Y.Z@.@..............s... ..o...1..M.....
#...#...*3
$9
subscribe
$8
whatever
:1

// omitted

11:16:26.436680 IP localhost.61043 > localhost.6379: Flags [P.], seq 220:222, ack 236, win 12752, options [nop,nop,TS val 596420907 ecr 596420907], length 2
E..6q4@.@............s....p.......1..*.....
#..+#..+*2     <----  partial tx due to contamination

11:16:26.436692 IP localhost.61043 > localhost.6379: Flags [P.], seq 222:264, ack 236, win 12752, options [nop,nop,TS val 596420907 ecr 596420907], length 42
E..^r.@.@............s....p.......1..R.....
#..+#..+*2     <----  normal tx
051a2777-c81f-455c-a607-4f48fe5a59ce

// omitted

11:16:26.436715 IP localhost.6379 > localhost.61043: Flags [P.], seq 236:283, ack 306, win 12749, options [nop,nop,TS val 596420907 ecr 596420907], length 47
E..c.|@.@..............s......p...1..W.....
#..+#..+-ERR Protocol error: invalid multibulk length   <--- Protocol error

// omitted

11:16:26.436786 IP localhost.6379 > localhost.61043: Flags [R], seq 2952692539, win 0, length 0
E..(+.@.@..............s...;....P.......     <--- RST

[marcosnils]

LGTM. Whenever another collaborator approves we should definitely merge.

@itugs thanks for the thorough investigation.

[HeartSaVioR]

I've just try tcpdump and found there's some issue but I don't think about difference of BinaryJedisPubSub and JedisPubSub.
Amazing works!

[xetorthio]

Yes. Good catch! Lets merge it!

[marcosnils]

@itugs did you run the tests before commiting?. Seems like something else broke because if this change

[HeartSaVioR]

@itugs BinaryJedisPubSub.proceed() and BinaryJedisPubSub.proceedWithPatterns() should execute flush() after calling psubscribe() / subscribe().

[itugs]

Oh, I'll follow that test failure. Sorry to bother you.

[itugs]

I revised and PublishSubscribeCommandsTest successed.
Some tests are still fail but I think it's due to env of CI.

[HeartSaVioR]

No problem. We also know random test failures, and we re-run tests.

[HeartSaVioR]

OK, test failure issue is gone.

[marcosnils]

Amazing! @HeartSaVioR. Can you merge?

[HeartSaVioR]

@marcosnils I'll merge it within 1 hour. :)

[HeartSaVioR]

Merged to master, 2.7, 2.8 respectively.

Thanks for your amazing work! @itugs

[itugs]

Thanks! I'm very exciting to be contributing to Jedis.