Your script redirects stderr to the original stdout of the script. To redirect both to /dev/null, the 2>&1 must follow the >/dev/null like this:
nohup caffeinate -s bacula-idle-watch.sh >/dev/null 2>&1 </dev/null & Bacula 9.2.0 collects the stderr from ClientBeforeJob commands, but Bacula clients running 7.4.x would have worked because they discard it. If that doesn't fix it, then try redirecting the shell's stdio as well: exec >/dev/null 2>&1 </dev/null nohup caffeinate -s bacula-idle-watch.sh & __Martin >>>>> On Fri, 4 Jan 2019 11:36:06 -0800, David Brodbeck said: > > This is driving me nuts because I feel like it should be straightforward > and I must be missing something basic. > > I want to launch the caffeinate command on OS X before starting a job. > Caffeinate takes a command as an argument, then goes into the background > and keeps the machine awake until the command exits. I use this after > waking machines up using a WOL script. > > When tested from the command line, caffeinate immediately backgrounds > itself. However, when I try to run it as a Bacula ClientBeforeJob script, > bacula-fd waits around forever for caffeiniate to exit. > > Here's what I've tried so far: > - Having bacula run a script that then runs caffeinate. > - Having bacula run a script that then runs caffeinate using nohup. > - Having the script redirect stdin, stdout, and stderr of caffeinate to > /dev/null > - Adding an ampersand after the script in the bacula ClientBeforeJob > specification. > > What invariably happens is the bash process created by bacula becomes a > zombie and waits for caffeinate to exit. Inspecting the caffeinate process > with lsof shows all of the file handles are redirected to /dev/null as > expected, so I don't think this is a case of stdin or stdout causing > problems. In all cases the only way to get bacula to finish the backup is > to kill the script that caffeinate is running. > > I can't figure out why I can't get bacula-fd to move on after the script > goes into the background. When I run the script manually from the command > line it backgrounds immediately. > > The oddest thing is this worked fine on clients using bacula-fd version > 7.4.x, but fails on a client using 9.2.0. > > Here's the script bacula-fd runs, as it currently stands: > > -- > #!/bin/bash > > PATH=/bin:/usr/bin:/usr/local/bin > > # Script to prevent system sleep while bacula is working. > # see bacula-idle-watch.sh for details. > > nohup caffeinate -s bacula-idle-watch.sh 2>&1 >/dev/null </dev/null & > -- > > Here's the contents of bacula-idle-watch.sh; it just waits to exit until > there's no bacula network connection anymore. caffeinate will terminate > once the script exits. > > -- > #!/bin/sh > > # This script delays a few minutes, then loops, checking for bacula-fd > # connections. When there are none, it exits. > # This is meant to be run with caffeinate in a bacula before-job script, > # e.g. "caffeinate -s bacula-idle-watch.sh" > # This will prevent the machine from idle-sleeping until bacula finishes. > > PATH=/bin:/usr/bin:/usr/sbin > > # We put a long delay here in case it takes bacula a while to get going. > sleep 300 > > # Now loop while looking at the network connection. > # We limit checks to once every five minutes because worst-case the machine > # just waits an extra five minutes to sleep. > while ( netstat -an | grep '\.9102.*ESTABLISHED' >/dev/null ) ; do > sleep 300 > done > > # Once the script exits, the wake-lock is released. > exit 0 > -- > > > -- > David Brodbeck > System Administrator, Department of Mathematics > University of California, Santa Barbara > _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
