down_retry_delay default is too low

[mrsimo]

Hi there! We were running some tests to figure out how much memcached downtime could handle our app, and since that downtime, we've gotten a bit familiar with down_retry_delay.

The default value for this value is 1s, but coupled with a socket_max_failures of 2, and a socket_timeout of 0.5, I think that makes the default for down_retry_delay not very good.

If the memcached server is down in a way that we exhaust the socket_timeout, considering the retries it does, down_retry_delay doesn't ever do anything.

Stopping one of our memcached servers caused all memcached queries to it to take 1s, which slowed our unicorns to a point where we ran out of them.

We changed the value of down_retry_delay to 120s, and now memcached servers going down doesn't have any visible effects.

I would suggest changing the default of this value to something bigger.

[petergoldstein]

Fixed by #650