On Wed, 2002-01-30 at 10:56, Tom Lane wrote:
>
> FWIW, I believe that in typical scenarios there *is* no competition as
> the syscache never gets full enough to have anything age out. In the
> regression tests my little stats addition shows no run with more than
> 266 cache entries accumulated; the average end-of-run cache population
> is 75 entries. Syscache is currently configured to allow 5000 entries
> before it starts to drop stuff.
Are there _any_ tests where it does start to drop stuff ?
In other words - is the stuff-dropping part tested reasonably recently
(or at all) ?
> The regression tests are probably not representative, but if anything
> I'd expect them to hit a wider variety of tables on an average run than
> typical applications do.
>
> Bottom line: it's not apparent to me why the cache policy should be
> anything but straight LRU across both positive and negative entries.
In other words we should cache Frequently Asked Questions and not
Frequently Found Answers ;)
-------------
Hannu