Giters

pgvector / pgvector

Open-source vector similarity search for Postgres

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Hosted Providers

ankane opened this issue · comments

commented

Add hosted providers that support pgvector to this thread.

Initial list:

commented

@ankane Funny, I was planning to send a PR shortly to request adding Supabase to the list. Been working with their team to get that integrated:
supabase/postgres#472

Thanks again for making a great extension. Currently using this to store and query OpenAI Embeddings.

commented

Awesome, big thanks for adding it!

commented

Thanks @m-abdelwahab, great to hear!

commented

Need that Google CloudSQL 8)

commented

Pigsty - Free RDS PG Alternative

now add pgvector as default extension in v2.0.2

commented

We added the extension to spilo to use with postgres-operator in kubernetes.

https://github.com/stablecog/spilo

Can just use the docker image in the helm chart values, or build your own:

https://hub.docker.com/r/stablecog/spilo

15.3+ has pgvector for Postgres 11-15

commented

fingers crossed for RDS 🤞

commented

anyone has setup pg_vector on Railway? Didn't found anything about installing extensions there :/

commented

anyone has setup pg_vector on Railway? Didn't found anything about installing extensions there :/

@NicolasLopes7 It's not possible yet. You can vote and comment on these feature requests:

commented

No digital ocean hosted PG yet. Heres an upvote support request for others that want it: https://ideas.digitalocean.com/app-framework-services/p/saving-openai-embeddings-with-pgvector

commented

Timescale supports pgVector in its cloud offering.

image

commented

Hi!

pgvector is available on Stackhero PostgreSQL cloud hosting solution with PostgreSQL 15 🥳🥳🥳

commented

And it is also available for Heroku via Stackhero PostgreSQL add-on 👌

commented

Also available on hosted PostgresML databases.

commented

pgvector is now available on Azure:

  • Azure Database for PostgreSQL Flexible Server

Follow this page to enable vector in the allowed extensions and then simply CREATE EXTENSION vector; on your Flexible Server.

  • Azure Cosmos DB for PostgreSQL

Just connect to your cluster and issue SELECT create_extension('vector');

Main readme update submitted as #129

commented

@mulander Seems like extension vector is not yet available in Azure Flexible Server (also not present in this doc). Do you know when it will be in general availability?

commented

@mulander Seems like extension vector is not yet available in Azure Flexible Server (also not present in this doc). Do you know when it will be in general availability?

@jcoc611-gvt What region are you checking and is it on a new or deployed cluster? The documentation update (including the doc you linked) is scheduled to be published on the 23rd of May.

commented

Please upvote the issue to request this extension on GCP Cloud SQL https://issuetracker.google.com/issues/278521774

commented

@mulander Seems like extension vector is not yet available in Azure Flexible Server (also not present in this doc). Do you know when it will be in general availability?

@jcoc611-gvt What region are you checking and is it on a new or deployed cluster? The documentation update (including the doc you linked) is scheduled to be published on the 23rd of May.

@mulander Seems like new clusters can enable vector whereas existing ones cannot.

In any case, it seems like pgvector is crashing on INSERT or UPDATE large vectors (e.g. 1536 dimensions). Logs look something like this:

LOG: server process (PID 3517) was terminated by signal 4: Illegal instruction

Even on a simple table such as

CREATE TABLE test (id bigserial PRIMARY KEY,  v vector(1536));

This works fine for trivially small vectors e.g. 1-2 dimensions.

Perhaps pgvector was compiled with optimizations not supported in Azure flexible database environment? Or is it a matter of configuring server parameters/using a SKU with more RAM?

Edit: seems like one needs to update track_activity_query_size on Azure to fit the embedding array for it to work.

commented

@jcoc611-gvt I think it'd be good to move this to an Azure support channel. Illegal instruction typically means pgvector was compiled with AVX instructions that aren't available on the machine (and will only appear with vectors of 8+ or 16+ dimensions).

commented

Hi @jcoc611-gvt and @ankane,

I am from the Azure Cosmos DB for PostgreSQL team, and I am also in touch with @mulander and Azure Database for PostgreSQL Flexible Server devs in our internal thread, investigating the issue.

