Re: Page at a time index scan

On Wed, 3 May 2006, Heikki Linnakangas wrote:

> On Tue, 2 May 2006, Tom Lane wrote:
>
>> Heikki Linnakangas <hlinnaka@iki.fi> writes:
>>> On Tue, 2 May 2006, Tom Lane wrote:
>>>> Backwards scan may break this whole concept; are you sure you've thought
>>>> it through?
>>
>>> I think so. The patch doesn't change the walk-left code. Do you have
>>> something specific in mind?
>>
>> I'm worried about synchronization, particularly what happens if the page
>> gets deleted from under you while you don't have it pinned.
>
> AFAICS, shouldn't happen. The half-dead system makes sure that a page won't
> get deleted while a scan might still be interested in it. It doesn't depend
> on pins.

As Tom pointed out elsewhere in this thread, the above explanation is
wrong because half-dead state only applies to upper-level pages. I thought
that half-dead means that the page is dead and removed from the tree, but
not yet recycled because some transaction might still be interested in it.
Now I see that that state is actually called "deleted".

The point remains, however. A page won't get deleted while a scan
might still be interested in it, because deleted pages are not
immediately recycled (except on vacuum full), and the left and right
sibling pointers stay intact until no transaction can be interested in it.

- Heikki