Possible bug in scheduler/transport agent interaction

[Andy Poling <andy@jhunix.hcf.jhu.edu>]

This is taking place with zmailer 2.2 running under Irix 4.0.4

I haven't really nailed this down yet, except through tracing of the
wedged processes when it happens, but here's my take on it.

There appears to be a bug in the interaction between scheduler and its
child proceses - the transport agents.  What I am seeing is scheduler is
wait()ing on its children, but the children are blocking on a full pipe
between them and scheduler.  Can you say deadlock?

This seems to only become evident if you have rather large amounts of
mail queued for a particular channel (local in our case) -- leading to
the transport agent filling the pipe with diagnostics for scheduler before
processing the entire queue.

I'm not terribly familiar with the zmailer code, but why scheduler is
wait()ing instead of reading seems to be the question to ask...

-Andy

Andy Poling                              Internet: andy@jhunix.hcf.jhu.edu
UNIX Systems Programmer                  Bitnet: ANDY@JHUNIX
Homewood Academic Computing              Voice: (410)516-8096    
Johns Hopkins University                 UUCP: uunet!mimsy!aplcen!jhunix!andy