Editing a file in-place with SED seems to prevent any further append operations by processes that are already running

Question

I have a log file that is written to by a server. I wrote a bash script to send me an email if there is an error in the server. I would now like to remove the lines containing the errors so I don't keep getting emails. I accomplish this by doing the following:

sed -i "/WARNING/d" logs/console.log

After running sed however, no more changes are written to the log. I'm guessing this is because running sed closes any open file descriptors or something. However, when I edit the file and manually remove the warning lines with vi I don't have this problem.

I have also tried redirecting the server output myself with both '>' and '>>' operators and after editing the file with sed the same thing happens (i.e. they are no longer updated).

Answer 1

When sed rewrites the log file the server process probably gets an IO error and doesn't try to reopen the log file. I don't know if this approach can work out. sed definitely doesn't have flags to tweak the way it rewrites files with the -i flag, and I don't know if the server can be tweaked to be more resilient when appending to the log.

So your best option might be a different approach: save the timestamp of the last error look for errors after that timestamp. Something like this:

ts=
file=console.log
while :; do
    if test "$ts"; then
        if sed -e "1,/$ts/d" $file | grep -q WARNING; then
            sed -e "1,/$ts/d" $file | sendmail ...
            ts=$(tail -n 1 $file | cut -f1 -d' ')
        fi
    else
        if grep -q WARNING $file; then
            sendmail ... < $file
            ts=$(tail -n 1 $file | cut -f1 -d' ')
        fi
    fi
    sleep 15
done

This script is just to give you an idea, it can be improved.