Synchronize access to @client_threads in LogStash::Outputs::Tcp

[lmars]

From original PR elastic/logstash#1270

---

Logstash was consistently failing with the following error when a client connected to a TCP output:

NoMethodError: undefined method `write' for nil:NilClass
       register at /opt/logstash/lib/logstash/outputs/tcp.rb:92
           each at org/jruby/RubyArray.java:1613
       register at /opt/logstash/lib/logstash/outputs/tcp.rb:91
           call at org/jruby/RubyProc.java:271
         encode at /opt/logstash/lib/logstash/codecs/json.rb:45
        receive at /opt/logstash/lib/logstash/outputs/tcp.rb:143
         handle at /opt/logstash/lib/logstash/outputs/base.rb:86
     initialize at (eval):21
           call at org/jruby/RubyProc.java:271
         output at /opt/logstash/lib/logstash/pipeline.rb:266
   outputworker at /opt/logstash/lib/logstash/pipeline.rb:225
  start_outputs at /opt/logstash/lib/logstash/pipeline.rb:152

Logstash at the time was receiving a heavy stream of events, so this is likely the result of a race condition accessing @client_threads which is not thread safe, so this change synchronizes access to it across threads.

[elasticsearch-release]

Jenkins standing by to test this. If you aren't a maintainer, you can ignore this comment. Someone with commit access, please review this and clear it for Jenkins to run; then say 'jenkins, test it'.

[colinsurprenant]

old issue followup: sorry about dropping this, it is still valid. I'll do a few online comments, if you are inclined on following up pease do, otherwise we can take what you proposed, make some modifications and merge that while preserving your commit for authorship.

[colinsurprenant]

instead of doing a "long" lock, I'd suggest keeping the client_thread var usage and simply do a short lock to update the @client_threads ivar like that

@client_threads_lock.synchronize do
  @client_threads << client_thread
end