Re: GIN improvements part2: fast scan

On 9.2.2014 22:51, Erik Rijkers wrote:
> On Sun, February 9, 2014 22:35, Tomas Vondra wrote:
>> On 3.2.2014 07:53, Oleg Bartunov wrote:
>>> PS. I used data delicious-rss-1250k.gz from
>>> http://randomwalker.info/data/delicious/
>>
>> I'm working on extending the GIN testing to include this test (and I'll
>> use it to test both for GIN and hstore-v2 patches).
>>
>> I do have the dataset, but I need the queries too - how did you generate
>> the queries for your benchmark? Do you have some query generator at hand?
>>
>> In your Dublin talk I see just this query type
>>
>> select count(*) from hs where h @> 'tags=>{{term=>NYC}}';
>>
>> but that seems inadequate for representative benchmark. Are there other
>> types of queries that need to be tested / might be interesting? E.g.
>> queries with multiple search terms etc.?
>>
>
> There is the hstore operators table (Table F-6) that you can perhaps use to generate queries? (I am working on this too.)
> At least it contains already a handful of queries.
>
> To get at the bits in that table, perhaps the perl program attached here helps:
>
> http://www.postgresql.org/message-id/f29d70631e2046655c40dfcfce6db3c3.squirrel@webmail.xs4all.nl
>
> YMMV, I guess...

It seems to me the purpose of the program is to demonstrate some
differences between expected and actual result of an operator. If that's
true then it's not really of much use - it's not a query generator. I
need a script that generates large number of 'reasonable' queries, i.e.
queries that make sense and resemble what the users actually do.

Tomas