Question

I’m using libreoffice/openoffice as a headless process to handle some document conversion tasks on my server that I “submit” via unoconv. Once in a while, the process that actually does the work, soffice.bin, seems to get wedged. I tried playing around with strace and saw that when launching new unoconv instances, they still connect and talk to the soffice process, just that nothing else happens after the ‘bad’ document goes in. If it were so simply as to just detect that soffice does not talk to incoming sockets any more, it’d be easy to write a watchdog. But it’s not that simple, apparently. Any ideas how to tell when things have gone south?

Answer 1

Here’s what I set up as a cron job:

def monitor_unoconv
  retval = false
  target_dir = "/tmp/monitor_unoconv"
  begin
    Timeout::timeout(30) do
      FileUtils.mkdir_p(target_dir)
      FileUtils.cp(File.dirname(__FILE__) + "/../hello.odt", target_dir)
      Dir.chdir target_dir do
        retval = system("unoconv -f html hello.odt")
      end
    end
  rescue => e
    STDERR.puts "Caught error #{e.inspect}"
    retval = false
  end
  if !retval
    STDERR.puts "soffice process appears hung.  Killing it"
    STDERR.puts `killall soffice.bin`
    sleep 5
    STDERR.puts `killall -9 soffice.bin`
  end
end

Seems to work ok.