Latest TDB2 design and code...

Rusty Russell rusty at rustcorp.com.au
Mon Sep 13 03:47:05 MDT 2010


On Sun, 12 Sep 2010 11:22:27 am tridge at samba.org wrote:
> Hi Ronnie,
> 
>  > But I really really really would like to see
>  > cheap read-only snapshots, and cheap compute delta between two snapshots.
> 
> why? I suspect snapshots will add a lot of complexity to the code, so
> what is the payoff for that complexity?
> 
> I suspect you want this for inter-cluster ctdb. Perhaps instead you
> can put the snapshot support in ctdbd itself, and use a minimal set of
> hooks in tdb that allow you to record changes, similar to the locking
> hooks that Rusty is proposing.

Or a TDB wrapper layer which records "here's what I did" (you could put it
in a TDB!).

But offline Tridge suggested stealing the ctdb replay code which does this
for cross-cluster transactions.  That would also work nicely for tracing...

Best of all, the TDB2 extensible attribute mechanism makes this easy to
add later.

Thanks,
Rusty.


More information about the samba-technical mailing list