2010/8/20 Dennis J. <djacobfeuerborn@gmail.com>:
>  On 08/20/2010 10:15 AM, Jérôme Loyet wrote:
>>
>> 2010/8/20 Thomas Fritz<fritztho@gmail.com>:
>>>
>>> 2010/8/20 Jérôme Loyet<ml@fatbsd.com>
>>>>
>>>> Hi,
>>>>
>>>> I was thinking about all that.
>>>>
>>>> There is maybe a feature which can help in this case.
>>>>
>>>> what if there is a "pid" directive per pool. In this case FPM will
>>>> write in this PID file each process PIDs dedicated to a pool. So that,
>>>> if you want to restart a pool, you'll be able to do it using the kill
>>>> command.
>>>>
>>>> [global]
>>>> pid=/var/run/php-fpm.pid ; master process PID
>>>> ...
>>>> [www.foo.bar]
>>>> pid=/home/www/www.foo.bar/logs/php-fpm.pid
>>>> ...
>>>> [www.domain.com]
>>>> pid=/home/www/www.domain.com/logs/php-fpm.pid
>>>> ...
>>>>
>>>> In /var/run/php-fpm.pid, you'll have only the master process PID.
>>>> You'll be able to stop/reload everything as usual. Nothing change
>>>> here.
>>>
>>> As i understand you: There is a php-fpm reload which only reloads the
>>> configuration files - does a graceful restart?
>>
>> yes. man php-fpm tells you about this:
>>
>> Once started, php-fpm then responds to several POSIX signals:
>>
>>  SIGINT,SIGTERM      immediate termination
>>  SIGQUIT             graceful stop
>>  SIGUSR1             re-open log file
>>  SIGUSR2             graceful reload of all workers + reload of fpm
>> conf/binary
>>
>>>> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
>>>> of the processes dedicated to the pool www.foo.bar using :
>>>>
>>>> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>>>
>>> Which will only kill all child processes of that specific pool but not
>>> the
>>> "master process" of that pool itself, right?
>>
>> yes
>>
>>>> The same for www.domain.com by using :
>>>> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>>>>
>>>> This will allow processes to restart a pool, to reload php.ini content
>>>> but it WON'T reload any FPM related conf.
>>>
>>> As i understand, it WILL reload pool specific php.ini, ENV and other
>>> configuration values!?
>>
>> nope. It will reload the php.ini content (/usr/local/php/lib/php.ini
>> for example) but it WON'T reload anything present on the php-fpm.conf
>> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
>> have to reload the master process which will restart all pool
>> processes. There is nothing to do about this.
>>
>>> If that works that easy and the fpm process recreates its child process
>>> after killing without problems - GREAT!
>>
>> it already works that easy. kill by hand all the FPM child processes
>> and you'll see that they'll be recreated by the master process. I'm
>> just proposing to note thoses PID in one file per pool to make
>> operations easier. That's all.
>>
> This sounds racy. Are you going to update the pid file every time a process
> dies and a replacement is created? What if I try to do the "kill
> <worker-pid-file>" at the wrong moment? If I execute this after a new
> process is created but before the pid file has been updated then this
> process potentially still runs a different config.
> This could lead to very subtle issues that are almost impossible to
> reproduce and track down.
>
> A better way would be to use the same approach but implement this killing of
> processes in php-fpm itself. That way php-fpm could suspend the creation of
> new process, kill the workers and then enable the creation of new processes
> again (similar to what preempt_disable() and preempt_enable() do in the
> kernel). e.g.:
>
> disable_worker_forking();
> kill_pool_workers();
> reload_config();
> enable_worker_forking();

You're right. With the patch I just proposed (changing the process
title), you can achieve this easely without doing anything else:

root@wild# ps a | grep fpm
26028 pts/3 S+ 0:00 php-fpm: master process
(/usr/local/php-trunk/etc/php-fpm.conf)
26030 pts/3 S+ 0:00 php-fpm: pool www_direct
26031 pts/3 S+ 0:00 php-fpm: pool www_direct
26032 pts/3 S+ 0:00 php-fpm: pool www_direct
26067 pts/3 S+ 0:00 php-fpm: pool www_chroot
26068 pts/3 S+ 0:00 php-fpm: pool www_chroot
26069 pts/3 S+ 0:00 php-fpm: pool www_chroot

# gracefull restart a pool
root@wild# pkill -QUIT -f "php-fpm: pool www_chroot"

# hard restart a pool
root@wild# pkill -f "php-fpm: pool www_chroot"

Maybe for those who need this, a hack can be done on the FPM init.d
script as proposed earlier in this thread.

++ Jerome
>
> Regards,
>  Dennis
>

+1 for the titles, but I would really like to be able to control
separate pools via fpm itself (reload configs, restarting pools in
case of fs hangs: nfs for example, killing the processes is not really
a solution, starting new processes and killing current ones on
background is the proper solution I think, and this should be done by
fpm itself). I think that the easiest way to communicate all this with
fpm would be via a simple fifo: here is an example [1] that I played
with one night some time ago (no fpm). Something like "start
www_chroot", "stop www_chroot", "restart www_chroot", "reload
www_chroot" would be nice to have :)

1: http://www.xaxo.eu/en/patches_php-ctrlfifo.html

On 20 Авг, 19:48, Jérôme Loyet <m...@fatbsd.com> wrote:
> 2010/8/20 Dennis J. <djacobfeuerb...@gmail.com>:
>
>
>
>
>
> >  On 08/20/2010 10:15 AM, Jérôme Loyet wrote:
>
> >> 2010/8/20 Thomas Fritz<fritz...@gmail.com>:
>
> >>> 2010/8/20 Jérôme Loyet<m...@fatbsd.com>
>
> >>>> Hi,
>
> >>>> I was thinking about all that.
>
> >>>> There is maybe a feature which can help in this case.
>
> >>>> what if there is a "pid" directive per pool. In this case FPM will
> >>>> write in this PID file each process PIDs dedicated to a pool. So that,
> >>>> if you want to restart a pool, you'll be able to do it using the kill
> >>>> command.
>
> >>>> [global]
> >>>> pid=/var/run/php-fpm.pid ; master process PID
> >>>> ...
> >>>> [www.foo.bar]
> >>>> pid=/home/www/www.foo.bar/logs/php-fpm.pid
> >>>> ...
> >>>> [www.domain.com]
> >>>> pid=/home/www/www.domain.com/logs/php-fpm.pid
> >>>> ...
>
> >>>> In /var/run/php-fpm.pid, you'll have only the master process PID.
> >>>> You'll be able to stop/reload everything as usual. Nothing change
> >>>> here.
>
> >>> As i understand you: There is a php-fpm reload which only reloads the
> >>> configuration files - does a graceful restart?
>
> >> yes. man php-fpm tells you about this:
>
> >> Once started, php-fpm then responds to several POSIX signals:
>
> >>  SIGINT,SIGTERM      immediate termination
> >>  SIGQUIT             graceful stop
> >>  SIGUSR1             re-open log file
> >>  SIGUSR2             graceful reload of all workers + reload of fpm
> >> conf/binary
>
> >>>> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
> >>>> of the processes dedicated to the poolwww.foo.barusing :
>
> >>>> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>
> >>> Which will only kill all child processes of that specific pool but not
> >>> the
> >>> "master process" of that pool itself, right?
>
> >> yes
>
> >>>> The same forwww.domain.comby using :
> >>>> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>
> >>>> This will allow processes to restart a pool, to reload php.ini content
> >>>> but it WON'T reload any FPM related conf.
>
> >>> As i understand, it WILL reload pool specific php.ini, ENV and other
> >>> configuration values!?
>
> >> nope. It will reload the php.ini content (/usr/local/php/lib/php.ini
> >> for example) but it WON'T reload anything present on the php-fpm.conf
> >> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
> >> have to reload the master process which will restart all pool
> >> processes. There is nothing to do about this.
>
> >>> If that works that easy and the fpm process recreates its child process
> >>> after killing without problems - GREAT!
>
> >> it already works that easy. kill by hand all the FPM child processes
> >> and you'll see that they'll be recreated by the master process. I'm
> >> just proposing to note thoses PID in one file per pool to make
> >> operations easier. That's all.
>
> > This sounds racy. Are you going to update the pid file every time a process
> > dies and a replacement is created? What if I try to do the "kill
> > <worker-pid-file>" at the wrong moment? If I execute this after a new
> > process is created but before the pid file has been updated then this
> > process potentially still runs a different config.
> > This could lead to very subtle issues that are almost impossible to
> > reproduce and track down.
>
> > A better way would be to use the same approach but implement this killing of
> > processes in php-fpm itself. That way php-fpm could suspend the creation of
> > new process, kill the workers and then enable the creation of new processes
> > again (similar to what preempt_disable() and preempt_enable() do in the
> > kernel). e.g.:
>
> > disable_worker_forking();
> > kill_pool_workers();
> > reload_config();
> > enable_worker_forking();
>
> You're right. With the patch I just proposed (changing the process
> title), you can achieve this easely without doing anything else:
>
> root@wild# ps a | grep fpm
> 26028 pts/3    S+     0:00 php-fpm: master process
> (/usr/local/php-trunk/etc/php-fpm.conf)
> 26030 pts/3    S+     0:00 php-fpm: pool www_direct
> 26031 pts/3    S+     0:00 php-fpm: pool www_direct
> 26032 pts/3    S+     0:00 php-fpm: pool www_direct
> 26067 pts/3    S+     0:00 php-fpm: pool www_chroot
> 26068 pts/3    S+     0:00 php-fpm: pool www_chroot
> 26069 pts/3    S+     0:00 php-fpm: pool www_chroot
>
> # gracefull restart a pool
> root@wild# pkill -QUIT -f "php-fpm: pool www_chroot"
>
> # hard restart a pool
> root@wild# pkill -f "php-fpm: pool www_chroot"
>
> Maybe for those who need this, a hack can be done on the FPM init.d
> script as proposed earlier in this thread.
>
> ++ Jerome
>
>
>
> > Regards,
> >  Dennis