We also believe the offending line is OPTFLAGS = -march=native, and we will try overriding it to avoid any potential issues in the future, especially when/if the extension is built on one machine/architecture and distributed to other machines (this happens when the span of supported VM sizes is large).

For us to investigate the issue and help you better and faster, @jcoc611-gvt, could you please share with us which VM size you were using when you tested the below excerpt:

CREATE TABLE test (id bigserial PRIMARY KEY,  v vector(1536));
commented

Hi @aytekinar, please create a new issue for this.

commented

It looks like RDS support does not include Aurora: Aurora extension support

While I couldn't find a documented way to make a request, this email address appears to work: rds-aurora-extensions-request@amazon.com

commented

https://github.com/different-ai/embedbase hosted version uses pgvector in supabase :)

commented

It seems to me that Google has recently secretly added support for pgvector 0.4.2 in POSTGRES_15_2.R20230530.01_00 maintenance update! The list of extensions in the documentation has not been updated yet to reflect this but running SELECT * FROM pg_available_extensions; on a recently updated Cloud SQL instance shows this:

CleanShot 2023-06-19 at 16 56 34

Yas!

commented

We've added pgvector to our supported extensions at Aiven, starting from PG13, see https://docs.aiven.io/docs/products/postgresql/reference/list-of-extensions. You may need to apply a maintenance upgrade to get it in your existing service.

commented

anyone have an update on Aurora pgvector support on RDS?

commented

It seems to me that Google has recently secretly added support for pgvector 0.4.2 in POSTGRES_15_2.R20230530.01_00 maintenance update! The list of extensions in the documentation has not been updated yet to reflect this but running SELECT * FROM pg_available_extensions; on a recently updated Cloud SQL instance shows this:

CleanShot 2023-06-19 at 16 56 34

Yas!

It’s now in the official documentation !!!

commented

image
I faced this issue when i was installing pgvector. i am using windows 10 . can anyone help me please

commented

+1 pgvector for AWS Aurora

commented

So I guess this announcement from AWS on May 3 about pgvector support doesn't apply to Aurora? https://aws.amazon.com/about-aws/whats-new/2023/05/amazon-rds-postgresql-pgvector-ml-model-integration/

commented

Correct, pgvector works on Amazon RDS for PostgresSQL but not on Amazon Aurora PostgresSQL.

commented

Although it is not documented, this seems to work on DigitalOcean Managed Postgres as well now. At least with Postgres 15. It didn't work with Postgres 12 though.

commented

DigitalOcean now lists pgvector on their page of supported extensions (for PostgreSQL 13 and newer): https://docs.digitalocean.com/products/databases/postgresql/details/supported-extensions/

commented

The pgvector extension is available on Aurora PostgreSQL as of July 13 (https://aws.amazon.com/about-aws/whats-new/2023/07/amazon-aurora-postgresql-pgvector-vector-storage-similarity-search) on versions 15.3, 14.8, 13.11, 12.15 and higher. Here's a blog that shows you how to build an AI powered chat bot application and perform sentiment analysis using pgvector and Amazon Aurora PostgreSQL - https://aws.amazon.com/blogs/database/leverage-pgvector-and-amazon-aurora-postgresql-for-natural-language-processing-chatbots-and-sentiment-analysis/

commented

(off topic) @guillim is the GUI you show something from google cloud or a standalone product?

commented

Looks like Azure Postgres Flex Server only supports pgvector .4,0

commented

EDB's BigAnimal Postgres-as-a-Service supports pgvector:

BigAnimal is EDB’s managed Postgres service that allows you to run Postgres on all major cloud platforms. Now, you can enable pgvector on your databases and start experimenting!

commented

pgvector is supported on Railway via the community made pgvector project template, which has been updated to the latest version today!

(Thanks @brody192!)

edit: incorrectly called the template an "official" template

commented

thanks for the shout-out, but I feel it necessary to point out that it's not official, it's not published by Railway or pgvector themselves, I'm just a Railway community member :)

commented

The Heroku Postgres add-on seems to support it for tiers starting from standard.

commented

https://nexai.site now supports pgvector as a projects semantic database.

nexai-dashboard-settings