Re: Question about RUM-index

On Wed, Jun 15, 2016 at 6:56 AM, Andreas Joseph Krogh <andreas@visena.com> wrote:

Hi.

 

First; Is this the correct forum to ask questions about the Postgres Pro's new RUM-index?

 

​-general is fairly broad in scope and the authors do monitor here as far as I am aware.  I'm not sure this is strictly related to that extension though, anyway.​

 

 

If not, please point me to the right forum.

 

I'm trying to do this (done with GIN):

 

create extension if not exists btree_gin;
drop table if EXISTS delivery;

create table delivery(    id BIGSERIAL primary key,    fts_all TSVECTOR not null,    folder_id BIGINT NOT NULL,    sent TIMESTAMP not null,    message varchar not null
);


create index gin_idx on delivery using GIN(fts_all, folder_id);

​In general I'm not sure what you are so focused on multi-column indexes through this email.​

[snip]

 

​What index?  You don't have "sent" in the one CREATE INDEX statement you specified above.​

 

 

So I wonder if RUM can do any better?

What I don't understand is how to have "folder_id" as part of the RUM-index so that I can search in an array of folders using the index, AND have the whole result sorted by "sent"-timestamp also using the RUM-index.

 

 

​A bit out of my experience here...but see comment on multi-column indexes above.​

 

In the (limited) documentation sorting using timestamp is done like this:

 

 

 

 

ORDER BY sent <-> '2000-01-01'::TIMESTAMP

 

which I don't understand; Why must one specify a value here, and how does that value affect the result?

 

 

​You don't (I think...); <ORDER BY sent> is perfectly valid, it will use the "<=, and >=" operators in the btree operator family to perform the ordering...see comment regarding multi-column indexes.  Even if you stick sent into such an index it would not (I guess it could...but your where clause doesn't filter on it so doing so seem counter-productive) be a leading column so it is unlikely that the index would be helpful in supplying tuples to the sort node in order - the sort would still have work to perform.

 

Maybe its simple ignorance but AFAIK the sorting node never consults an index to perform its work.  Where indexes come into to play is the node under the sort is able to supply its tuples in sorted order then the sort node will have nothing to do.

 

https://www.postgresql.org/docs/9.6/static/indexes-ordering.html

 

"""

In addition to simply finding the rows to be returned by a query, an index may be able to deliver them in a specific sorted order. This allows a query's ORDER BY specification to be honored without a separate sorting step. Of the index types currently supported by PostgreSQL, only B-tree can produce sorted output — the other index types return matching rows in an unspecified, implementation-dependent order.

​"""​

​HTH

 

David J.​

--

Andreas Joseph Krogh

CTO / Partner - Visena AS

Mobile: +47 909 56 963

andreas@visena.com

www.visena.com