Nice to see all the proposed modifications but I pray though that someone is keeping an eye on the big picture and not ending us up with an overly complicated, and thus, inaccessible and difficult to use item.

Jerome, I like the idea that you had for renaming the processes (this
makes it similar to nginx too). Do you think this will get merged in
for the next point release?

On Aug 20, 1:37 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>
>
>
>
>
>
>
> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>
> >> 2010/8/20 Gints Murāns <gints.mur...@gmail.com>:
> >> > Yes, I agree that it would solve it. I have a startup script already
> >> > in mind for this :) I mean command line could look something like
> >> > this: "/etc/init.d/php-fpm restart-pool www"
>
> >> I also have in mind changing the processes label. Because every
> >> process are labeled "php-fpm opt_args" in ps or top outpout. I'd like
> >> to have
>
> >> php-fpm: master process
> >> php-fpm: poolwww.foo.bar
> >> php-fpm: poolwww.foo.bar
> >> php-fpm: poolwww.foo.bar
> >> php-fpm: poolwww.domain.com
> >> php-fpm: poolwww.domain.com
> >> php-fpm: poolwww.domain.com
>
> > +1 Great!
>
> If some of you can try the following path, it would be great. It
> changes the process title according to their type:
>
> http://bugs.php.net/52660
>
> PS output sample:
> 21744 pts/3    S+     0:00 php-fpm: master process
> (/usr/local/php-trunk/etc/php-fpm.conf)
> 21745 pts/3    S+     0:00 php-fpm: pool www_chroot
> 21746 pts/3    S+     0:00 php-fpm: pool www_direct
>
> ++ Jerome
>
>
>
>
>
> >> it'll help and will be more clear
>
> >> > On Aug 20, 11:15 am, Jérôme Loyet <m...@fatbsd.com> wrote:
> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>
> >> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>
> >> >> >> Hi,
>
> >> >> >> I was thinking about all that.
>
> >> >> >> There is maybe a feature which can help in this case.
>
> >> >> >> what if there is a "pid" directive per pool. In this case FPM will
> >> >> >> write in this PID file each process PIDs dedicated to a pool. So
> >> >> >> that,
> >> >> >> if you want to restart a pool, you'll be able to do it using the
> >> >> >> kill
> >> >> >> command.
>
> >> >> >> [global]
> >> >> >> pid=/var/run/php-fpm.pid ; master process PID
> >> >> >> ...
> >> >> >> [www.foo.bar]
> >> >> >> pid=/home/www/www.foo.bar/logs/php-fpm.pid
> >> >> >> ...
> >> >> >> [www.domain.com]
> >> >> >> pid=/home/www/www.domain.com/logs/php-fpm.pid
> >> >> >> ...
>
> >> >> >> In /var/run/php-fpm.pid, you'll have only the master process PID..
> >> >> >> You'll be able to stop/reload everything as usual. Nothing change
> >> >> >> here.
>
> >> >> > As i understand you: There is a php-fpm reload which only reloads the
> >> >> > configuration files - does a graceful restart?
>
> >> >> yes. man php-fpm tells you about this:
>
> >> >> Once started, php-fpm then responds to several POSIX signals:
>
> >> >>  SIGINT,SIGTERM      immediate termination
> >> >>  SIGQUIT             graceful stop
> >> >>  SIGUSR1             re-open log file
> >> >>  SIGUSR2             graceful reload of all workers + reload of fpm
> >> >> conf/binary
>
> >> >> >> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
> >> >> >> of the processes dedicated to the poolwww.foo.barusing:
>
> >> >> >> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>
> >> >> > Which will only kill all child processes of that specific pool but
> >> >> > not the
> >> >> > "master process" of that pool itself, right?
>
> >> >> yes
>
> >> >> >> The same forwww.domain.combyusing :
> >> >> >> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>
> >> >> >> This will allow processes to restart a pool, to reload php.ini
> >> >> >> content
> >> >> >> but it WON'T reload any FPM related conf.
>
> >> >> > As i understand, it WILL reload pool specific php.ini, ENV and other
> >> >> > configuration values!?
>
> >> >> nope. It will reload the php.ini content (/usr/local/php/lib/php.ini
> >> >> for example) but it WON'T reload anything present on the php-fpm.conf
> >> >> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
> >> >> have to reload the master process which will restart all pool
> >> >> processes. There is nothing to do about this.
>
> >> >> > If that works that easy and the fpm process recreates its child
> >> >> > process
> >> >> > after killing without problems - GREAT!
>
> >> >> it already works that easy. kill by hand all the FPM child processes
> >> >> and you'll see that they'll be recreated by the master process. I'm
> >> >> just proposing to note thoses PID in one file per pool to make
> >> >> operations easier. That's all.
>
> >> >> >> what do you think ?
>
> >> >> >> ++ jerome
>
> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
> >> >> >> > Hi!
> >> >> >> > If such a graceful reload/restart does not disrupt any pool, such
> >> >> >> > a
> >> >> >> > graceful
> >> >> >> > reload/restart of configuration changes in a pool (and i think
> >> >> >> > thats
> >> >> >> > most
> >> >> >> > common use case) is more than sufficient.
> >> >> >> > But in case a pool hangs ( i do not know if it possible, if it can
> >> >> >> > happen
> >> >> >> > and when how php-fpm handles this - i do not know anything from
> >> >> >> > the
> >> >> >> > technical design behind php-fpm) a restart of that specific pool
> >> >> >> > would
> >> >> >> > be
> >> >> >> > better. Or is it possible to check for such cases in the graceful
> >> >> >> > restart
> >> >> >> > option?
> >> >> >> > But anyway. php-fpm is great and i am really happy that they
> >> >> >> > included it
> >> >> >> > in
> >> >> >> > 5.3.3 than 5.4! Thank you very much for your hard work!
> >> >> >> > Kind regards
>
> >> >> >> > ---
> >> >> >> > Thomas FRITZ
> >> >> >> > web http://fritzthomas.com
> >> >> >> > twitter http://twitter.com/thomasf
>
> >> >> >> > 2010/8/19 Jérôme Loyet <m...@fatbsd.com>
>
> >> >> >> >> 2010/8/19 iain wright <iainw...@usc.edu>:
> >> >> >> >> > Agreed, we are close to deploying 8 CMS sites on a new server
> >> >> >> >> > each
> >> >> >> >> > running
> >> >> >> >> > in its own php-fpm pool, seems silly to have to restart all of
> >> >> >> >> > them
> >> >> >> >> > if
> >> >> >> >> > something happened to 1 site.
>
> >> >> >> >> it's definetely not possible by design. If you really want to
> >> >> >> >> separate
> >> >> >> >> each site, launch 8 instances of FPM with only one pool each. But
> >> >> >> >> it
> >> >> >> >> would not be that *beautiful*. With the include feature on the
> >> >> >> >> fpm
> >> >> >> >> conf file, you can share values between differents FPM conf
> >> >> >> >> files.
>
> >> >> >> >> > That being said, phpfpm has worked like a charm and I love it.
>
> >> >> >> >> Love to hear that :)
>
> >> >> >> >> > Iain Wright
> >> >> >> >> > LAN Administrator - CBIS
> >> >> >> >> > USC Clinical Translational Science Institute
>
> >> >> >> >> > On Thu, Aug 19, 2010 at 8:43 AM, Diemuzi <diem...@gmail.com>
> >> >> >> >> > wrote:
>
> >> >> >> >> >> Changing per pool values is a big part for me. I would rather
> >> >> >> >> >> restart
> >> >> >> >> >> only the pool I changed the values in so that I do not
> >> >> >> >> >> possibly
> >> >> >> >> >> disrupt the other pools running.
>
> >> >> >> >> >> On Jul 30, 5:38 pm, Gints Murāns <gints.mur...@gmail.com>
> >> >> >> >> >> wrote:
> >> >> >> >> >> > What happens with other pools when one of the pools have
> >> >> >> >> >> > gateway
> >> >> >> >> >> > timeout? I mean when there is some long script running, that
> >> >> >> >> >> > takes
> >> >> >> >> >> > up
> >> >> >> >> >> > all the pool resources. I guess that would have to affect
> >> >> >> >> >> > other
> >> >> >> >> >> > pools.
> >> >> >> >> >> > So I want to restart this one, why should I those others?
> >> >> >> >> >> > Especially
> >> >> >> >> >> > when there is a lot of webpages and a lot of pools.
>
> >> >> >> >> >> > On Jul 26, 9:35 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>
> >> >> >> >> >> > > 2010/7/26 GM <g...@gm.lv>:
>
> >> >> >> >> >> > > > I would like to know that too.
>
> >> >> >> >> >> > > can you describe why you want that ? what is the need ?
> >> >> >> >> >> > > Before
> >> >> >> >> >> > > coding
> >> >> >> >> >> > > anything I'd like to be sure the solution is the right
> >> >> >> >> >> > > one.
>
> >> >> >> >> >> > > > On Jul 25, 12:44 pm, Thomas Fritz <fritz...@gmail.com>
> >> >> >> >> >> > > > wrote:
> >> >> >> >> >> > > >> I think i can remember that i have somewhere read about
> >> >> >> >> >> > > >> that
> >> >> >> >> >> > > >> feature or
> >> >> >> >> >> > > >> feature wish!
> >> >> >> >> >> > > >> Is it possible to restart only a specific pool and not
> >> >> >> >> >> > > >> the
> >> >> >> >> >> > > >> whole
> >> >> >> >> >> > > >> FPM? Is it
> >> >> >> >> >> > > >> planned to support that?
>
> >> >> >> >> >> > > >> Thanks
>
> >> >> >> >> >> > > >> Kind regards
>
> >> >> >> >> >> > > >> ---
>
> >> >> >> >> >> > > >> *Thomas FRITZ*
> >> >> >> >> >> > > >> *web*http://fritzthomas.com
> >> >> >> >> >> > > >> *twitter*http://twitter.com/thomasf
>
> >> >> > --
>
> >> >> > ---
> >> >> > Thomas FRITZ
> >> >> > webhttp://fritzthomas.com
> >> >> > twitterhttp://twitter.com/thomasf
>
> > --
>
> > ---
> > Thomas FRITZ
> > webhttp://fritzthomas.com
> > twitterhttp://twitter.com/thomasf

