At work we use cronolog for automatic rotation of log files for a number of processes since those processes just write to STDOUT and STDERR instead of using a proper logging library. Unfortunately, that means when running the script/program we have to redirect STDERR to STDOUT, and then pipe the results to cronolog, since cronolog reads from STDIN. The result looks something along the lines of the following:
ruby main.rb &2>1 | cronolog /logs/main.log /logs/main-%Y-%m-%d.log
The problem with this is if that errors are few and far between, as one hopes they should be, then it might be really tricky to find the errors amongst the other logging. Ideally, I thought it would be nice to have STDOUT go to one log file, and STDERR get written to a err file for the process.
After some digging into From Bash to Z Shell I found something about process substitution in the Bash shell. After a little experimentation and tweaking, I came up with the following:
ruby main.rb > >(/usr/sbin/cronolog /logs/main.log /logs/main-%Y-%m-%d.log) 2> >(/usr/sbin/cronolog /logs/main.err /logs/main-%Y-%m-%d.err)
This allows me to use cronolog with both the STDOUT and STDERR streams. By using cronolog in the process substitution, it allows the output streams to be treated as input streams to cronolog, where as before I had to combine them into one stream and then pipe the single stream to cronolog as in the first example.
Hope this can help someone else, and save some hours of digging.