Re: fsfs recover and missing revprop

[Jeremy Whitlock <jc...@gmail.com>]

> That sounds good.  IIRC, the current implementation does an exponential search
> for the largest missing revision, then a binary search for the largest existing
> revision.
>
> So the recover instead should do an incremental search from revision 0 to ensure
> that all revs and revprops exist.  I think this is necessary because you can
> have a backup tool that backed up all the revs but was in the middle of backing
> up the revprops, or vis-versa.

Blair,

The initial patch I've submitted to this does not alter the way you've
done your exponential search and works completely within the
constraints you've set.  I can easily modify your code and extend it
to do the revs/revprops check for each revision, which makes the most
sense and could then be used for 'svnadmin verify'.  Do you mind
looking at the second patch I submitted to Issue 2992 and reviewing it
as-is?  Once we get that patch applied, I'll expand it.

Take care,

Jeremy

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: fsfs recover and missing revprop

[Blair Zajac <bl...@orcaware.com>]

Jeremy Whitlock wrote:
>> That sounds good.  IIRC, the current implementation does an exponential search
>> for the largest missing revision, then a binary search for the largest existing
>> revision.
>>
>> So the recover instead should do an incremental search from revision 0 to ensure
>> that all revs and revprops exist.  I think this is necessary because you can
>> have a backup tool that backed up all the revs but was in the middle of backing
>> up the revprops, or vis-versa.
> 
> Blair,
> 
> The initial patch I've submitted to this does not alter the way you've
> done your exponential search and works completely within the
> constraints you've set.  I can easily modify your code and extend it
> to do the revs/revprops check for each revision, which makes the most
> sense and could then be used for 'svnadmin verify'.  Do you mind
> looking at the second patch I submitted to Issue 2992 and reviewing it
> as-is?  Once we get that patch applied, I'll expand it.
> 
> Take care,
> 
> Jeremy

Jeremy,

Thanks for the patch, I'll look at it this week.

BTW, the fsfs recover work was done by Malcom.

Regards,
Blair

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org