On 21.09.2009 13:41, Jeff Trawick wrote: > On Mon, Sep 21, 2009 at 4:01 AM, Rainer Jung <rainer.j...@kippdata.de > <mailto:rainer.j...@kippdata.de>> wrote: > > On 21.09.2009 09:11, Paul Querna wrote: > > While, this is obviously a terrible thing for a general purpose > > module, I would like to propose that we add a 'FCGIIgnoreVirtualHost' > > configuration option, which would set the vhost field to a constant > > value, so FCGI processes would be shared between virtual hosts. > > +1, I assume you meant this directive to be used per virtual host, not > as a global setting? > > Or maybe even a 'FCGIGroup xyz' that lets you choose the string used > instead of the virtual host name (if set). So you can combine according > to your needs? On the positive side that would be more flexible, on the > negative side you would always need to add an argument to the directive > (or find a better directive name, that makes it more obvious, that with > a missing argument it defines a default group). > > > I like this in general. Having to supply an argument doesn't sound > negative to me. > > It is probably inevitable, but the term "Group" here is potentially > confused with the existing term "Class" (as used in > DefaultMinClassProcessCount and DefaultMaxClassProcessCount).
I noticed an enty in our Chnagelog for mod_fcgid that actually dates back to August 2007 (r753578): 1. Patch from Gabriel Barazer, gabriel at oxeva.fr When setting multiple virtual hosts with the same SuexecUserGroup user and group, the process manager use the same process pool for both virtual hosts. This means if one virtual host has a DefaultInitEnv and the other has different values set, a fastcgi request from any of these virtual host can go to the same processes, which is inconsistent (a request from virtualhost a with DefaultInitEnv VAL "a", can go to a process spawned with virtualhost b with DefaultInitEnv VAL "b" set) So this reminds us, that if we want to share processes over the boundaries of VirtualHosts, we have a problem with all configuration which afftecs those processes and is possible per vhost. So in order to make the results predictable we need to look for clear semantics of such a process sharing. I currently do not have one. Regards, Rainer