2010/8/30 Davy <dcampano@gmail.com>:
> Jerome, I like the idea that you had for renaming the processes (this
> makes it similar to nginx too).  Do you think this will get merged in
> for the next point release?

don't know yet. It'll depend on the PHP release schedule which I don't know yet.

>
> On Aug 20, 1:37 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>
>>
>>
>>
>>
>>
>>
>> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>>
>> >> 2010/8/20 Gints Murāns <gints.mur...@gmail.com>:
>> >> > Yes, I agree that it would solve it. I have a startup script already
>> >> > in mind for this :) I mean command line could look something like
>> >> > this: "/etc/init.d/php-fpm restart-pool www"
>>
>> >> I also have in mind changing the processes label. Because every
>> >> process are labeled "php-fpm opt_args" in ps or top outpout. I'd like
>> >> to have
>>
>> >> php-fpm: master process
>> >> php-fpm: poolwww.foo.bar
>> >> php-fpm: poolwww.foo.bar
>> >> php-fpm: poolwww.foo.bar
>> >> php-fpm: poolwww.domain.com
>> >> php-fpm: poolwww.domain.com
>> >> php-fpm: poolwww.domain.com
>>
>> > +1 Great!
>>
>> If some of you can try the following path, it would be great. It
>> changes the process title according to their type:
>>
>> http://bugs.php.net/52660
>>
>> PS output sample:
>> 21744 pts/3    S+     0:00 php-fpm: master process
>> (/usr/local/php-trunk/etc/php-fpm.conf)
>> 21745 pts/3    S+     0:00 php-fpm: pool www_chroot
>> 21746 pts/3    S+     0:00 php-fpm: pool www_direct
>>
>> ++ Jerome
>>
>>
>>
>>
>>
>> >> it'll help and will be more clear
>>
>> >> > On Aug 20, 11:15 am, Jérôme Loyet <m...@fatbsd.com> wrote:
>> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>
>> >> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>>
>> >> >> >> Hi,
>>
>> >> >> >> I was thinking about all that.
>>
>> >> >> >> There is maybe a feature which can help in this case.
>>
>> >> >> >> what if there is a "pid" directive per pool. In this case FPM will
>> >> >> >> write in this PID file each process PIDs dedicated to a pool. So
>> >> >> >> that,
>> >> >> >> if you want to restart a pool, you'll be able to do it using the
>> >> >> >> kill
>> >> >> >> command.
>>
>> >> >> >> [global]
>> >> >> >> pid=/var/run/php-fpm.pid ; master process PID
>> >> >> >> ...
>> >> >> >> [www.foo.bar]
>> >> >> >> pid=/home/www/www.foo.bar/logs/php-fpm.pid
>> >> >> >> ...
>> >> >> >> [www.domain.com]
>> >> >> >> pid=/home/www/www.domain.com/logs/php-fpm.pid
>> >> >> >> ...
>>
>> >> >> >> In /var/run/php-fpm.pid, you'll have only the master process PID.
>> >> >> >> You'll be able to stop/reload everything as usual. Nothing change
>> >> >> >> here.
>>
>> >> >> > As i understand you: There is a php-fpm reload which only reloads the
>> >> >> > configuration files - does a graceful restart?
>>
>> >> >> yes. man php-fpm tells you about this:
>>
>> >> >> Once started, php-fpm then responds to several POSIX signals:
>>
>> >> >>  SIGINT,SIGTERM      immediate termination
>> >> >>  SIGQUIT             graceful stop
>> >> >>  SIGUSR1             re-open log file
>> >> >>  SIGUSR2             graceful reload of all workers + reload of fpm
>> >> >> conf/binary
>>
>> >> >> >> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
>> >> >> >> of the processes dedicated to the poolwww.foo.barusing:
>>
>> >> >> >> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>>
>> >> >> > Which will only kill all child processes of that specific pool but
>> >> >> > not the
>> >> >> > "master process" of that pool itself, right?
>>
>> >> >> yes
>>
>> >> >> >> The same forwww.domain.combyusing :
>> >> >> >> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>>
>> >> >> >> This will allow processes to restart a pool, to reload php.ini
>> >> >> >> content
>> >> >> >> but it WON'T reload any FPM related conf.
>>
>> >> >> > As i understand, it WILL reload pool specific php.ini, ENV and other
>> >> >> > configuration values!?
>>
>> >> >> nope. It will reload the php.ini content (/usr/local/php/lib/php.ini
>> >> >> for example) but it WON'T reload anything present on the php-fpm.conf
>> >> >> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
>> >> >> have to reload the master process which will restart all pool
>> >> >> processes. There is nothing to do about this.
>>
>> >> >> > If that works that easy and the fpm process recreates its child
>> >> >> > process
>> >> >> > after killing without problems - GREAT!
>>
>> >> >> it already works that easy. kill by hand all the FPM child processes
>> >> >> and you'll see that they'll be recreated by the master process. I'm
>> >> >> just proposing to note thoses PID in one file per pool to make
>> >> >> operations easier. That's all.
>>
>> >> >> >> what do you think ?
>>
>> >> >> >> ++ jerome
>>
>> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>> >> >> >> > Hi!
>> >> >> >> > If such a graceful reload/restart does not disrupt any pool, such
>> >> >> >> > a
>> >> >> >> > graceful
>> >> >> >> > reload/restart of configuration changes in a pool (and i think
>> >> >> >> > thats
>> >> >> >> > most
>> >> >> >> > common use case) is more than sufficient.
>> >> >> >> > But in case a pool hangs ( i do not know if it possible, if it can
>> >> >> >> > happen
>> >> >> >> > and when how php-fpm handles this - i do not know anything from
>> >> >> >> > the
>> >> >> >> > technical design behind php-fpm) a restart of that specific pool
>> >> >> >> > would
>> >> >> >> > be
>> >> >> >> > better. Or is it possible to check for such cases in the graceful
>> >> >> >> > restart
>> >> >> >> > option?
>> >> >> >> > But anyway. php-fpm is great and i am really happy that they
>> >> >> >> > included it
>> >> >> >> > in
>> >> >> >> > 5.3.3 than 5.4! Thank you very much for your hard work!
>> >> >> >> > Kind regards
>>
>> >> >> >> > ---
>> >> >> >> > Thomas FRITZ
>> >> >> >> > web http://fritzthomas.com
>> >> >> >> > twitter http://twitter.com/thomasf
>>
>> >> >> >> > 2010/8/19 Jérôme Loyet <m...@fatbsd.com>
>>
>> >> >> >> >> 2010/8/19 iain wright <iainw...@usc.edu>:
>> >> >> >> >> > Agreed, we are close to deploying 8 CMS sites on a new server
>> >> >> >> >> > each
>> >> >> >> >> > running
>> >> >> >> >> > in its own php-fpm pool, seems silly to have to restart all of
>> >> >> >> >> > them
>> >> >> >> >> > if
>> >> >> >> >> > something happened to 1 site.
>>
>> >> >> >> >> it's definetely not possible by design. If you really want to
>> >> >> >> >> separate
>> >> >> >> >> each site, launch 8 instances of FPM with only one pool each.. But
>> >> >> >> >> it
>> >> >> >> >> would not be that *beautiful*. With the include feature on the
>> >> >> >> >> fpm
>> >> >> >> >> conf file, you can share values between differents FPM conf
>> >> >> >> >> files.
>>
>> >> >> >> >> > That being said, phpfpm has worked like a charm and I love it.
>>
>> >> >> >> >> Love to hear that :)
>>
>> >> >> >> >> > Iain Wright
>> >> >> >> >> > LAN Administrator - CBIS
>> >> >> >> >> > USC Clinical Translational Science Institute
>>
>> >> >> >> >> > On Thu, Aug 19, 2010 at 8:43 AM, Diemuzi <diem...@gmail.com>
>> >> >> >> >> > wrote:
>>
>> >> >> >> >> >> Changing per pool values is a big part for me. I would rather
>> >> >> >> >> >> restart
>> >> >> >> >> >> only the pool I changed the values in so that I do not
>> >> >> >> >> >> possibly
>> >> >> >> >> >> disrupt the other pools running.
>>
>> >> >> >> >> >> On Jul 30, 5:38 pm, Gints Murāns <gints.mur...@gmail.com>
>> >> >> >> >> >> wrote:
>> >> >> >> >> >> > What happens with other pools when one of the pools have
>> >> >> >> >> >> > gateway
>> >> >> >> >> >> > timeout? I mean when there is some long script running, that
>> >> >> >> >> >> > takes
>> >> >> >> >> >> > up
>> >> >> >> >> >> > all the pool resources. I guess that would have to affect
>> >> >> >> >> >> > other
>> >> >> >> >> >> > pools.
>> >> >> >> >> >> > So I want to restart this one, why should I those others?
>> >> >> >> >> >> > Especially
>> >> >> >> >> >> > when there is a lot of webpages and a lot of pools.
>>
>> >> >> >> >> >> > On Jul 26, 9:35 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>>
>> >> >> >> >> >> > > 2010/7/26 GM <g...@gm.lv>:
>>
>> >> >> >> >> >> > > > I would like to know that too.
>>
>> >> >> >> >> >> > > can you describe why you want that ? what is the need ?
>> >> >> >> >> >> > > Before
>> >> >> >> >> >> > > coding
>> >> >> >> >> >> > > anything I'd like to be sure the solution is the right
>> >> >> >> >> >> > > one.
>>
>> >> >> >> >> >> > > > On Jul 25, 12:44 pm, Thomas Fritz <fritz...@gmail.com>
>> >> >> >> >> >> > > > wrote:
>> >> >> >> >> >> > > >> I think i can remember that i have somewhere read about
>> >> >> >> >> >> > > >> that
>> >> >> >> >> >> > > >> feature or
>> >> >> >> >> >> > > >> feature wish!
>> >> >> >> >> >> > > >> Is it possible to restart only a specific pool and not
>> >> >> >> >> >> > > >> the
>> >> >> >> >> >> > > >> whole
>> >> >> >> >> >> > > >> FPM? Is it
>> >> >> >> >> >> > > >> planned to support that?
>>
>> >> >> >> >> >> > > >> Thanks
>>
>> >> >> >> >> >> > > >> Kind regards
>>
>> >> >> >> >> >> > > >> ---
>>
>> >> >> >> >> >> > > >> *Thomas FRITZ*
>> >> >> >> >> >> > > >> *web*http://fritzthomas.com
>> >> >> >> >> >> > > >> *twitter*http://twitter.com/thomasf
>>
>> >> >> > --
>>
>> >> >> > ---
>> >> >> > Thomas FRITZ
>> >> >> > webhttp://fritzthomas.com
>> >> >> > twitterhttp://twitter.com/thomasf
>>
>> > --
>>
>> > ---
>> > Thomas FRITZ
>> > webhttp://fritzthomas.com
>> > twitterhttp://twitter.com/thomasf
>

