On Mon, Nov 12, 2012 at 5:17 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
> On 12 November 2012 16:51, Robert Haas <robertmhaas@gmail.com> wrote:
>
>> Although there may be some workloads that access very large numbers of
>> tables repeatedly, I bet that's not typical.
>
> Transactions with large numbers of DDL statements are typical at
> upgrade (application or database release level) and the execution time
> of those is critical to availability.
>
> I'm guessing you mean large numbers of tables and accessing each one
> multiple times?
Yes, that is what I meant.
>> Rather, I bet that a
>> session which accesses 10,000 tables is most likely to access them
>> just once each - and right now we don't handle that case very well;
>> this is not the first complaint about big relcaches causing problems.
>
> pg_restore frequently accesses tables more than once as it runs, but
> not more than a dozen times each, counting all types of DDL.
Hmm... yeah. Some of those accesses are probably one right after
another so any cache-flushing behavior would be fine; but index
creations for example might happen quite a bit later in the file,
IIRC.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company