On 2020-07-08 at 15:39 -0700, David Gasaway wrote:
> On Wed, Jul 8, 2020 at 6:46 AM Ivan Shapovalov <[email protected]>
> wrote:
> 
> 
> > This specific filesystem is used to store a borg repository.
> > 
> 
> I currently use borg + s3ql myself.  I don't think I'd recommend
> it.  In my
> experience, borg check operations don't fare well in this scenario
> (as in,
> downloads of the entire repository).  I plan to switch to
> s3ql_backup.py +
> s3ql, but the prospect of the uploads needed to start from scratch
> are
> daunting.
> 

If I understand you right, then a `borg check` _has_ to download the
entire repository, because, well, it's an integrity check :)

But yeah, I don't hold any illusions — though I don't see much of a
choice either. I have a 8+ TB dataset containing some VMs, some random
assorted small files which come and go, some user shares with arbitrary
data and I need to snapshot+backup all that stuff for point-in-time
recovery.

Using S3QL's built-in quasi-snapshotting is a possibility, but I'm not
sure if rsync is going to be to any efficient with delta transfers of
VM images.

I guess I'll try my luck with various block sizes, see how large the
metadata grows and how efficient borg reads will be. I still need to
profile borg I/O patterns rigorously, but I believe that small random
reads are the main issue. Worst case I guess I'll have to write a
custom caching layer for partial blocks, maybe borg-aware.

-- 
Ivan Shapovalov / intelfx /

-- 
You received this message because you are subscribed to the Google Groups 
"s3ql" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/s3ql/709919d29857d843aea829e35939f45fae7071ef.camel%40intelfx.name.

Reply via email to