Horton

Re: Use read streams in pg_visibility

Noah Misch <noah@leadboat.com>

View thread View on postgr.es
From: Noah Misch <noah@leadboat.com>
To: Nazir Bilal Yavuz <byavuz81@gmail.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2024-09-10T22:38:38Z
Lists: pgsql-hackers 
On Tue, Sep 10, 2024 at 02:35:46PM +0300, Nazir Bilal Yavuz wrote:
> Your patch is correct. I wrongly assumed it would catch blockno bug,
> the attached version catches it. I made blockno = 0 invisible and not
> frozen before copying the vm file. So, in the blockno buggy version;
> callback will skip that block but the main loop in the
> collect_corrupt_items() will not skip it. I tested it with your patch
> and there is exactly 1 blockno difference between expected and result
> output.

Pushed.  I added autovacuum=off so auto-analyze of a system catalog can't take
a snapshot that blocks VACUUM updating the vismap.  I doubt that could happen
under default settings, but this lets us disregard the possibility entirely.

I also fixed the mix of tabs and spaces inside test file string literals.

Commits

  1. Optimize pg_visibility with read streams.

  2. Revert "Optimize pg_visibility with read streams."

  3. Fix stack variable scope from previous commit.

  4. Add block_range_read_stream_cb(), to deduplicate code.