Re: FW: Subversion: [#6256] replication daemon

On 10 Jul 2002, Keith Bostic <bostic@sleepycat.com> wrote:
> > From: Martin Pool <mbp@samba.org>
> >
> > Somebody has promised to talk about a system very vaguely like that at
> > a symposium in Sydney this month. It looks like they basically just
> > echo every operation to all nodes.
>
> Does anybody know what happens when two processes update the same
> data item on different masters with conflicting information? The
> documentation doesn't say.

No, I don't know.

TDB has a simpler transaction model than DB (hence t for trivial).
Applications can explicitly lock either a key, or the whole database,
when they want to do an update. If you don't lock, then updates are
atomic but not isolated (if i have my terminology straight); whoever
gets in last wins.

I would guess (only a guess) that this system replicates those locks
across the network, so they can get the same behaviour whether
distributed or local.

My other question would be how they handle network partitions;
obviously you can't allow both databases to continue writing.

I'm going to AOSS4; I will be interested to see the talk.

-- 
Martin 
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org