Re: [PATCH v4 17/18] qapi: backup: add immutable-source parameter

[Vladimir Sementsov-Ogievskiy]

24.02.2022 16:05, Hanna Reitz wrote:
> On 16.02.22 20:46, Vladimir Sementsov-Ogievskiy wrote:
> > We are on the way to implement internal-backup with fleecing scheme,
> > which includes backup job copying from fleecing block driver node
> > (which is target of copy-before-write filter) to final target of
> > backup. This job doesn't need own filter, as fleecing block driver node
> > is a kind of snapshot, it's immutable from reader point of view.
> >
> > Let's add a parameter for backup to not insert filter but instead
> > unshare writes on source. This way backup job becomes a simple copying
> > process.
> >
> > Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> > ---
> >   qapi/block-core.json      | 11 ++++++-
> >   include/block/block_int.h |  1 +
> >   block/backup.c            | 61 +++++++++++++++++++++++++++++++++++----
> >   block/replication.c       |  2 +-
> >   blockdev.c                |  1 +
> >   5 files changed, 69 insertions(+), 7 deletions(-)
>
> I’m not really technically opposed to this, but I wonder what the actual 
> benefit of this is.  It sounds like the only benefit is that we don’t need a 
> filter driver, but what’s the problem with such a filter driver?

Hmm. Yes, that's the only benefit: less extra components -> more stability.

But I doubt now does it really worth extra parameter.. More parameters that 
actually change nothing for the user -> less stability :)

Ok, I think I at least should postpone it for now, this series is too fat even 
without this patch.

The only possible problem - will permission conflict happen in the next test 
without this patch? But if it will, the solution should exist to solve it 
without user interaction. I'll check and try to avoid this new parameter.

> (And if we just want to copy data off of a immutable node, I personally would 
> go for the mirror job instead, but it isn’t like I could give good technical 
> reasons for that personal bias.)

I still hope that in far good future mirror will work through block/block-copy 
like backup, and there would be no difference what to use for immutable source 
copying.


Thanks a lot for reviewing!

--
Best regards,
Vladimir