php fastcgi process dies every 3-4 hours?

I am running php with fastcgi. Everything appears to be working fine
except for some errors in my logs, which I get about every 3-4 hours:

[Tue Aug 19 17:20:48 2003] [warn] FastCGI: (dynamic) server
"/var/www/php/php.fcgi" (pid 22390) terminated by calling exit with
status '0'

[Tue Aug 19 17:20:48 2003] [warn] FastCGI: (dynamic) server
"/var/www/php/php.fcgi" (uid 45, gid 45) restarted (pid 26326)


From my understanding, fastcgi is supposed to keep 5 servers running by
default even if there is no demand for them. This is a very low traffic
server right now (really just 2 users for webmail). I leave my webmail
window open all night, which refreshes every 30 seconds so it shouldn't be
idle anyway. Also it starts right back up, so I do not think it has
anything to do with idle timeouts.

It seems funny that it exists with success, like if it will only run for
so long no matter what.

Any ideas? I have tried the latest php and mod_fastcgi cvs snapshots. It
appears to be a problem with php (since mod_fastcgi shouldn't be causing
the php process to exit, at least I don't think), but of course I can't be
certain.

after digging in some more I found out that it dies every 500 requests, so
I found a place in the php fastcgi code where it will exit after 500
requests.

I can't seem to find any comments in the code as to why this is so. Any
ideas?

I would like to see this error less, so I may up the limit.
However when I do agressive testing using the apache bench program, it
will die too much and stay dead for a while (due to some fcgi timeouts it
looks like), which seems bad as it could fail under high loads (although
it would start up again eventually).



On Wed, 20 Aug 2003 12:10:17 -0500, Nick Bartos wrote:

> I am running php with fastcgi. Everything appears to be working fine
> except for some errors in my logs, which I get about every 3-4 hours:
>
> [Tue Aug 19 17:20:48 2003] [warn] FastCGI: (dynamic) server
> "/var/www/php/php.fcgi" (pid 22390) terminated by calling exit with
> status '0'
>
> [Tue Aug 19 17:20:48 2003] [warn] FastCGI: (dynamic) server
> "/var/www/php/php.fcgi" (uid 45, gid 45) restarted (pid 26326)
>
>
> From my understanding, fastcgi is supposed to keep 5 servers running by
> default even if there is no demand for them. This is a very low traffic
> server right now (really just 2 users for webmail). I leave my webmail
> window open all night, which refreshes every 30 seconds so it shouldn't be
> idle anyway. Also it starts right back up, so I do not think it has
> anything to do with idle timeouts.
>
> It seems funny that it exists with success, like if it will only run for
> so long no matter what.
>
> Any ideas? I have tried the latest php and mod_fastcgi cvs snapshots. It
> appears to be a problem with php (since mod_fastcgi shouldn't be causing
> the php process to exit, at least I don't think), but of course I can't be
> certain.

I swear it never fails. No matter how much research I do before posting,
shortly after I post a question I seem to find the answer shortly
thereafter.

Anyway at the top of the php-fcgi main source file it mentions an
environment variable that can override the default 500 requests, to help
with possible memory leaks in php. I guess I can understand that
(although I would think the apache mod_php would need to do the same, if
thinking was consistant across the sapi's. not sure if it does though).
Anyway perhaps I have more faith in php than the developers do, but I have
increased the default to 1,000,000 requests. I may back that down later.

I would think that the default of 500 should perhaps be raised a bit. It
seems that during a burst of high traffic you could get 1500 (500 * 3)
requests in 30 seconds, which would cause mod_fastcgi to kill php-fcgi for
a while because it was restarting to quickly. I had this happen in
several test cases (which are not by any means real-world, but I am a
supporter of overkill most of the time). Of course mod_fastcgi will start
php-fcgi up again, but not after a few minutes. During that time the
apache won't be able to surve up any php documents.

After raising the limit to 1,000,000 I ran over a million requests (1 at a
time, as fast as my script could pump them out), and all seemed well. I
did not get any break in answering requests, or any errors in the logs
(execpt of course an error after 1 millino requests saying that php-cgi
died, 1 per million requests). Also I was able to browse other php stuff
on the server while running the test with not too much lag, and it never
skipped a beat.



On Wed, 20 Aug 2003 12:36:10 -0500, Nick Bartos wrote:

> after digging in some more I found out that it dies every 500 requests,
> so I found a place in the php fastcgi code where it will exit after 500
> requests.
>
> I can't seem to find any comments in the code as to why this is so. Any
> ideas?
>
> I would like to see this error less, so I may up the limit. However when
> I do agressive testing using the apache bench program, it will die too
> much and stay dead for a while (due to some fcgi timeouts it looks
> like), which seems bad as it could fail under high loads (although it
> would start up again eventually).
>
>
>
> On Wed, 20 Aug 2003 12:10:17 -0500, Nick Bartos wrote:
>
>> I am running php with fastcgi. Everything appears to be working fine
>> except for some errors in my logs, which I get about every 3-4 hours:
>>
>> [Tue Aug 19 17:20:48 2003] [warn] FastCGI: (dynamic) server
>> "/var/www/php/php.fcgi" (pid 22390) terminated by calling exit with
>> status '0'
>>
>> [Tue Aug 19 17:20:48 2003] [warn] FastCGI: (dynamic) server
>> "/var/www/php/php.fcgi" (uid 45, gid 45) restarted (pid 26326)
>>
>>
>> From my understanding, fastcgi is supposed to keep 5 servers running by
>> default even if there is no demand for them. This is a very low
>> traffic server right now (really just 2 users for webmail). I leave my
>> webmail window open all night, which refreshes every 30 seconds so it
>> shouldn't be idle anyway. Also it starts right back up, so I do not
>> think it has anything to do with idle timeouts.
>>
>> It seems funny that it exists with success, like if it will only run
>> for so long no matter what.
>>
>> Any ideas? I have tried the latest php and mod_fastcgi cvs snapshots.
>> It appears to be a problem with php (since mod_fastcgi shouldn't be
>> causing the php process to exit, at least I don't think), but of course
>> I can't be certain.