[mapserver-dev] potential fcgi hang problem

[Bart van den Eijnden (OSGIS)]

Hi Frank,

I was told by our sys admin that the 600 ms cannot be changed 
unfortunately. Also if I look at the settings here:

http://www.fastcgi.com/mod_fastcgi/docs/mod_fastcgi.html

I don't really spot an option to change this setting nor the restart 
behaviour.

Best regards,
Bart

Frank Warmerdam wrote:
> bartvde at osgis.nl wrote:
>> Hi list,
>>
>> we are running Mapserver as a FCGI, and if someone tries an illegal mode
>> or something similar, it will cause all the MS threads to hang. We are
>> guessing this is because of the exit(0)'s in Mapserver's error handler.
>>
>> Is it possible for someone who is more knowledgeable to look into 
>> this? TIA.
>>
>> So e.g. trying a &mode=ma&layers=all request for about 10 times will 
>> cause
>> everything to freeze up.
>>
>> What we get is:
>>
>> [Wed Jul 23 09:53:28 2008] [warn] FastCGI: (dynamic) server
>> "/var/www/fcgi-bin/mapserv.fcgi" has failed to remain running for 30
>> seconds given 3 attempts, its restart interval has been backed off to 
>> 600
>> seconds
>>
>> [Wed Jul 23 10:04:11 2008] [warn] FastCGI: (dynamic) server
>> "/var/www/fcgi-bin/mapserv.fcgi" (pid 32486) terminated by calling exit
>> with status '0'
>> [Wed Jul 23 10:04:11 2008] [warn] FastCGI: (dynamic) server
>> "/var/www/fcgi-bin/mapserv.fcgi" has remained running for more than 30
>> seconds, its restart interval has been restored to 1 seconds
>
> Bart,
>
> Interesting.  My general thinking has been that the mapserv fastcgi
> could just exit if something goes wrong so we don't have to worry about
> things being in a bad state before processing another request.  That the
> system would just start a new fastcgi instance in it's place and that the
> only loss is that connections and stuff would need to be reopened.
>
> But I can see how errors could trigger this "back off restart" behavior
> by the apache mod_fcgi.
>
> I can think of two obvious approaches to deal with this.
>
> 1) Look into mod_fcgi settings and see if the restart-backoff stuff 
> can be
> reduced or removed.
>
> 2) We could start trying to continue processing a new request after an
> error in mapserv.  The main issue is that this will require touching
> many threads of control - for instance in the main mapserv.c logic.
> Everywhere (or at least at points of interest) where writeError() is
> called we would need to modify it to avoid an exit() and to be ready to
> process a new request after writing the error.
>
> My suggestion would be to first look at the apache side of things and
> see what can be done.  If that isn't helpful, we could do some error
> recovery stuff but it seems like the work could be endless, and it will
> add mess and complication to a mapserv.c mainline that is already
> horrendous.
>
> Best regards,


-- 
Bart van den Eijnden
OSGIS, Open Source GIS
bartvde at osgis.nl
http://www.osgis.nl