[H-SASIG] rdiff-image-cron /etc/rdiff-image/rdiff-image.conf FAILED!

[Russell Stuart]

On Tue, 2010-02-02 at 15:15 +1000, Raymond Smith wrote:
> When does your backup script delete files? Do deletes only happen with
> each full backup or do they happen more often?

When the number of backups on S3 exceeds the number required to
implement the backup cycle specified in the configuration file.
Currently the backup cycle in the configuration file is:

24x1hour,7x1day,5x1week,6x30days,6x182days

> Also, any idea if S3 directory creation suffers from the same
> eventual-consistency problem as files?

Yes directories must suffer from the same problem.  It is a consequence
of the CAP Theorem.  See:
http://devblog.streamy.com/2009/08/24/cap-theorem/

> > So is it only the noisy emails.  There is a simple solution to them if
> > it continues - just don't report files appearing out of thin air as
> > evidence of someone else modifying the backups.  Check for file
> > deletions and modifications only.
> 
> This would be a great improvement -- noisy system emails will end up
> being treated just like spam.

I will fix it one way or the other if this problem continues.  But I am
not 100% certain I know what is going on.  It is not the easiest thing
to get visibility into.  I log every interaction with S3, and this log
is what I using to figure out what happening.  The log agrees with the
error message, however it is possible both could be wrong ..

> What I meant to say here was that if we prefixed with a
> unique-to-(virtual)machine artifact then there would be no problem
> with different instances of excalibur overwriting each other. We might
> then dispense with the backup is stuffed check.

There are a couple of issues with that.  Firstly, storing those unused
backups on S3 forever more costs money.  Secondly it makes restoring
even harder - how to you know which "artifact" is the most recent one?
Restoring is difficult as it is - I had write a wrapper around because
it appears everyone found too complex to do by hand.

The backup system is still a work in progress right now.  I have done 5
releases since I originally announced it here.  Nothing major - they
fixing minor little irritations as you can see from the change log.
There is at least one more release to come, and it won't address this
problem.