Re: [WIP] [B-Tree] Keep indexes sorted by heap physical location

On Thu, Aug 18, 2016 at 6:27 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Claudio Freire <klaussfreire(at)gmail(dot)com> writes:
>> On Thu, Aug 18, 2016 at 6:04 PM, Kevin Grittner <kgrittn(at)gmail(dot)com> wrote:
>>> Speaking of performance side effects, does this avoid O(N^2)
>>> performance on index tuple insertion with duplicate values, for all
>>> insertion orderings? For example, does it descend directly to the
>>> right leaf page for the insert rather than starting at the front of
>>> the block of duplicate values and scanning to the right for a
>>> block with space, with a random chance to split a full block on
>>> each page it moves through?
>
>> Yes, but only on non-unique indexes.
>
> How's that work if the existing entries aren't in TID order (which they
> will not be, in a pre-existing index)? Or are you assuming you can blow
> off on-disk compatibility of indexes?
>
> regards, tom lane

This patch does blow off on-disk compatibility, but the plan is to
re-add it later on.

A flag in the meta page would indicate whether it's a sorted index or
not, and the only way to get a sorted index would be with a reindex.
The code would have to support both for a while until whatever point
we'd figure we can drop support for old format indexes.

Since this is a sort order change I see no alternative, either the
whole index is sorted by TID or not.