Exelent would be if fpm status page would have some authorization and
behind it there should have an option to kill pools. As well status
page could show all pools, not just one it is running in.

On Aug 30, 8:40 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
> 2010/8/30 Davy <dcamp...@gmail.com>:
>
> > Jerome, I like the idea that you had for renaming the processes (this
> > makes it similar to nginx too).  Do you think this will get merged in
> > for the next point release?
>
> don't know yet. It'll depend on the PHP release schedule which I don't know yet.
>
>
>
> > On Aug 20, 1:37 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>
> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>
> >> >> 2010/8/20 Gints Murāns <gints.mur...@gmail.com>:
> >> >> > Yes, I agree that it would solve it. I have a startup script already
> >> >> > in mind for this :) I mean command line could look something like
> >> >> > this: "/etc/init.d/php-fpmrestart-pool www"
>
> >> >> I also have in mind changing the processes label. Because every
> >> >> process are labeled "php-fpm opt_args" in ps or top outpout. I'd like
> >> >> to have
>
> >> >> php-fpm: master process
> >> >> php-fpm: poolwww.foo.bar
> >> >> php-fpm: poolwww.foo.bar
> >> >> php-fpm: poolwww.foo.bar
> >> >> php-fpm: poolwww.domain.com
> >> >> php-fpm: poolwww.domain.com
> >> >> php-fpm: poolwww.domain.com
>
> >> > +1 Great!
>
> >> If some of you can try the following path, it would be great. It
> >> changes the process title according to their type:
>
> >>http://bugs.php.net/52660
>
> >> PS output sample:
> >> 21744 pts/3    S+     0:00 php-fpm: master process
> >> (/usr/local/php-trunk/etc/php-fpm.conf)
> >> 21745 pts/3    S+     0:00 php-fpm: pool www_chroot
> >> 21746 pts/3    S+     0:00 php-fpm: pool www_direct
>
> >> ++ Jerome
>
> >> >> it'll help and will be more clear
>
> >> >> > On Aug 20, 11:15 am, Jérôme Loyet <m...@fatbsd.com> wrote:
> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>
> >> >> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>
> >> >> >> >> Hi,
>
> >> >> >> >> I was thinking about all that.
>
> >> >> >> >> There is maybe a feature which can help in this case.
>
> >> >> >> >> what if there is a "pid" directive per pool. In this case FPM will
> >> >> >> >> write in this PID file each process PIDs dedicated to a pool. So
> >> >> >> >> that,
> >> >> >> >> if you want torestarta pool, you'll be able to do it using the
> >> >> >> >> kill
> >> >> >> >> command.
>
> >> >> >> >> [global]
> >> >> >> >> pid=/var/run/php-fpm.pid ; master process PID
> >> >> >> >> ...
> >> >> >> >> [www.foo.bar]
> >> >> >> >> pid=/home/www/www.foo.bar/logs/php-fpm.pid
> >> >> >> >> ...
> >> >> >> >> [www.domain.com]
> >> >> >> >> pid=/home/www/www.domain.com/logs/php-fpm.pid
> >> >> >> >> ...
>
> >> >> >> >> In /var/run/php-fpm.pid, you'll have only the master process PID.
> >> >> >> >> You'll be able to stop/reload everything as usual. Nothing change
> >> >> >> >> here.
>
> >> >> >> > As i understand you: There is a php-fpm reload which only reloads the
> >> >> >> > configuration files - does a gracefulrestart?
>
> >> >> >> yes. man php-fpm tells you about this:
>
> >> >> >> Once started, php-fpm then responds to several POSIX signals:
>
> >> >> >>  SIGINT,SIGTERM      immediate termination
> >> >> >>  SIGQUIT             graceful stop
> >> >> >>  SIGUSR1             re-open log file
> >> >> >>  SIGUSR2             graceful reload of all workers + reload of fpm
> >> >> >> conf/binary
>
> >> >> >> >> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
> >> >> >> >> of the processes dedicated to the poolwww.foo.barusing:
>
> >> >> >> >> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>
> >> >> >> > Which will only kill all child processes of that specific pool but
> >> >> >> > not the
> >> >> >> > "master process" of that pool itself, right?
>
> >> >> >> yes
>
> >> >> >> >> The same forwww.domain.combyusing:
> >> >> >> >> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>
> >> >> >> >> This will allow processes torestarta pool, to reload php.ini
> >> >> >> >> content
> >> >> >> >> but it WON'T reload any FPM related conf.
>
> >> >> >> > As i understand, it WILL reload pool specific php.ini, ENV and other
> >> >> >> > configuration values!?
>
> >> >> >> nope. It will reload the php.ini content (/usr/local/php/lib/php..ini
> >> >> >> for example) but it WON'T reload anything present on the php-fpm..conf
> >> >> >> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
> >> >> >> have to reload the master process which willrestartall pool
> >> >> >> processes. There is nothing to do about this.
>
> >> >> >> > If that works that easy and the fpm process recreates its child
> >> >> >> > process
> >> >> >> > after killing without problems - GREAT!
>
> >> >> >> it already works that easy. kill by hand all the FPM child processes
> >> >> >> and you'll see that they'll be recreated by the master process. I'm
> >> >> >> just proposing to note thoses PID in one file per pool to make
> >> >> >> operations easier. That's all.
>
> >> >> >> >> what do you think ?
>
> >> >> >> >> ++ jerome
>
> >> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
> >> >> >> >> > Hi!
> >> >> >> >> > If such a graceful reload/restartdoes not disrupt any pool, such
> >> >> >> >> > a
> >> >> >> >> > graceful
> >> >> >> >> > reload/restartof configuration changes in a pool (and i think
> >> >> >> >> > thats
> >> >> >> >> > most
> >> >> >> >> > common use case) is more than sufficient.
> >> >> >> >> > But in case a pool hangs ( i do not know if it possible, if it can
> >> >> >> >> > happen
> >> >> >> >> > and when how php-fpm handles this - i do not know anything from
> >> >> >> >> > the
> >> >> >> >> > technical design behind php-fpm) arestartof that specific pool
> >> >> >> >> > would
> >> >> >> >> > be
> >> >> >> >> > better. Or is it possible to check for such cases in the graceful
> >> >> >> >> >restart
> >> >> >> >> > option?
> >> >> >> >> > But anyway. php-fpm is great and i am really happy that they
> >> >> >> >> > included it
> >> >> >> >> > in
> >> >> >> >> > 5.3.3 than 5.4! Thank you very much for your hard work!
> >> >> >> >> > Kind regards
>
> >> >> >> >> > ---
> >> >> >> >> > Thomas FRITZ
> >> >> >> >> > web http://fritzthomas.com
> >> >> >> >> > twitter http://twitter.com/thomasf
>
> >> >> >> >> > 2010/8/19 Jérôme Loyet <m...@fatbsd.com>
>
> >> >> >> >> >> 2010/8/19 iain wright <iainw...@usc.edu>:
> >> >> >> >> >> > Agreed, we are close to deploying 8 CMS sites on a new server
> >> >> >> >> >> > each
> >> >> >> >> >> > running
> >> >> >> >> >> > in its own php-fpm pool, seems silly to have torestartall of
> >> >> >> >> >> > them
> >> >> >> >> >> > if
> >> >> >> >> >> > something happened to 1 site.
>
> >> >> >> >> >> it's definetely not possible by design. If you really want to
> >> >> >> >> >> separate
> >> >> >> >> >> each site, launch 8 instances of FPM with only one pool each. But
> >> >> >> >> >> it
> >> >> >> >> >> would not be that *beautiful*. With the include feature on the
> >> >> >> >> >> fpm
> >> >> >> >> >> conf file, you can share values between differents FPM conf
> >> >> >> >> >> files.
>
> >> >> >> >> >> > That being said, phpfpm has worked like a charm and I love it.
>
> >> >> >> >> >> Love to hear that :)
>
> >> >> >> >> >> > Iain Wright
> >> >> >> >> >> > LAN Administrator - CBIS
> >> >> >> >> >> > USC Clinical Translational Science Institute
>
> >> >> >> >> >> > On Thu, Aug 19, 2010 at 8:43 AM, Diemuzi <diem...@gmail.com>
> >> >> >> >> >> > wrote:
>
> >> >> >> >> >> >> Changing per pool values is a big part for me. I would rather
> >> >> >> >> >> >>restart
> >> >> >> >> >> >> only the pool I changed the values in so that I do not
> >> >> >> >> >> >> possibly
> >> >> >> >> >> >> disrupt the other pools running.
>
> >> >> >> >> >> >> On Jul 30, 5:38 pm, Gints Murāns <gints.mur....@gmail.com>
> >> >> >> >> >> >> wrote:
> >> >> >> >> >> >> > What happens with other pools when one of the pools have
> >> >> >> >> >> >> > gateway
> >> >> >> >> >> >> > timeout? I mean when there is some long script running, that
> >> >> >> >> >> >> > takes
> >> >> >> >> >> >> > up
> >> >> >> >> >> >> > all the pool resources. I guess that would have to affect
> >> >> >> >> >> >> > other
> >> >> >> >> >> >> > pools.
> >> >> >> >> >> >> > So I want torestartthis one, why should I those others?
> >> >> >> >> >> >> > Especially
> >> >> >> >> >> >> > when there is a lot of webpages and a lot of pools.
>
> >> >> >> >> >> >> > On Jul 26, 9:35 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>
> >> >> >> >> >> >> > > 2010/7/26 GM <g...@gm.lv>:
>
> >> >> >> >> >> >> > > > I would like to know that too.
>
> >> >> >> >> >> >> > > can you describe why you want that ? what is the need ?
> >> >> >> >> >> >> > > Before
> >> >> >> >> >> >> > > coding
> >> >> >> >> >> >> > > anything I'd like to be sure the solution is the right
> >> >> >> >> >> >> > > one.
>
> >> >> >> >> >> >> > > > On Jul 25, 12:44 pm, Thomas Fritz <fritz...@gmail.com>
> >> >> >> >> >> >> > > > wrote:
> >> >> >> >> >> >> > > >> I think i can remember that i have somewhere read about
> >> >> >> >> >> >> > > >> that
> >> >> >> >> >> >> > > >> feature or
> >> >> >> >> >> >> > > >> feature wish!
> >> >> >> >> >> >> > > >> Is it possible torestartonly a specific pool and not
> >> >> >> >> >> >> > > >> the
> >> >> >> >> >> >> > > >> whole
> >> >> >> >> >> >> > > >> FPM? Is it
> >> >> >> >> >> >> > > >> planned to support that?
>
> >> >> >> >> >> >> > > >> Thanks
>
> >> >> >> >> >> >> > > >> Kind regards
>
> >> >> >> >> >> >> > > >> ---
>
> >> >> >> >> >> >> > > >> *Thomas FRITZ*
> >> >> >> >> >> >> > > >> *web*http://fritzthomas.com
> >> >> >> >> >> >> > > >> *twitter*http://twitter.com/thomasf
>
> >> >> >> > --
>
> >> >> >> > ---
> >> >> >> > Thomas FRITZ
> >> >> >> > webhttp://fritzthomas.com
> >> >> >> > twitterhttp://twitter.com/thomasf
>
> >> > --
>
> >> > ---
> >> > Thomas FRITZ
> >> > webhttp://fritzthomas.com
> >> > twitterhttp://twitter.com/thomasf
>
>

2010/9/17 Gints Murāns <gints.murans@gmail.com>:
> Exelent would be if fpm status page would have some authorization and
> behind it there should have an option to kill pools. As well status
> page could show all pools, not just one it is running in.

you're welcome to open a request on bugs.php.net ... and don't forget
to provide a patch :D

