A small application of mine uses tee(1) in pipelines a lot to log the output of subshells. it very often hangs leaving a zombie process that is the parent of a tee(1) process. The ps -aef output for the tee(1) generally looks like this:
dougl 11469284 12517574 0 14:07:43 pts/8 0:00 tee /tmp//sh-np-4134995703

Typical use in this scenario is from within a function:
"${@}" 2>&1 | tee >(sed -n -e "s/^/${timestamp} ${caller} #### /g" -ep 2>&1 >> ${iss_log_file}) | (printf "${caller} " ; cat -)

The "${@}" are the args to the function and very often a pipeline itself.

Has anyone else experienced this? Is there a "natural" fix? (Think shell environment, configuration, or the like. Spawning a coprocess to keep an eye on such things is certainly possible, but seems wrong somehow.)

I'm on AIX 6.1. Bash is version 4.2.1.

Thanks.

,Doug

one thing I can see is the handling of stderr:
this would be better I think:
(sed -n -e "s/^/${timestamp} ${caller} #### /g" -ep >> ${iss_log_file} 2>&1 )
or you can try:
exec 2>&1
before this line (if it was made in a function)

Well, darn! Now I have to look closely at that use of stderr again.

There are cases where the calling function does want error output too.

Thanks.

Tick. Tick. Tick.

I'm back; I've looked. This is indeed one of those cases. Thanks for the critical eye.

this is not about stderr, but the order of the evaluation of redirections.
http://www.tldp.org/LDP/abs/html/io-redirection.html