On Sun, Mar 04, 2007 at 08:24:59PM +0000, Mike wrote:
Igor Sysoev wrote:
On Tue, Feb 20, 2007 at 01:36:10PM +0000, Chris McGrath wrote:
I'm using runit to supervise nginx, and to get it to work properly, I
had to set
daemon = off;
In the nginx.conf. The documentation says that this shouldn't be used
in production. I was wondering why this was and if it is safe to keep
running like this.
The "daemon off" and "master_process off" were initially created
for development only. However, some people started to run nginx under
daemontools and so on using "daemon off". Non-daemonized nginx runs
without problems except you can not do online upgrade.
But "master_process off" should be never used in production.
Is there any way that this can be changed in the future? Non-disruptive
upgrades to nginx are an awesome feature that we would love to exploit,
but we are also using Runit to handle the service (and Launchd is
handling Runit).
It's the technical issue. Runit fork()s and exec()utes nginx so it
becomes the parent of the nginx master process. But when nginx does
an online upgrade, the master process fork()s and exec()utes a new binary
too. When old master process exists, the new master process' parent
is init process (#1) and Runit can not control it.
The possible workaround is to use SO_REUSEPORT for listen sockets, but
I did not investigate it.
5 messages in ru.sysoev.nginxRe: runit and nginx
