Working directory and recovery methods
--------------------------------------

Whilst a backup is taking place, the server will create a directory in the
storage area for the client with a number and a timestamp. For example:

/var/spool/burp/<client>/0000027 2015-04-12 01:24:29

During backup phases 1 (file system scan) and 2 (send actual data), there will
be a symlink pointing to the directory called 'working':

/var/spool/burp/<client>/working -> 0000027 2015-04-12 01:24:29

During backup phases 3 (manifest generation) and 4 (shuffling), the symlink
will be renamed to 'finishing':

/var/spool/burp/<client>/finishing -> 0000027 2015-04-12 01:24:29

When the backup is complete, the symlink will be renamed to 'current':

/var/spool/burp/<client>/current -> 0000027 2015-04-12 01:24:29


If there is some error that happens during the backup and the backup is
interrupted, the burp server needs to decide what happens next. The behaviour
of what happens next is triggered the next time that the client contacts the
server.

If the interruption left a 'finishing' symlink behind, the server will attempt
to carry on and complete the backup. Part of phase 4 may involve operations
that alter the immediately previous backup (it may need to convert its files
into reverse deltas), so once it is 'finishing', burp can only try to move
forwards. Note that when the backup is 'finishing', no more data is required
from the client.

If the interruption left a 'working' symlink behind, the server will check the
'working_dir_recovery_method' server-side option to decide what to do next.
There are three choices:

 delete: Just delete the old working directory.

 resume: Simply continue the previous backup from the point at which it
         left off. NOTE: If the client has changed its include/exclude
         configuration since the backup was interrupted, the recovery method
         will automatically switch to 'delete'.

Note also that if the backup was interrupted in phase 1 (file system scan), the
recovery method will automatically switch to 'delete'. 


Setting 'delete' is the safest option (and also the default), because it means
that an entire backup run has to finish uninterrupted for a backup to be
considered complete. If you want to do Windows bare metal restores, you should
choose this option so that you don't end up with a mixture of files from
different VSS snapshots.

For 'resume', burp will use the working directory's original client file
system scan in order to request the remaining files that it needs to finish
the backup.
If you are using Windows (or some other OS that similarly generates browsable
file system snapshots for backing up), it will mean that files have been copied
from more than one snapshot, so the restore will be inconsistent if you are
considering a bare metal restore.
Additionally, there are occasionally bugs reported in the resume code, which
can be avoided if you use 'delete'.
Personally, I use 'resume' for my backups and have had no problems.


Some people have a lot of data, and need to give an initial backup extra help
to complete. They might set 'resume' until they have that initial backup, and
switch to 'delete' after that.