Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:47954
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain gmail.com designates 209.85.218.219 as permitted sender)
DomainKey-Status: bad
DomainKey-Signature: a=rsa-sha1; c=nofws;
        d=gmail.com; s=gamma;
        h=mime-version:in-reply-to:references:date:message-id:subject:from:to
         :cc:content-type;
        b=NAmC27/vuoMcu2woe0l6icHgw+8lJSq4jl+IbKaPLzWbnCHQYnUiuuAuP+ZFEAPvge
         9sZ97fP/PD+VQoAYw6h3hRWzSLwmwa+bCvdcx+zDVd8VsW4ytpIxkRJPEjGszlqPqy7g
         YP7OwhS6eFKoF3zuBC0pTX9WD4V8o1CTvFsMA=
MIME-Version: 1.0
In-Reply-To: <k2n3bea96c41004150041mf49f3004nf913435b70e7199b@mail.gmail.com>
References: <j2g7b32197c1004060011n5b78f062kf05bde7d0b2b9993@mail.gmail.com>
	 <j2n3bea96c41004140201ia1cffc4fr1c191a343af73931@mail.gmail.com>
	 <4BC58D10.4050609@lerdorf.com>
	 <y2x7b32197c1004141858sa672049dxda3e8f1404e97b04@mail.gmail.com>
	 <y2m3bea96c41004142303x190f5b45hc15952527b13a446@mail.gmail.com>
	 <o2rb54d0abe1004150009p3c1142ben2b415e1e429179bc@mail.gmail.com>
	 <w2yb54d0abe1004150014h895ee40cne7559d51bda3bd44@mail.gmail.com>
	 <o2i3bea96c41004150025td9b9ed89l72eb7154a80432c4@mail.gmail.com>
	 <k2pb54d0abe1004150039rbd189087z4a09a726796a74b1@mail.gmail.com>
	 <k2n3bea96c41004150041mf49f3004nf913435b70e7199b@mail.gmail.com>
Date: Thu, 15 Apr 2010 09:56:11 +0200
Message-ID: <m2zb54d0abe1004150056ze63f98c2zf4dbdae715f50c19@mail.gmail.com>
To: =?UTF-8?B?SsOpcsO0bWUgTG95ZXQ=?= <jerome@loyet.net>
Cc: =?UTF-8?B?5ruV6Lev6YGl?= <tly.phpfans@gmail.com>, 
	Rasmus Lerdorf <rasmus@lerdorf.com>, internals@lists.php.net
Content-Type: multipart/alternative; boundary=0015175ccffe09c1a5048441d467
Subject: Re: [PHP-DEV] One suggestion to PHP-FPM
From: tyra3l@gmail.com (Ferenc Kovacs)

--0015175ccffe09c1a5048441d467
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

On Thu, Apr 15, 2010 at 9:41 AM, J=C3=A9r=C3=B4me Loyet <jerome@loyet.net> =
wrote:

> 2010/4/15 Ferenc Kovacs <tyra3l@gmail.com>:
> >
> >
> > On Thu, Apr 15, 2010 at 9:25 AM, J=C3=A9r=C3=B4me Loyet <jerome@loyet.n=
et> wrote:
> >>
> >> 2010/4/15 Ferenc Kovacs <tyra3l@gmail.com>:
> >> >
> >> >
> >> > On Thu, Apr 15, 2010 at 9:09 AM, Ferenc Kovacs <tyra3l@gmail.com>
> wrote:
> >> >>
> >> >>
> >> >> 2010/4/15 J=C3=A9r=C3=B4me Loyet <jerome@loyet.net>
> >> >>>
> >> >>> 2010/4/15 =E6=BB=95=E8=B7=AF=E9=81=A5 <tly.phpfans@gmail.com>:
> >> >>> > We check error log after our server crashed, and we found that p=
hp
> >> >>> > heap
> >> >>> > memory is out of limit,
> >> >>> > so I think imagick use emalloc to allocate php heap memory, to
> check
> >> >>> > the
> >> >>> > zend_memory_usage is
> >> >>> > not a method for imagick, other circumstances which occupy huge
> >> >>> > memory
> >> >>> > can
> >> >>> > use this method to
> >> >>> > ensure the system have enough free memory.
> >> >>>
> >> >>> OK,
> >> >>>
> >> >>> one thing I didn't understand is "why should you want check de zen=
d
> >> >>> memory usage when there is a memory_limit parameter in PHP.ini " ?
> >> >>> does'nt it do the same at PHP layer when allocating ?
> >> >>>
> >> >>
> >> >> Obviously you didn't read the original mail, and the links inside:
> >> >>
> >> >>
> >> >>
> http://groups.google.com/group/highload-php-en/browse_thread/thread/1882a=
3b2257dcc5c/ea73892cea011541?lnk=3Dgst&q=3Dgaochunhui#ea73892cea011541
> >> >>
> >> >> The problem here is that PHP wouldn't free any allocated memory
> untill
> >> >> the process exits, so there's the risk of memory exhaustion if too
> >> >> many PHP processes were running, even though each process wouldn't
> >> >> exceed PHP's memory_limit.
> >> >>
> >> >> My suggestion is more about releasing the allocated memory as soon =
as
> >> >> possible. That is, this option is similar to "max_requests".
> >> >>
> >> >> PHP-FPM would kill the PHP process if the requests a process handle=
d
> >> >> exceed max_requests, and similarly, PHP-FPM should kill the PHP
> >> >> process whose memory usage exceeds "exit_on_memory_exceeds".
> >> >>
> >> >> So one of your lib (for example imagick) leaks memory, on the long
> run,
> >> >> it
> >> >> will exhaust the memory limit, and will kill a totaly request.
> >> >> You can set that how many request should be served with one worker,
> but
> >> >> you can't soft limit it's memory consumption.
> >> >> This is what the patch does:
> >> >> if you set the hard limit: (memory_limit) you can guarante that no
> >> >> process
> >> >> will use more memory, because if it tries, it will fail.
> >> >> and you can set soft limit, if that reached, the process will die a=
nd
> >> >> respawn after finishing the current request.
> >> >>
> >> >> Tyrael
> >> >
> >> > Btw: did you know about
> >> > http://php.net/manual/en/function.apache-child-terminate.php I just
> >> > noticed
> >> > a few days ago, obviously, you can't use with cgi/fastcgi/php-fpm bu=
t
> >> > for
> >> > apache, it does exactly what you need, except that you have to execu=
te
> >> > from
> >> > userland.
> >>
> >> This function is interesting but it's called from php scripts. When
> >> you are a sysadmin who host hundreds of different customers, you can't
> >> garanty each customer will call this function each time he uses
> >> consomption scripts.
> >
> > Yeah, I know, and I think I did mention that, but this sounds like a go=
od
> > feature in the language, but sadly it has too many limitation in it's
> > current form. (no windows support, only works with apache)
>
> I see how soft_limit could work (after processing a page, memory is
> checked and if it's more than the soft limit, the process is killed).
> But how can you make the hard_limit to work ? You'll have to setup
> memory check for each process by a master process

This is what the proposed php-fpm patch does, or am I missing something?

> , because the child
> won't be able to check it's memory realtime (or I missed something)
>
> ++ Jerome
>
If we talking about my proposal(to kill/set to kill the worker from inside
the request), then maybe you are right, I don't know much about the interna=
l
memory management system, but you are saying, that there is no way to tell
the worker's memory consumption from inside the php userland?
there are functions like memory_get_usage(), memory_get_peak_usage(), and
there are ways in every OS to tell the ~ memory consumption of a given
process.
Am I missing something?

Tyrael

--0015175ccffe09c1a5048441d467--