Saulius Grazulis wrote:
>> In practice, half-completed commits are under txns anyway, and are put
>> into revs with a single mv, so you either get them or not regardless.
>
> Thanks for explanation, it sound very reassureing.
In the worst-case (if it does get done in the wrong order) the current file
mentions a new revision that was not synced, and then you have to open in
in a text editor and type in the highest revision that you did get in
(which is easy to find with a glance into revs/). FSFS is pretty robust to
ordinary incremental backup strategies.
>> if one wants to get really fancy, if you
>> * read 'current'
>> * sync revs
>> * write the value of 'current' you originally read
>>
>> and than the target mirror is also consistent the entire time, since
>> having revisions showing up under revs that aren't referenced yet is
>> harmless.
>
> I guess re-running rsync some time later would be a simplier and would
> also advance the 'current' (yes, I understand that I can again be missing
> the very latest revision, but if the goal is to keep backup reasonably
> close to the head, re-rsyncing sounds to me like a simplier solution, in
> the light of what you explained above -- right?).
Might be, though what I described is also pretty simple. The advantage to
doing this is nothing for a backup, but it's nice for mirroring as then the
target is also valid during the entire time that the backup is running.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Thu Apr 14 15:12:35 2005