Entire website impacted by fastcgi application error?

I’ve got a situation where I’m running two fastcgi
servers, with 3 processes per server. So URL-A goes to
serverA’s queue, and is handled by one of the three
processes for that server. Pretty standard stuff.

Now I get a transient condition in the backend that
causes every process in serverA to exit (uncleanly).
When fcgi_pm starts a new process it exits
immediately. Very soon this behavior trigger’s
fcgi_pm’s "bad" designation where it says "serverA has
failed to remain running for longer than 30 seconds
given 3 attempts – changing its restart delay to 600
seconds".

That is ok and expected (actually, it’s not ok but
that isn’t fastcgi’s fault!).

However, as traffic continues to come in we see the
apache worker processes attempt to connect to serverA
and hang there waiting for a timeout. This has two
effects: 1) users to the site see a long delay and
then a status 500 response and 2) very quickly all or
almost all of the apache workers are in the "waiting
for connection to serverA" state.

The net effect is now my whole server is almost
completely out of operation, rather than just the
URL-A application.

I note that fcgi_pm sets the "bad" flag on serverA
when it is in this state. Is there a reason why
mod_fastcgi can’t notice that flag and short-circuit
the connection attempt?

Has anyone seen this scenario? How did you get around
it? Are there existing configuration / build options
that would help?



__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
___________________________________
fastcgi-developers mailing list
http://fastcgi.com/fastcgi-developers/