++ jerome

>
> On Aug 30, 8:40 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>> 2010/8/30 Davy <dcamp...@gmail.com>:
>>
>> > Jerome, I like the idea that you had for renaming the processes (this
>> > makes it similar to nginx too).  Do you think this will get merged in
>> > for the next point release?
>>
>> don't know yet. It'll depend on the PHP release schedule which I don't know yet.
>>
>>
>>
>> > On Aug 20, 1:37 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>
>> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>>
>> >> >> 2010/8/20 Gints Murāns <gints.mur...@gmail.com>:
>> >> >> > Yes, I agree that it would solve it. I have a startup script already
>> >> >> > in mind for this :) I mean command line could look something like
>> >> >> > this: "/etc/init.d/php-fpmrestart-pool www"
>>
>> >> >> I also have in mind changing the processes label. Because every
>> >> >> process are labeled "php-fpm opt_args" in ps or top outpout. I'd like
>> >> >> to have
>>
>> >> >> php-fpm: master process
>> >> >> php-fpm: poolwww.foo.bar
>> >> >> php-fpm: poolwww.foo.bar
>> >> >> php-fpm: poolwww.foo.bar
>> >> >> php-fpm: poolwww.domain.com
>> >> >> php-fpm: poolwww.domain.com
>> >> >> php-fpm: poolwww.domain.com
>>
>> >> > +1 Great!
>>
>> >> If some of you can try the following path, it would be great. It
>> >> changes the process title according to their type:
>>
>> >>http://bugs.php.net/52660
>>
>> >> PS output sample:
>> >> 21744 pts/3    S+     0:00 php-fpm: master process
>> >> (/usr/local/php-trunk/etc/php-fpm.conf)
>> >> 21745 pts/3    S+     0:00 php-fpm: pool www_chroot
>> >> 21746 pts/3    S+     0:00 php-fpm: pool www_direct
>>
>> >> ++ Jerome
>>
>> >> >> it'll help and will be more clear
>>
>> >> >> > On Aug 20, 11:15 am, Jérôme Loyet <m...@fatbsd.com> wrote:
>> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>
>> >> >> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>>
>> >> >> >> >> Hi,
>>
>> >> >> >> >> I was thinking about all that.
>>
>> >> >> >> >> There is maybe a feature which can help in this case.
>>
>> >> >> >> >> what if there is a "pid" directive per pool. In this case FPM will
>> >> >> >> >> write in this PID file each process PIDs dedicated to a pool.. So
>> >> >> >> >> that,
>> >> >> >> >> if you want torestarta pool, you'll be able to do it using the
>> >> >> >> >> kill
>> >> >> >> >> command.
>>
>> >> >> >> >> [global]
>> >> >> >> >> pid=/var/run/php-fpm.pid ; master process PID
>> >> >> >> >> ...
>> >> >> >> >> [www.foo.bar]
>> >> >> >> >> pid=/home/www/www.foo.bar/logs/php-fpm.pid
>> >> >> >> >> ...
>> >> >> >> >> [www.domain.com]
>> >> >> >> >> pid=/home/www/www.domain.com/logs/php-fpm.pid
>> >> >> >> >> ...
>>
>> >> >> >> >> In /var/run/php-fpm.pid, you'll have only the master process PID.
>> >> >> >> >> You'll be able to stop/reload everything as usual. Nothing change
>> >> >> >> >> here.
>>
>> >> >> >> > As i understand you: There is a php-fpm reload which only reloads the
>> >> >> >> > configuration files - does a gracefulrestart?
>>
>> >> >> >> yes. man php-fpm tells you about this:
>>
>> >> >> >> Once started, php-fpm then responds to several POSIX signals:
>>
>> >> >> >>  SIGINT,SIGTERM      immediate termination
>> >> >> >>  SIGQUIT             graceful stop
>> >> >> >>  SIGUSR1             re-open log file
>> >> >> >>  SIGUSR2             graceful reload of all workers + reload of fpm
>> >> >> >> conf/binary
>>
>> >> >> >> >> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
>> >> >> >> >> of the processes dedicated to the poolwww.foo.barusing:
>>
>> >> >> >> >> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>>
>> >> >> >> > Which will only kill all child processes of that specific pool but
>> >> >> >> > not the
>> >> >> >> > "master process" of that pool itself, right?
>>
>> >> >> >> yes
>>
>> >> >> >> >> The same forwww.domain.combyusing:
>> >> >> >> >> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>>
>> >> >> >> >> This will allow processes torestarta pool, to reload php.ini
>> >> >> >> >> content
>> >> >> >> >> but it WON'T reload any FPM related conf.
>>
>> >> >> >> > As i understand, it WILL reload pool specific php.ini, ENV and other
>> >> >> >> > configuration values!?
>>
>> >> >> >> nope. It will reload the php.ini content (/usr/local/php/lib/php.ini
>> >> >> >> for example) but it WON'T reload anything present on the php-fpm.conf
>> >> >> >> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
>> >> >> >> have to reload the master process which willrestartall pool
>> >> >> >> processes. There is nothing to do about this.
>>
>> >> >> >> > If that works that easy and the fpm process recreates its child
>> >> >> >> > process
>> >> >> >> > after killing without problems - GREAT!
>>
>> >> >> >> it already works that easy. kill by hand all the FPM child processes
>> >> >> >> and you'll see that they'll be recreated by the master process. I'm
>> >> >> >> just proposing to note thoses PID in one file per pool to make
>> >> >> >> operations easier. That's all.
>>
>> >> >> >> >> what do you think ?
>>
