On Fri, Jan 7, 2011 at 06:30, Mister IT Guru <misteritguru AT gmx DOT com> wrote:

On 07/01/2011 12:23, James Harper wrote:
>>> Suggestion:
>>>
>>> Schedule the day's Incremental, then schedule the VirtualFull, say,
> 30
>>> minutes later.
>>>
>>> Put a RunBeforeJob script on the incremental that creates a lockfile
> (in
>>> a properly race-safe manner, of course) for the client.
>>>
>>> Put a RunAfterJob script on the incremental that removes the
> lockfile.
>>> Put a RunBeforeJob script on the VirtualFull job that checks for
>>> presence of the client's lockfile, and, if it finds it still
> present,
>>> sleeps for five minutes before checking again, and does not return
> until
>>> the lockfile has been gone for two consecutive checks (thus making
>>> certain there is a minimum of five minutes for attribute metadata
> from
>>> the job to be flushed).
>>>
>>>
>> Brilliant - sounds workable, I just don't know if my bacula skills are
>> up to it, I'm still very fresh to it, but the theory of your
> suggestion
>> is the closest I guess we can come. I will look into in - Thank you
>> bacula list :)
> I'm not completely sure, but I think that Bacula figures out what media
> it is going to use before it calls RunBeforeJob. This would mean that if
> you schedule your VirtualFull while your Incremental is running, the
> VirtualFull will not include the Incremental backup, no matter how much
> you wait inside the VirtualFull's RunBeforeJob script.
>
> Does anyone know for sure?
>
> James

I'm thinking that a workaround for this would be a script that checks if
the previous incremental has finished, and if it hasn't exit the job,
and then schedule a new job for $timenow+30mins. And then in 30 mins,
it'll check again. That way, no overlap"

What I just said works on paper, but may not be able to actually run in
bacula, I say that because google will index this msg and people may not
read the whole thread when they find this:)

I have a specific perl script that runs, and fires off the top x (in our case 3) vfulls each morning queued to run, based on what incrementals are seen running successfully the night before, and has had no fulls within x (in my case 90) days.

We did this because of the poor scheduling capability of bacula, and the desire to have constant aniversary based backups, instead of any kind of set "weekend full" backup schedule, which would be impossible to complete in a window.

I am happy to post if anyone wants.

-Blake