[patch] Retry timeout should be allowed to be zero

[Tim Starling]

Immediate reconnection after a failure is a reasonable policy, especially in cases where there is only one memcached server. We use twemproxy, so our PHP clients typically have only one server configured. So, libmemcached should allow MEMCACHED_BEHAVIOR_RETRY_TIMEOUT to be set to zero. This is pretty straightforward -- a patch is attached, tested against libmemcached 1.0.17 with current PECL memcached.

Since the retry timeout is causing tens of failures per second for us, root cause unknown, we are planning on deploying this patch immediately as a workaround. See https://bugzilla.wikimedia.org/show_bug.cgi?id=56882

The patch simply wraps the section of memcached_mark_server_for_timeout() which sets next_retry with

    if (server->root->retry_timeout != 0)

and allows it to be set to zero in behavior.cc

[Massive Media (493pocbrcycmdw7yksonho9o2qzz-o18bz-d18ecat4t1b76tkfi3vttrkfngli)]

You can set it to -1, it should work too

[Brian Aker (brianaker)]

Looking at this.

[Massive Media (493pocbrcycmdw7yksonho9o2qzz-o18bz-d18ecat4t1b76tkfi3vttrkfngli)]

Please also check https://answers.launchpad.net/libmemcached/+question/239497 , it's related to it and we're having issues with this as well.