[Bacula-devel] Change in "AfterJob" and "WriteBootstrap" execution order from 1.38 to 2.0.3 ?

[Timo N. <tim...@te...>]

I made a complete re-install of my file/backup server, from FC5 to
CentOS (RHEL) 5.
At this time, I also upgraded Bacula from 1.38.11 to 2.0.3, the latest
stable release available as rpms.

My configuration uses RunAfterJob to invoke a shell script that emails the
bootstap file to me. The idea is to get a copy of the bootstrap file
immediately stored somewhere outside the physical backup server.

This used to work (yes, I checked the file contents of the old emails) in
the older system.
However, after upgrade to 2.0.3, the RunAfterJob emails me the bootsrap file
of the previous run of this backup job, which means that the order of
writing the bootsrap file and execution of the runafterjob command has
changed.
Otherwise I propably wouldn't have noticed this at all, but I started
investigating why I didn't receive the email after the very first backup -
the file couldn't be emailed since it didn't exist yet. After the following
runs, I now keep receiving the one-step-too-old file.

Director conf file has the following in job definition
----
  Schedule = "WeeklyCycle"
  Write Bootstrap = "/var/bacula/dogbert.bsr"
  RunAfterJob  = "/etc/bacula/after_dogbert"
----

/etc/bacula/after_dogbert that is referred above is essentially this:
----
#!/bin/sh
# Emails bootstrap
/usr/bin/nail -s Dogbert-Bootstrap-S -a /var/bacula/dogbert.bsr xxx@yyy.zzz
</dev/null >/dev/null
----

This is (part of) what I got into log after the first run of this job:
----
30-Jun 01:41 dogbert-dir: AfterJob: run command "/etc/bacula/after_dogbert"
30-Jun 01:41 dogbert-dir: AfterJob: /var/bacula/dogbert.bsr: No such file or
directory
  Scheduled time:         30-Jun-2007 01:05:00
  Start time:             30-Jun-2007 01:05:04
  End time:               30-Jun-2007 01:41:10
----

This makes me think, that there has been a code change from 1.38 to 2.0 that
changes the order in which RunAfterJob and WriteBootstrap are processed. I
find the old order more logical - RunAfter really meant running after
everything(?) else was done.

Could some of the developers confirm if there has been an accidental or
intentional change on this, and is it reasonable to expect it to start
behaving in the old way again (or has it already been fixed in 2.1 tree,
maybe)?


Regards,
Timo