public inbox for [email protected]  

help / color / mirror / Atom feed

From: Jonathan S. Katz <[email protected]>
To: PostgreSQL Hackers <[email protected]>
Cc: Nathan Bossart <[email protected]>
Subject: Large expressions in indexes can't be stored (non-TOASTable)
Date: Tue, 3 Sep 2024 12:35:42 -0400
Message-ID: <[email protected]> (raw)

Hi,

I ran into an issue (previously discussed[1]; quoting Andres out of 
context that not addressing it then would "[a]ll but guarantee that 
we'll have this discussion again"[2]) when trying to build a very large 
expression index that did not fit within the page boundary. The 
real-world use case was related to a vector search technique where I 
wanted to use binary quantization based on the relationship between a 
constant vector (the average at a point-in-time across the entire data 
set) and the target vector[3][4]. An example:

CREATE INDEX ON embeddings
   USING hnsw((quantization_func(embedding, $VECTOR)) bit_hamming_ops);

However, I ran into the issue in[1], where pg_index was identified as 
catalog that is missing a toast table, even though `indexprs` is marked 
for extended storage. Providing a very simple reproducer in psql below:

----
CREATE TABLE def (id int);
SELECT array_agg(n) b FROM generate_series(1,10_000) n \gset
CREATE OR REPLACE FUNCTION vec_quantizer (a int, b int[]) RETURNS bool 
AS $$ SELECT true $$ LANGUAGE SQL IMMUTABLE;
CREATE INDEX ON def (vec_quantizer(id, :'b'));

ERROR:  row is too big: size 29448, maximum size 8160
---

This can come up with vector searches as vectors can be quite large - 
the case I was testing involved a 1536-dim floating point vector (~6KB), 
and the node parse tree pushed past the page boundary by about 2KB.

One could argue that pgvector or an extension can build in capabilities 
to handle quantization internally without requiring the user to provide 
a source vector (pgvectorscale does this). However, this also limits 
flexibility to users, as they may want to bring their own quantization 
functions to vector searches, e.g., as different quantization techniques 
emerge, or if a particular technique is more suitable for a person's 
dataset.

Thanks,

Jonathan

[1] 
https://www.postgresql.org/message-id/flat/84ddff04-f122-784b-b6c5-3536804495f8%40joeconway.com
[2] 
https://www.postgresql.org/message-id/20180720000356.5zkhvfpsqswngyob%40alap3.anarazel.de
[3] https://github.com/pgvector/pgvector
[4] https://jkatz05.com/post/postgres/pgvector-scalar-binary-quantization/


Attachments:

  [application/pgp-signature] OpenPGP_signature.asc (840B, ../[email protected]/2-OpenPGP_signature.asc)
  download

view thread (2+ messages)

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected], [email protected], [email protected]
  Subject: Re: Large expressions in indexes can't be stored (non-TOASTable)
  In-Reply-To: <[email protected]>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox