Re: Different compression methods for FPI

Justin Pryzby <pryzby@telsasoft.com>

From: Justin Pryzby <pryzby@telsasoft.com>
To: Andrey Borodin <x4mmm@yandex-team.ru>
Cc: Michael Paquier <michael@paquier.xyz>, pgsql-hackers@postgresql.org, Thomas Munro <thomas.munro@gmail.com>
Date: 2021-03-15T18:09:18Z
Lists: pgsql-hackers

Attachments

On Sun, Mar 14, 2021 at 07:31:35PM -0500, Justin Pryzby wrote:
> On Sat, Mar 13, 2021 at 08:48:33PM +0500, Andrey Borodin wrote:
> > > 13 марта 2021 г., в 06:28, Justin Pryzby <pryzby@telsasoft.com> написал(а):
> > > Updated patch with a minor fix to configure.ac to avoid warnings on OSX.
> > > And 2ndary patches from another thread to allow passing recovery tests.
> > > Renamed to WAL_COMPRESSION_*
> > > Split LZ4 support to a separate patch and support zstd.  These show that
> > > changes needed for a new compression method have been minimized, although not
> > > yet moved to a separate, abstracted compression/decompression function.
> > 
> > Thanks! Awesome work!
> > 
> > > These two patches are a prerequisite for this patch to progress:
> > > * Run 011_crash_recovery.pl with wal_level=minimal
> > > * Make sure published XIDs are persistent
> > > 
> > > I don't know if anyone will consider this patch for v14 - if not, it should be
> > > set to v15 and revisit in a month.  
> > 
> > I want to note, that fixes for 011_crash_recovery.pl are not strictly necessary for this patch set.
> > The problem in tests arises if we turn on wal_compression, absolutely independently from wal compression method.
> > We turn on wal_compression in this test only for CI purposes.
> 
> I rearranged the patches to reflect this.
> Change to zlib and zstd to level=1.
> Add support for negative "zstd fast" levels.
> Use correct length accounting for "hole" in LZ4 and ZSTD.
> Fixed Solution.pm for zstd on windows.
> Switch to zstd by default (for CI).
> Add docs.

Changes:
- Allocate buffer sufficient to accommodate any supported compression method;
- Use existing info flags argument rather than adding another byte for storing
  the compression method; this seems to be what was anticipated by commit
  57aa5b2bb and what Michael objected to.

I think the existing structures are ugly, so maybe this suggests using a GUC
assign hook to support arbitrary compression level, and maybe other options.

-- 
Justin

Commits

  1. Add more error context to RestoreBlockImage() and consume it

  2. Add support for LZ4 with compression of full-page writes in WAL

  3. Extended statistics on expressions

  4. Be clear about whether a recovery pause has taken effect.

  5. Add GUC to enable compression of full page images stored in WAL.