From: Jim Blandy [mailto:firstname.lastname@example.org]
> I wonder what kind of meaning one can assign to transactions if one
> has a working copy assembled from multiple servers. Are transactions
> meaningful across multiple independent servers?
In the abstract sense, yes, transactions can be meaningful across separate
servers if the transaction involves both of the servers. Some systems
require this. Distributed transactions in SQL databases come to mind.
An example case for subversion might be:
* committing an APR change in the same change set with a working copy
Do we care? Even if we think we care, do we care enough to do the necessary
2 phase commit footwork?
I'd postulate that this isn't worth caring about unless we managed to
leverage someone else's distributed transaction coordinator.
Distributed transactions just get you into trouble when you try to scale
systems as well.
If subversion were to ever support replicating the version store out to
other servers, you certainly wouldn't want the replication to be transacted,
that just limits your scalability limit. Besides, this is a version store,
what better system could there possibly be for a merge replication scheme?
:) The store admin gets bugged when conflicts need manual resolution.
Received on Sat Oct 21 14:36:16 2006