Thread

Re: [PATCH] Add zstd compression for TOAST using extended header format

Dharin Shah <dharinshah95@gmail.com> — 2025-12-13T19:32:27Z

Hello,

Apologies for the spam, updated the patch with the tests corrected.

Thanks,
Dharin

On Sat, Dec 13, 2025 at 6:31 PM Dharin Shah <dharinshah95@gmail.com> wrote:

> Hello PG Hackers,
>
> Want to submit a patch that implements zstd compression for TOAST data
> using a 20-byte TOAST pointer format, directly addressing the concerns
> raised in prior discussions [1
> <https://www.postgresql.org/message-id/flat/CAFAfj_F4qeRCNCYPk1vgH42fDZpjQWKO%2Bufq3FyoVyUa5AviFA%40mail.gmail.com#e41c78674adfa4d16b2fa82e59faf9aa>
> ][2
> <https://www.postgresql.org/message-id/flat/CAJ7c6TOtAB0z1UrksvGTStNE-herK-43bj22=5xVBg7S4vr5rQ@mail.gmail.com>
> ][3
> <https://www.postgresql.org/message-id/flat/YoMiNmkztrslDbNS@paquier.xyz>
> ].
>
> A bit of a background in the 2022 thread [3
> <https://www.postgresql.org/message-id/flat/YoMiNmkztrslDbNS@paquier.xyz>],
> Robert Haas suggested:
> "we had better reserve the fourth bit pattern for something extensible
> e.g. another byte or several to specify the actual method"
>
> i.e. something like:
> 00 = PGLZ
> 01 = LZ4
> 10 = reserved for future emergencies
> 11 = extended header with additional type byte
>
> Michael also asked whether we should have "something a bit more extensible
> for the design of an extensible varlena header."
>
> This patch implements that idea.
> The format:
>
>   struct varatt_external_extended {
>       int32   va_rawsize;     /* same as legacy */
>       uint32  va_extinfo;     /* cmid=3 signals extended format */
>       uint8   va_flags;       /* feature flags */
>       uint8   va_data[3];     /* va_data[0] = compression method */
>       Oid     va_valueid;     /* same as legacy */
>       Oid     va_toastrelid;  /* same as legacy */
>   };
>
> *A few notes:*
>
> - Zstd only applies to external TOAST, not inline compression. The 2-bit
> limit in va_tcinfo stays as-is for inline data, where pglz/lz4 work fine
> anyway. Zstd's wins show up on larger values.
> - A GUC use_extended_toast_header controls whether pglz/lz4 also use the
> 20-byte format (defaults to off for compatibility, can enable it if you
> want consistency).
> - Legacy 16-byte pointers continue to work - we check the vartag to
> determine which format to read.
>
> The 4 extra bytes per pointer is negligible for typical TOAST data sizes,
> and it gives us room to grow.
>
> Regards,
> Dharin
>