[syslog-ng] Restarting destination program is dangerous

[Bryan Henderson]

The new feature that restarts a 'program' destination when it exits is
pretty dangerous.  Whatever caused the program to exit might be
persistent, so that every time you start it you exit, causing an
infinite loop of forking and execing.  I discovered this the first
time I tried to use current Syslog-ng, when the specified destination
program didn't exist.

I think this should be optional behavior.

Another idea that would help with this is not to do the restart until
you have a message to deliver.  That way, you'd have at most one fork
per message.

In lieu of an option, I am presently running modified Syslog-ng in
which I simply remove the log_pipe_init() from afprogram_dd_exit().
Is that sufficient?


I also think that, philosophically, Syslog-ng shouldn't be involved in
restarting failed programs, because it isn't a problem special to
syslog.  I think it's better done by a generic monitor/restarter
program.  For example, I have a program called 'respawn' that spawns a
child and restarts it when it dies.  It has various parameters to
limit how often it's willing to fork.  I tell Syslog-ng destination
program(respawn logstuff ...).

-- 
Bryan Henderson                                   San Jose, California