why no parallel index build? and diff time in building indexes

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

why no parallel index build? and diff time in building indexes

Bill Ross

I get it that there are situations where a parallel index build wouldn't work, but it seems to be impossible - how can that be right?

    id1 character varying(10),
    id2 character varying(10),

CREATE INDEX pairs_h_1_idx ON pr.pairs_h USING btree (id1);
CREATE INDEX pairs_h_2_idx ON pr.pairs_h USING btree (id2);

->

COPY 213531445
Time: 301621.653 ms (05:01.622)
CREATE INDEX
Time: 331532.765 ms (05:31.533)
CREATE INDEX
Time: 645480.485 ms (10:45.480)

and

COPY 437443831
Time: 351852.578 ms (05:51.853)
CREATE INDEX
Time: 179233.868 ms (02:59.234)
CREATE INDEX
Time: 438404.415 ms (07:18.404)

If not within the db, might it be possible to do it in a tool I could write, then somehow put it all directly in the db?

Within the db, a language addition (e.g. make CREATE take a list of actions) might be needed.

Question 2 is why the second index takes much longer than the 1st?

Finally, shout out to anyone from the early 90's.

Thanks,

Bill Ross

--
Phobrain.com