Mercurial and incremental backups...

[Peter Arrenbrecht]

On Fri, May 30, 2008 at 10:45 AM, Marcin Kasperski
<Marcin.Kasperski at softax.com.pl> wrote:
>
> How effective - in terms of transfer and datasize overhead - is
> incremental backup of .hg/ directories?
>
> And what is the most effective method in case one would like to
> optimize it as far as possible? Generating consecutive bundles and
> uploading them? But is it easy to automate?

Sounds like a good approach. Should be fairly easy, either using tags
or a separate clone that records the state of the last backup. Then
simply bundle against this clone/tag and update the clone/tag
afterwards. Don't know how well that plays with multiple branches,
though. I use the separate clone approach to keep track of what I need
to ship my customer in the next instalment (don't have access to his
repo, but like a backup it's a one-way street).

>
> In general - do you have any recommendations for effective backup
> of Mercurial repos?
>
> (background: I am just considering backup to S3 via jungledisk or sth
>  similar, there you pay for both the data transfer and the total size
>  of files stored, it is in particular bad idea to make frequent small
>  changes to big files as changed files are transferred as a whole)

Then you definitely want to backup bundles, not the raw .hg/store. I
have been tossing about the idea of automatically splitting revlog
files when they reach a certain size. My intention was to improve the
effectiveness of the local hardlinks for multiple clones, but it would
help incremental backups as well. Nothing to show yet, though. Just
ideas. But if there is sufficient interest, I might give it a go. So
far nobody showed much interest.

-parren