I am trying to tidy up the following snippets, design goals are to log all output from a script, and should not be a wrapper. Less lines are better.

I don't care about user inputs (at this stage), target scripts are run non-interactively.

The snippet needs to

  • output stdout to log, and always echo to console
  • output stderr to log, and echo to console iff debugging is enabled
  • stderr messages should be prefixed with time stamps and other usefulness

At the moment I have the following which only tests out under recent versions of bash (4.2+?) like in Ubuntu precise, but misbehaves on CentOS6.


# copy stdout to log always and echo to console
exec >  >(tee -a ${DEBUG_LOG})           

# copy stderr to log only, unless debugging is enabled
[ $DEBUG_TEST = "true" ] \
  && exec 2> >(tee -a ${DEBUG_LOG} >&2) \
  || exec 2>> ${DEBUG_LOG}

Then this...

# Expand escaped characters, wrap at 70 chars on spaces, 
# and indent wrapped lines
msg_log() { 
  echo -e "$(date +%T) ${0##*/}: $1" \
    | fold -w70 -s | sed '2~1s/^/  /' >&2; 
msg_con() { 
  if [ "${DEBUG_TEST}" = "true" ]; then 
    msg_log "$1"
    echo -e "$1" | fold -w70 -s | sed '2~1s/^/  /'; 

Instead of echo I can call one of those msg procedures, e.g., msg_con "hello world".
Also script output will then go to stderr by setting as an environment variable at call time, e.g., DEBUG_TEST=true myscript.

I have read that exec may not work in some shells such as busybox. There is a mkfifo and fork combination at https://stackoverflow.com/a/5200754 that does something similar but I'd rather not use fork unless absolutely needed.

Prefer bash examples please, but something that works under sh or is more portable would be nice. Any ideas?

4 Answers 4

function startLogging {
    exec > >(gawk -v pid=$$ '{ print strftime("%F-%T"),pid,$0; fflush(); }' | tee -a $logfile)
    [ ! -z "$DEBUG" ] && exec 2>&1 || exec 2> >(gawk -v pid=$$ '{ print strftime("%F-%T"),pid,$0; fflush(); }' >>$logfile)
    echo "=== Log started for $$ at $(date +%F-%T) ==="

You need to have $logfile set to something

  • This is kinda cool, the way I'm reading it you use gawk to add the message headers. That will have the additional side effect of adding those to command outputs as well.
    – Glenn
    Commented Oct 5, 2013 at 9:14
  • The reason to use gawk to add a timestamp (and process id) to each log line is because it will get executed each time output is written and therefore update the timestamp. Also it's important to use gawk and not awk because I think the strftime function is a GNU extension.
    – Angelo
    Commented Oct 9, 2013 at 17:41

exec > filename should work in sh, and it actually works in busybox v1.15.3 (Nov 2011). But the process substitution >(command) is unportable since it's bash extension. Just avoid using it in scripts. Why >> isn't enough for you?

exec 1>>${DEBUG_LOG}
exec 2>>${DEBUG_LOG}

Another solution is to specify redirection outside of your scripts. When your script is invoked in background (by cron, or system script, etc.), they should be called like this

./my_script 1>>${DEBUG_LOG} 2>>${DEBUG_LOG}

When you invoke the script manually and you want to see the output, just call it without redirections.

  • 1
    The question asker wants the output of the script to go to both the console and the log file.
    – user26112
    Commented May 4, 2013 at 11:29

These two examples will do what your stated objectives are

echo -n $(date) >> $DEBUG_LOG
command 2>&1 | tee -a $DEBUG_LOG


echo -n $(date) >> $DEBUG_LOG
command >> $DEBUG_LOG 2>&1

You could use tee command or script command both are really useful.

  • script looks cool, but does not really do what I need, the behavior has to modifiable during runtime.
    – Glenn
    Commented Oct 5, 2013 at 8:54

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .