Also see this in ALPHA at CERN. Eventually there are so many defunct elogd processes that the user runs out of "maxproc" quota and automatic submission 
of elog messages starts to fail. (and the users complain, reboot all computers, etc).

The elogd we use is this:
https://bitbucket.org/ritt/elog/commits/44800a769b99599db7620779e2142b1161c694fc?at=master

The best I can tell, the main elogd is spawning something but does not reap finished subprocesses (wait() syscall). My guess it is spawning ImageMagik stuff 
to create preview images.

K.O.