Kevin Yank <[email protected]> wrote: > Regarding zero-downtime deploys: > > On 12 Mar 2015, at 5:45 pm, Eric Wong <[email protected]> wrote: > > > Best bet would be to run with double the workers temporarily unless > > you're too low on memory (and swapping) or backend (DB) connections or > > any other resource. > > I’d like to take this approach as I do have enough memory to spare. > How do you usually implement this? Any good write-ups or sample > configs you can point me to?
Only send SIGUSR2 to the master, leaving you with two masters and two sets of workers. Skip (automated) sending of SIGTTOU signals to lower worker count to the old master. Eventually, you'll decide to send SIGQUIT to the old master to stop it (or the new one, if you decide the new code is broken). You can still combine this with SIGWINCH (or SIGTTOU) to stop traffic flow to the old master, too. Thanks for following up on your logrotate/eye issue, by the way.
