Comments on: I wish I could be at PGWest

By: Merlin Moncure

Merlin Moncure — Wed, 10 Nov 2010 16:02:54 +0000

Basically, the idea AIUI is to optimize the case where all tuples in the page are known visible, that is, unchanged under any current running transaction. This will allow the lookup back to the heap to check visibility to be skipped, and the data returned directly out of the index. The higher the ratio of reads to writes on a particular page is, the more this optimization pays. Effectively, this will give you covering indexes at the expense of maintaining the visibility bit (which is more complicated than it sounds).

By: Rob Wultsch

Rob Wultsch — Wed, 20 Oct 2010 01:04:55 +0000

@Baron:
I talked with Bruce Momjian and Josh Berkus at OpenSQL Camp about index only queries. There is evidently a push to have indexes where all tuples are visible be able to perform index only queries. I am sure that has a use case, I just have not run across it.

By: Andy

Andy — Tue, 19 Oct 2010 16:39:39 +0000

@Ryan. Ah thanks for the explanation. I did not know that.

By: Ryan

Ryan — Tue, 19 Oct 2010 16:35:13 +0000

@Andy: The indexes do not contain visibility information. In order for postgres to confirm that the information found in the index is actually supposed to be visible to the client, it must look up the visibility information in the table first. This can be quite expensive.

By: Xaprb

Xaprb — Tue, 19 Oct 2010 16:32:59 +0000

No.