Occasionally stale server selection about redis_failover

[ryanlecompte]

Hmm, looking at the code I don't see how this could be a problem. The basic idea is that we should always #pop as many times as #client_for is called. Is there a particular redis call that's causing this for you? Is it happening in a simple call like #get or #set or in more complicated/nested calls like in #multi?

[netshade]

Loops on #incr, that's the only command. The client is initialized with :master_only => true . Setup was three node managers, one zookeeper, three instances of redis. Test script looked like:

loop do
 expected += 1
 actual = $client.incr("key")
 if expected != actual
  abort
  end
end

The comparison stuff worked as expected, but in the course of running that script while killing random servers, the client logger would report it had received the new master, but would immediately thereafter attempt to connect to the old master and write, either causing connection errors or read-only errors.

( Ruby 1.9.3 )