>> >> >> >> >> ++ jerome
>>
>> >> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>> >> >> >> >> > Hi!
>> >> >> >> >> > If such a graceful reload/restartdoes not disrupt any pool, such
>> >> >> >> >> > a
>> >> >> >> >> > graceful
>> >> >> >> >> > reload/restartof configuration changes in a pool (and i think
>> >> >> >> >> > thats
>> >> >> >> >> > most
>> >> >> >> >> > common use case) is more than sufficient.
>> >> >> >> >> > But in case a pool hangs ( i do not know if it possible, if it can
>> >> >> >> >> > happen
>> >> >> >> >> > and when how php-fpm handles this - i do not know anything from
>> >> >> >> >> > the
>> >> >> >> >> > technical design behind php-fpm) arestartof that specific pool
>> >> >> >> >> > would
>> >> >> >> >> > be
>> >> >> >> >> > better. Or is it possible to check for such cases in the graceful
>> >> >> >> >> >restart
>> >> >> >> >> > option?
>> >> >> >> >> > But anyway. php-fpm is great and i am really happy that they
>> >> >> >> >> > included it
>> >> >> >> >> > in
>> >> >> >> >> > 5.3.3 than 5.4! Thank you very much for your hard work!
>> >> >> >> >> > Kind regards
>>
>> >> >> >> >> > ---
>> >> >> >> >> > Thomas FRITZ
>> >> >> >> >> > web http://fritzthomas.com
>> >> >> >> >> > twitter http://twitter.com/thomasf
>>
>> >> >> >> >> > 2010/8/19 Jérôme Loyet <m...@fatbsd.com>
>>
>> >> >> >> >> >> 2010/8/19 iain wright <iainw...@usc.edu>:
>> >> >> >> >> >> > Agreed, we are close to deploying 8 CMS sites on a new server
>> >> >> >> >> >> > each
>> >> >> >> >> >> > running
>> >> >> >> >> >> > in its own php-fpm pool, seems silly to have torestartall of
>> >> >> >> >> >> > them
>> >> >> >> >> >> > if
>> >> >> >> >> >> > something happened to 1 site.
>>
>> >> >> >> >> >> it's definetely not possible by design. If you really want to
>> >> >> >> >> >> separate
>> >> >> >> >> >> each site, launch 8 instances of FPM with only one pool each. But
>> >> >> >> >> >> it
>> >> >> >> >> >> would not be that *beautiful*. With the include feature on the
>> >> >> >> >> >> fpm
>> >> >> >> >> >> conf file, you can share values between differents FPM conf
>> >> >> >> >> >> files.
>>
>> >> >> >> >> >> > That being said, phpfpm has worked like a charm and I love it.
>>
>> >> >> >> >> >> Love to hear that :)
>>
>> >> >> >> >> >> > Iain Wright
>> >> >> >> >> >> > LAN Administrator - CBIS
>> >> >> >> >> >> > USC Clinical Translational Science Institute
>>
>> >> >> >> >> >> > On Thu, Aug 19, 2010 at 8:43 AM, Diemuzi <diem...@gmail..com>
>> >> >> >> >> >> > wrote:
>>
>> >> >> >> >> >> >> Changing per pool values is a big part for me. I would rather
>> >> >> >> >> >> >>restart
>> >> >> >> >> >> >> only the pool I changed the values in so that I do not
>> >> >> >> >> >> >> possibly
>> >> >> >> >> >> >> disrupt the other pools running.
>>
>> >> >> >> >> >> >> On Jul 30, 5:38 pm, Gints Murāns <gints.mur....@gmail.com>
>> >> >> >> >> >> >> wrote:
>> >> >> >> >> >> >> > What happens with other pools when one of the pools have
>> >> >> >> >> >> >> > gateway
>> >> >> >> >> >> >> > timeout? I mean when there is some long script running, that
>> >> >> >> >> >> >> > takes
>> >> >> >> >> >> >> > up
>> >> >> >> >> >> >> > all the pool resources. I guess that would have to affect
>> >> >> >> >> >> >> > other
>> >> >> >> >> >> >> > pools.
>> >> >> >> >> >> >> > So I want torestartthis one, why should I those others?
>> >> >> >> >> >> >> > Especially
>> >> >> >> >> >> >> > when there is a lot of webpages and a lot of pools.
>>
>> >> >> >> >> >> >> > On Jul 26, 9:35 pm, Jérôme Loyet <m....@fatbsd.com> wrote:
>>
>> >> >> >> >> >> >> > > 2010/7/26 GM <g...@gm.lv>:
>>
>> >> >> >> >> >> >> > > > I would like to know that too.
>>
>> >> >> >> >> >> >> > > can you describe why you want that ? what is the need ?
>> >> >> >> >> >> >> > > Before
>> >> >> >> >> >> >> > > coding
>> >> >> >> >> >> >> > > anything I'd like to be sure the solution is the right
>> >> >> >> >> >> >> > > one.
>>
>> >> >> >> >> >> >> > > > On Jul 25, 12:44 pm, Thomas Fritz <fritz...@gmail.com>
>> >> >> >> >> >> >> > > > wrote:
>> >> >> >> >> >> >> > > >> I think i can remember that i have somewhere read about
>> >> >> >> >> >> >> > > >> that
>> >> >> >> >> >> >> > > >> feature or
>> >> >> >> >> >> >> > > >> feature wish!
>> >> >> >> >> >> >> > > >> Is it possible torestartonly a specific pool and not
>> >> >> >> >> >> >> > > >> the
>> >> >> >> >> >> >> > > >> whole
>> >> >> >> >> >> >> > > >> FPM? Is it
>> >> >> >> >> >> >> > > >> planned to support that?
>>
>> >> >> >> >> >> >> > > >> Thanks
>>
>> >> >> >> >> >> >> > > >> Kind regards
>>
>> >> >> >> >> >> >> > > >> ---
>>
>> >> >> >> >> >> >> > > >> *Thomas FRITZ*
>> >> >> >> >> >> >> > > >> *web*http://fritzthomas.com
>> >> >> >> >> >> >> > > >> *twitter*http://twitter.com/thomasf
>>
>> >> >> >> > --
>>
>> >> >> >> > ---
>> >> >> >> > Thomas FRITZ
>> >> >> >> > webhttp://fritzthomas.com
>> >> >> >> > twitterhttp://twitter.com/thomasf
>>
>> >> > --
>>
>> >> > ---
>> >> > Thomas FRITZ
>> >> > webhttp://fritzthomas.com
>> >> > twitterhttp://twitter.com/thomasf
>>
>>
>

