Re: How is statement level read consistency implemented?

Roberts, Jon wrote:

> > > -With autovacuum, does it skip these rows still being referenced
> > > in a transaction or does it wait?
> >
> > It skips them, the idea being that a future vacuum will remove them.
>
> Awesome.  In a large data warehouse, the snapshot too old error is very
> annoying and I'm glad PostgreSQL is superior to Oracle in this regard.
> :)

Well, the disadvantage of the PostgreSQL way is that it keeps dead rows
around for longer than they're actually needed, and so it causes some
problems in pathological conditions -- for example when setting up large
replication sets with Slony, or during a pg_dump, no dead rows can be
removed.  Since the Slony thing can take a very long time, dead rows
start to pile up in a way that can really harm performance.

We're currently working on it so that it is less of a problem for 8.4 ...

--
Alvaro Herrera                                http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support