Andrew Sullivan ajs at crankycanuck.ca
Tue Feb 23 10:58:18 PST 2010
On Tue, Feb 23, 2010 at 01:21:44PM -0500, Brad Nicholson wrote:
> from Slony.  It's the sort of data you'll need to really pinpoint what
> is going on and how to fix it.

Excellent point.  Brad's point in particular about how all this
interacts with other transactions is super-important.  It took several
major releases of Postgres for all those refinements to make it into
the system views, as people learned more of what they needed to do.
Trying to re-build that information out of the Slony tables is
hobbling yourself.

Moreover, accessing the system views doesn't affect those statistics
the way looking at the Slony tables will.  If you start grovelling
through the Slony tables all the time to find out how many updates you
have, you're going to have all kinds of cache hits and misses (and
index hits and so on) that will affect the system views, which don't
have any clue why you're doing these queries.  So you'll actually
cause countable activity when you're doing your counting.  The first
rule of a measuring stick is that it shouldn't actually modify the
thing measured.

Best,

Andrew "Schrödinger's database" Sullivan

-- 
Andrew Sullivan
ajs at crankycanuck.ca


More information about the Slony1-general mailing list