i would make it more like apc - ship with a php file example, but use
RESTful hooks or even raw API that is accessed via php functions
(however that would make it easy for anyone to use, so maybe not.
unless there was a secret "api key" defined somewhere...)


2010/9/17 Jérôme Loyet <ml@fatbsd.com>:
> 2010/9/17 Gints Murāns <gints.murans@gmail.com>:
>> Exelent would be if fpm status page would have some authorization and
>> behind it there should have an option to kill pools. As well status
>> page could show all pools, not just one it is running in.
>
> you're welcome to open a request on bugs.php.net ... and don't forget
> to provide a patch :D
>
> ++ jerome
>
>>
>> On Aug 30, 8:40 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>>> 2010/8/30 Davy <dcamp...@gmail.com>:
>>>
>>> > Jerome, I like the idea that you had for renaming the processes (this
>>> > makes it similar to nginx too).  Do you think this will get merged in
>>> > for the next point release?
>>>
>>> don't know yet. It'll depend on the PHP release schedule which I don't know yet.
>>>
>>>
>>>
>>> > On Aug 20, 1:37 pm, Jérôme Loyet <m...@fatbsd.com> wrote:
>>> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>>
>>> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>>>
>>> >> >> 2010/8/20 Gints Murāns <gints.mur...@gmail.com>:
>>> >> >> > Yes, I agree that it would solve it. I have a startup script already
>>> >> >> > in mind for this :) I mean command line could look something like
>>> >> >> > this: "/etc/init.d/php-fpmrestart-pool www"
>>>
>>> >> >> I also have in mind changing the processes label. Because every
>>> >> >> process are labeled "php-fpm opt_args" in ps or top outpout. I'd like
>>> >> >> to have
>>>
>>> >> >> php-fpm: master process
>>> >> >> php-fpm: poolwww.foo.bar
>>> >> >> php-fpm: poolwww.foo.bar
>>> >> >> php-fpm: poolwww.foo.bar
>>> >> >> php-fpm: poolwww.domain.com
>>> >> >> php-fpm: poolwww.domain.com
>>> >> >> php-fpm: poolwww.domain.com
>>>
>>> >> > +1 Great!
>>>
>>> >> If some of you can try the following path, it would be great. It
>>> >> changes the process title according to their type:
>>>
>>> >>http://bugs.php.net/52660
>>>
>>> >> PS output sample:
>>> >> 21744 pts/3    S+     0:00 php-fpm: master process
>>> >> (/usr/local/php-trunk/etc/php-fpm.conf)
>>> >> 21745 pts/3    S+     0:00 php-fpm: pool www_chroot
>>> >> 21746 pts/3    S+     0:00 php-fpm: pool www_direct
>>>
>>> >> ++ Jerome
>>>
>>> >> >> it'll help and will be more clear
>>>
>>> >> >> > On Aug 20, 11:15 am, Jérôme Loyet <m...@fatbsd.com> wrote:
>>> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>>
>>> >> >> >> > 2010/8/20 Jérôme Loyet <m...@fatbsd.com>
>>>
>>> >> >> >> >> Hi,
>>>
>>> >> >> >> >> I was thinking about all that.
>>>
>>> >> >> >> >> There is maybe a feature which can help in this case.
>>>
>>> >> >> >> >> what if there is a "pid" directive per pool. In this case FPM will
>>> >> >> >> >> write in this PID file each process PIDs dedicated to a pool. So
>>> >> >> >> >> that,
>>> >> >> >> >> if you want torestarta pool, you'll be able to do it using the
>>> >> >> >> >> kill
>>> >> >> >> >> command.
>>>
>>> >> >> >> >> [global]
>>> >> >> >> >> pid=/var/run/php-fpm.pid ; master process PID
>>> >> >> >> >> ...
>>> >> >> >> >> [www.foo.bar]
>>> >> >> >> >> pid=/home/www/www.foo.bar/logs/php-fpm.pid
>>> >> >> >> >> ...
>>> >> >> >> >> [www.domain.com]
>>> >> >> >> >> pid=/home/www/www.domain.com/logs/php-fpm.pid
>>> >> >> >> >> ...
>>>
>>> >> >> >> >> In /var/run/php-fpm.pid, you'll have only the master process PID.
>>> >> >> >> >> You'll be able to stop/reload everything as usual. Nothing change
>>> >> >> >> >> here.
>>>
>>> >> >> >> > As i understand you: There is a php-fpm reload which only reloads the
>>> >> >> >> > configuration files - does a gracefulrestart?
>>>
>>> >> >> >> yes. man php-fpm tells you about this:
>>>
>>> >> >> >> Once started, php-fpm then responds to several POSIX signals:
>>>
>>> >> >> >>  SIGINT,SIGTERM      immediate termination
>>> >> >> >>  SIGQUIT             graceful stop
>>> >> >> >>  SIGUSR1             re-open log file
>>> >> >> >>  SIGUSR2             graceful reload of all workers + reload of fpm
>>> >> >> >> conf/binary
>>>
>>> >> >> >> >> In /home/www/www.foo.bar/logs/php-fpm.pid, you'll have only the PIDs
>>> >> >> >> >> of the processes dedicated to the poolwww.foo.barusing:
>>>
>>> >> >> >> >> kill `cat /home/www/www.foo.bar/logs/php-fpm.pid`
>>>
>>> >> >> >> > Which will only kill all child processes of that specific pool but
>>> >> >> >> > not the
>>> >> >> >> > "master process" of that pool itself, right?
>>>
>>> >> >> >> yes
>>>
>>> >> >> >> >> The same forwww.domain.combyusing:
>>> >> >> >> >> kill `cat /home/www/www.domain.com/logs/php-fpm.pid`
>>>
>>> >> >> >> >> This will allow processes torestarta pool, to reload php.ini
>>> >> >> >> >> content
>>> >> >> >> >> but it WON'T reload any FPM related conf.
>>>
>>> >> >> >> > As i understand, it WILL reload pool specific php.ini, ENV and other
>>> >> >> >> > configuration values!?
>>>
>>> >> >> >> nope. It will reload the php.ini content (/usr/local/php/lib/php.ini
>>> >> >> >> for example) but it WON'T reload anything present on the php-fpm.conf
>>> >> >> >> (/usr/local/php/etc/php-fpm.conf). To reload the php-fpm.conf you'l
>>> >> >> >> have to reload the master process which willrestartall pool
>>> >> >> >> processes. There is nothing to do about this.
>>>
>>> >> >> >> > If that works that easy and the fpm process recreates its child
>>> >> >> >> > process
>>> >> >> >> > after killing without problems - GREAT!
>>>
>>> >> >> >> it already works that easy. kill by hand all the FPM child processes
>>> >> >> >> and you'll see that they'll be recreated by the master process.. I'm
>>> >> >> >> just proposing to note thoses PID in one file per pool to make
>>> >> >> >> operations easier. That's all.
>>>
>>> >> >> >> >> what do you think ?
>>>
>>> >> >> >> >> ++ jerome
>>>
>>> >> >> >> >> 2010/8/20 Thomas Fritz <fritz...@gmail.com>:
>>> >> >> >> >> > Hi!
>>> >> >> >> >> > If such a graceful reload/restartdoes not disrupt any pool, such
>>> >> >> >> >> > a
>>> >> >> >> >> > graceful
>>> >> >> >> >> > reload/restartof configuration changes in a pool (and i think
>>> >> >> >> >> > thats
>>> >> >> >> >> > most
>>> >> >> >> >> > common use case) is more than sufficient.
>>> >> >> >> >> > But in case a pool hangs ( i do not know if it possible, if it can
>>> >> >> >> >> > happen
