On Thu, Sep 22, 2016, at 01:29 AM, Philipp Gesang wrote: > Minutes again, because IMO a fallback between different units of > measure would be confusing.
Yeah, I was thinking the same thing. :) The updated patch looks good to me. So it looks like what we're getting from this is two things: * making the "timeout" limit apply to idling clients * optionally configuring a separate, longer limit specifically for idling clients The former restores old behaviour that looks like it was accidentally lost during the conversion to use a socket rather than signals for idled IPC. The latter then allows administrators to be more lenient with clients that are idling, even if they aren't respecting the restart-after-29-minutes convention. Both of those seem good to me. And they will both also protect Cyrus from the possibility of idling connections dropping out and never being cleaned up properly (though if/when/how this occurs is still subject to speculation). Cheers, ellie