Question on pid caching

[Tymm Twillman <ttwillman at penguincomputing dot com>]

Is there a way to indicate to glibc that a cached pid should be tossed?  
Specifically, with the implementation of a remote fork (for a clustered 
system), we're finding bugs due to interactions with glibc becasue of 
the caching of process IDs.   Right now we're seeing this if getpid() is 
called before the remote fork. The specific version of glibc we're 
working with currently is 2.3.4 as distributed w/RH4, though we'll need 
to keep things going with future versions.

Please note this is a case of a different implementation of fork() -- 
I'm not trying to get into discussions over whether pids should be 
cached, just want to know how to indicate to glibc that a cached pid 
will no longer be valid; preferably cleanly and reasonably transparently 
for folks using the clustering libraries.

Thanks,

-Tymm