Re: gist vacuum seaq access

From: Heikki Linnakangas <hlinnakangas(at)vmware(dot)com>
To: Костя Кузнецов <chapaev28(at)ya(dot)ru>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: gist vacuum seaq access
Date: 2014-09-12 21:38:34
Message-ID: 541367DA.1000104@vmware.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 09/11/2014 12:16 PM, Костя Кузнецов wrote:
> After discussion of gist seaq access in vaccum there are 2 issue:
> Heikki says :
> Vacuum needs to memorize the current NSN when it begins
> 1) how i may getting corect NSN.

The "current NSN" is just current WAL insert location, so
GetXLogInsertRecPtr() will do it. Or if it's an unlogged table,
GistGetFakeNSN().

> Also i must setting F_DELETED flag on empty page and to clean parent from link
> on deleted_pages
> 2) how i may getting parent of page fast??

No idea. The B-tree code uses the page's high key to search the tree,
but that doesn't work too well for GiST. There is no high key on the
page to search with, although you could use e.g. the last key you remove
from the page for that. But a search from the root with that key might
need to visit many pages before finding the deleted page, so it's not
very fast.

I think you need to take a different approach. For example, leave the
empty page in the tree in vacuum, but remember its page number. After
vacuum has finished, make another full scan of the index, searching for
the downlinks of all the empty pages.

- Heikki

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tomas Vondra 2014-09-12 22:07:39 Re: bad estimation together with large work_mem generates terrible slow hash joins
Previous Message Robert Haas 2014-09-12 21:22:58 Re: bad estimation together with large work_mem generates terrible slow hash joins