Re: computational model of transactions

Erwin wrote:
> PS
>
> AFAICT, MVCC is only safe to use in transactions that don't update
> anything, i.e. transactions that do *nothing but reads*.

Erwin,

I was under the impression MVCC could assist in updates as well.

For example, consider updating two tables, A and B, concurrently, with the update to A starting first. The rows of Table A to be updated depend on values of Table B. (A has a FK pointing to B). The two updates can proceed concurrently if the update to Table A has a consistent view of table B, which it can get because the update to B is not in-place, that is, there are multiple versions of B.

But I might well be wrong. Received on Fri Aug 04 2006 - 16:51:20 CEST