On Tue, Nov 24, 2015 at 5:29 AM, Robert Haas <robertmhaas@gmail.com> wrote:
> On Mon, Nov 23, 2015 at 8:45 PM, Peter Geoghegan <pg@heroku.com> wrote:
>> Beyond that, CREATE INDEX and CLUSTER utility
>> cases will also need to be parallelized without all this executor
>> infrastructure.
>
> Or, alternatively, CREATE INDEX and CLUSTER could be refactored to use
> the executor. This is might sound crazy, but maybe it's not. Perhaps
> we could have the executor tree output correctly-formed index tuples
> that get funneled into a new kind of DestReceiver that puts them into
> the index. I don't know if that's a GOOD idea, but it's an idea.
Having CREATE INDEX use the executor seems like a useful idea for
reasons unrelated to parallelism.
The use case I have in mind is a table containing multiple years worth
of (approximately) time series data, where overwhelming majority of
queries are explicitly interested in recent data. Having a partial
index with WHERE tstamp > $some_recent_tstamp cutting out 90+% of
tuples was extremely helpful for performance for both index size
reasons and having to process less tuples. This index needs to be
periodically rebuilt with a newer timestamp constant, and the rebuild
would be a lot faster if it could use the existing index to perform an
index only scan of 10% of data instead of scanning and sorting the
full table.
Ants Aasma
--
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt
Web: http://www.postgresql-support.de
