Connection handling improvements v2

[arohter]

More robust zk client error handling: support progressively more severe retry logic.
Automatic fork and InheritedError handling:
Lazy reconnects reduce zookeeper and redis server load.
Reduces surface for kernel mutex deadlocks in forking env.
Trace connection debug logging.

This is redo of #66

[arohter]

FYI the use of the zk fork hook ( https://github.com/fanhattan/redis_failover/compare/conn_handling_v2_nofork...conn_handling_v2_forkhook ) still causes immediate deadlocks in our env, despite running latest versions of zk/zookeeper.

[eric]

Are you able to get a ruby thread dump of all threads in one of these deadlocks?

On Mon, Mar 17, 2014 at 12:52 AM, arohter notifications@github.com
wrote:

FYI the use of the zk fork hook ( https://github.com/fanhattan/redis_failover/compare/conn_handling_v2_nofork...conn_handling_v2_forkhook ) still causes immediate deadlocks in our env, despite running latest versions of zk/zookeeper.

Reply to this email directly or view it on GitHub:
#73 (comment)

[arohter]

Closing this PR as obsolete.
We've been unable to fix the zk deadlock issue, and have since moved to using https://github.com/coreos/etcd as the redis_failover backend in production. We've also improved the election logic and fixed various edge cases/failures that we've run into in our high-load use. New PR coming soon!