[rdiff-backup-users] Problem with Detection of Multiple rdiff-backup instances

[Dean Cording]

I've come across an issue with the way that rdiff-backup ensures that only one 
server is accessing a backup dataset.

When rdiff-backup starts it checks the metadata to see if another instance of 
rdiff-backup is performing a backup.  If it finds one then it checks the PID to 
see if the other instance is still running and if not, assumes the other 
instance has crashed so it regresses the previous incomplete backup.

I am running rdiff-backup on Amazon's cloud computing resource.  Whenever I 
want to backup I start a new virtual server, run rdiff-backup and then shut it 
down.

Recently I had a backup fail, probably because of a network outage.  All 
subsequent backups refuse to run because rdiff-backup believes the failed rdiff-
backup instance is still running - even though this is clearly impossible 
because it is a totally different instance of the virtual server.

This had me stumped for a while but I finally figured out what is happening.

Because I start a new virtual server instance each time and I run the backup 
from a script, everything happens in a consistent order.  As a result the 
instance of rdiff-backup running on the server for each backup session almost 
always has the same PID.  So when a backup fails, the subsequent backup looks 
at the metadata, finds the PID of the failed backup and sees that that PID is 
still running - not realising that the other instance is actually itself.

I'm not sure of a way of working around this problem as the virtual machine is 
always started from a known state and hasn't been running long enough to build 
up any entropy to generate unique random numbers between different sessions.