Thank you.

Sorry, I don't want to monitor the jobs all the time they're running, just "attach" to a already-running job once he's dead. We got a tool which checks the estimated maximum run time for each script and creates some seconds strace and a lsof report for the process before killing it.

Here is what I did so far:

open my $strace_logger,'>','/dev/null';
$strace_logger->autoflush(1);
[...]
print $strace_logger "SHORT STATUS MSG";
[download]

There is no output out of it (everything is going to /dev/null, so no I/O usage, but the print triggers a syscall visible in strace - including some bytes of the content.