>>> >> >> >> >> > and when how php-fpm handles this - i do not know anything from
>>> >> >> >> >> > the
>>> >> >> >> >> > technical design behind php-fpm) arestartof that specific pool
>>> >> >> >> >> > would
>>> >> >> >> >> > be
>>> >> >> >> >> > better. Or is it possible to check for such cases in the graceful
>>> >> >> >> >> >restart
>>> >> >> >> >> > option?
>>> >> >> >> >> > But anyway. php-fpm is great and i am really happy that they
>>> >> >> >> >> > included it
>>> >> >> >> >> > in
>>> >> >> >> >> > 5.3.3 than 5.4! Thank you very much for your hard work!
>>> >> >> >> >> > Kind regards
>>>
>>> >> >> >> >> > ---
>>> >> >> >> >> > Thomas FRITZ
>>> >> >> >> >> > web http://fritzthomas.com
>>> >> >> >> >> > twitter http://twitter.com/thomasf
>>>
>>> >> >> >> >> > 2010/8/19 Jérôme Loyet <m...@fatbsd.com>
>>>
>>> >> >> >> >> >> 2010/8/19 iain wright <iainw...@usc.edu>:
>>> >> >> >> >> >> > Agreed, we are close to deploying 8 CMS sites on a new server
>>> >> >> >> >> >> > each
>>> >> >> >> >> >> > running
>>> >> >> >> >> >> > in its own php-fpm pool, seems silly to have torestartall of
>>> >> >> >> >> >> > them
>>> >> >> >> >> >> > if
>>> >> >> >> >> >> > something happened to 1 site.
>>>
>>> >> >> >> >> >> it's definetely not possible by design. If you really want to
>>> >> >> >> >> >> separate
>>> >> >> >> >> >> each site, launch 8 instances of FPM with only one pool each. But
>>> >> >> >> >> >> it
>>> >> >> >> >> >> would not be that *beautiful*. With the include feature on the
>>> >> >> >> >> >> fpm
>>> >> >> >> >> >> conf file, you can share values between differents FPM conf
>>> >> >> >> >> >> files.
>>>
>>> >> >> >> >> >> > That being said, phpfpm has worked like a charm and I love it.
>>>
>>> >> >> >> >> >> Love to hear that :)
>>>
>>> >> >> >> >> >> > Iain Wright
>>> >> >> >> >> >> > LAN Administrator - CBIS
>>> >> >> >> >> >> > USC Clinical Translational Science Institute
>>>
>>> >> >> >> >> >> > On Thu, Aug 19, 2010 at 8:43 AM, Diemuzi <diem...@gmail.com>
>>> >> >> >> >> >> > wrote:
>>>
>>> >> >> >> >> >> >> Changing per pool values is a big part for me. I would rather
>>> >> >> >> >> >> >>restart
>>> >> >> >> >> >> >> only the pool I changed the values in so that I do not
>>> >> >> >> >> >> >> possibly
>>> >> >> >> >> >> >> disrupt the other pools running.
>>>
>>> >> >> >> >> >> >> On Jul 30, 5:38 pm, Gints Murāns <gints.mur....@gmail.com>
>>> >> >> >> >> >> >> wrote:
>>> >> >> >> >> >> >> > What happens with other pools when one of the pools have
>>> >> >> >> >> >> >> > gateway
>>> >> >> >> >> >> >> > timeout? I mean when there is some long script running, that
>>> >> >> >> >> >> >> > takes
>>> >> >> >> >> >> >> > up
>>> >> >> >> >> >> >> > all the pool resources. I guess that would have to affect
>>> >> >> >> >> >> >> > other
>>> >> >> >> >> >> >> > pools.
>>> >> >> >> >> >> >> > So I want torestartthis one, why should I those others?
>>> >> >> >> >> >> >> > Especially
>>> >> >> >> >> >> >> > when there is a lot of webpages and a lot of pools.
>>>
>>> >> >> >> >> >> >> > On Jul 26, 9:35 pm, Jérôme Loyet <m....@fatbsd.com> wrote:
>>>
>>> >> >> >> >> >> >> > > 2010/7/26 GM <g...@gm.lv>:
>>>
>>> >> >> >> >> >> >> > > > I would like to know that too.
>>>
>>> >> >> >> >> >> >> > > can you describe why you want that ? what is the need ?
>>> >> >> >> >> >> >> > > Before
>>> >> >> >> >> >> >> > > coding
>>> >> >> >> >> >> >> > > anything I'd like to be sure the solution is the right
>>> >> >> >> >> >> >> > > one.
>>>
>>> >> >> >> >> >> >> > > > On Jul 25, 12:44 pm, Thomas Fritz <fritz....@gmail.com>
>>> >> >> >> >> >> >> > > > wrote:
>>> >> >> >> >> >> >> > > >> I think i can remember that i have somewhere read about
>>> >> >> >> >> >> >> > > >> that
>>> >> >> >> >> >> >> > > >> feature or
>>> >> >> >> >> >> >> > > >> feature wish!
>>> >> >> >> >> >> >> > > >> Is it possible torestartonly a specific pool and not
>>> >> >> >> >> >> >> > > >> the
>>> >> >> >> >> >> >> > > >> whole
>>> >> >> >> >> >> >> > > >> FPM? Is it
>>> >> >> >> >> >> >> > > >> planned to support that?
>>>
>>> >> >> >> >> >> >> > > >> Thanks
>>>
>>> >> >> >> >> >> >> > > >> Kind regards
>>>
>>> >> >> >> >> >> >> > > >> ---
>>>
>>> >> >> >> >> >> >> > > >> *Thomas FRITZ*
>>> >> >> >> >> >> >> > > >> *web*http://fritzthomas.com
>>> >> >> >> >> >> >> > > >> *twitter*http://twitter.com/thomasf
>>>
>>> >> >> >> > --
>>>
>>> >> >> >> > ---
>>> >> >> >> > Thomas FRITZ
>>> >> >> >> > webhttp://fritzthomas.com
>>> >> >> >> > twitterhttp://twitter.com/thomasf
>>>
>>> >> > --
>>>
>>> >> > ---
>>> >> > Thomas FRITZ
>>> >> > webhttp://fritzthomas.com
>>> >> > twitterhttp://twitter.com/thomasf
>>>
>>>
>>
>

On Fri, Sep 17, 2010 at 9:13 AM, Michael Shadle <mike503@gmail.com> wrote:
> i would make it more like apc - ship with a php file example, but use
> RESTful hooks or even raw API that is accessed via php functions
> (however that would make it easy for anyone to use, so maybe not.
> unless there was a secret "api key" defined somewhere...)

to go even further on that tangent, that option could allow customers
to reload their -own- pools if it was API-based w/ some sort of method
for security.

(this is overcomplicating it of course... but could open up even more
granular control for people!)

2010/9/17 Michael Shadle <mike503@gmail.com>:
> On Fri, Sep 17, 2010 at 9:13 AM, Michael Shadle <mike503@gmail.com> wrote:
>> i would make it more like apc - ship with a php file example, but use
>> RESTful hooks or even raw API that is accessed via php functions
>> (however that would make it easy for anyone to use, so maybe not.
>> unless there was a secret "api key" defined somewhere...)
>
> to go even further on that tangent, that option could allow customers
> to reload their -own- pools if it was API-based w/ some sort of method
> for security.

that won't happened because if security breach it could open

>
> (this is overcomplicating it of course... but could open up even more
> granular control for people!)
>

2010/9/17 Jérôme Loyet <ml@fatbsd.com>:

>> to go even further on that tangent, that option could allow customers
>> to reload their -own- pools if it was API-based w/ some sort of method
>> for security.
>
> that won't happened because if security breach it could open

like i said, it was a tangent. i'm sure there could be some way of
shielding it. or hell, the api call would only be controlling the
workers underneath that same pool anyway. could be an ini option to
allow api calls or not :p