Vectorsearch Docker Image

About

Initially, I targeted the paradedb/paradedb:0.12.0-pg17 image for my task, but it became clear that we only required its pg_search and pgvector extensions, with the rest being unnecessary for our use case.

Additionally, an older release of paradedb included pgvectorscale from Timescale for its advanced algorithms. However, this was removed in the latest release. I decided to explore using it, which brings us to the current setup.

Build

I started with the Postgres 17 base image featuring pgvector (by time of writing, the latest version is pgvector/pgvector:0.8.0-pg17 ). Subsequently, I added pg_search and pgvectorscale as dependencies.

The bootstrap.sh script includes SQL commands to create these extensions for every potential database, ensuring a comprehensive setup.

Usage

1. I have published a docker image for vectorsearch at dockerhub

2. To build from the Dockerfile: make build. For more details regarding the build process, please refer to the makefile details.

To run the container:

docker run --name vectorsearch -p <5432>:5432 -v <$pwd/pgdata>:/var/lib/postgresql/data/ -e POSTGRES_USER=myuser -e POSTGRES_PASSWORD=mypassword -e POSTGRES_DB=mydatabase oaklight/vectorsearch

or use docker compose :

docker compose up -d

Docker Compose Sample Configuration

For easier management of the container, a sample docker-compose.yaml is provided as compose.yaml.sample . You can use this file as a starting point for configuring and running the service with Docker Compose.

Here is the content of compose.yaml.sample :

services:
  paradedb:
    image: oaklight/vectorsearch:latest-pg17
    container_name: vectorsearch
    environment:
      POSTGRES_USER: myuser
      POSTGRES_PASSWORD: mypassword
      POSTGRES_DB: mydatabase
    ports:
      - 5432:5432
    volumes:
      - $pwd/pgdata:/var/lib/postgresql/data/

How to Use the Sample Configuration

1. Copy compose.yaml.sample to compose.yaml:

   cp compose.yaml.sample compose.yaml

2. Modify compose.yaml as needed to match your environment and configuration preferences.

3. Run the following command to start the service using Docker Compose:

   docker compose up -d

This will start the vectorsearch container with the configurations specified in compose.yaml .

Note

The image is based on the official postgres image, so you can use the same commands to interact with the database.

For example, you could use psql to connect to the database:

psql -h <your_container_name> -U <username> -d <database>

You can also use pgAdmin to connect to the database.