A Docker image providing PostgreSQL 18 with vector extensions for AI/ML workloads.

• PostgreSQL 18 - Latest stable PostgreSQL version
• pgvector - Vector similarity search (from pgvector/pgvector:0.8.1-pg18-trixie)
• pgvectorscale - Time-series vector compression (built from source)
• Semantic versioning - Managed via Poetry
• Local automation - Via pypyr workflows
• Multi-arch builds - amd64 + arm64 via GitHub Actions

docker.io/imprvai/imprv_pgvectorscale:18-v0.1.0
docker.io/imprvai/imprv_pgvectorscale:18-latest

docker run --rm -it \
  -e POSTGRES_PASSWORD=secret \
  -p 5432:5432 \
  docker.io/imprvai/imprv_pgvectorscale:18-latest

On first initialization with a fresh data volume:

• ✅ pgvector is pre-installed (from base image)
• ✅ vectorscale is installed via init_vectorscale.sql
• ✅ vectorscale is preloaded via shared_preload_libraries

Verify extensions are available:

SELECT * FROM pg_available_extensions
WHERE name IN ('vector', 'vectorscale');

SELECT * FROM pg_extension
WHERE extname IN ('vector', 'vectorscale');

Version source of truth: pyproject.toml

[tool.poetry]
version = "X.Y.Z"

Bump version locally:

pypyr release part=patch  # or part=minor / part=major

This automated workflow will:

1. Bump version in pyproject.toml using Poetry
2. Build Docker images with tags:
   • 18-vX.Y.Z
   • 18-latest
3. Push images to Docker Hub
4. Update CHANGELOG.md with commits since previous tag
5. Commit pyproject.toml and CHANGELOG.md
6. Create git tag vX.Y.Z
7. Push commit and tag (if git credentials configured)

Commands can be run from the project root using pypyr shortcuts:

# Single-arch build (default, faster, for local testing)
pypyr build

# Multi-arch build (amd64 + arm64, requires docker buildx)
# Note: Multi-arch builds push images to registry (can't load locally)
pypyr build multiarch=true push=true

pypyr push

# Single-arch release (default, builds locally, then pushes)
pypyr release part=patch

# Multi-arch release (amd64 + arm64, builds and pushes to registry)
pypyr release part=patch multiarch=true

Includes: version bump + build + push + changelog + git tag/push

Note: When using multiarch=true, images are automatically pushed during build (required for multi-arch manifests). Single-arch builds are loaded locally first, then pushed separately.

Note:

• Shortcuts are defined in pyproject.toml under [tool.pypyr.shortcuts]
• Multi-arch builds require docker buildx and are slower (uses QEMU emulation)
• Multi-arch builds push images to registry (can't be loaded locally like single-arch)
• GitHub Actions always builds multi-arch by default

Workflow: .github/workflows/pr-build-test.yaml

• Automatic: Runs on Pull Requests (opened, updated, reopened)
• Paths: Only triggers on changes to docker/, .github/workflows/, pyproject.toml, or ops/

1. Build Docker image (without pushing)
2. Test PostgreSQL startup - Verifies container starts successfully
3. Test extensions - Verifies vector and vectorscale extensions are available and can be installed
4. Verify preload - Checks that vectorscale is properly configured

Note: Images are built with tag pr-{PR_NUMBER} for testing purposes only. No images are pushed to registry.

Workflow: .github/workflows/release.yaml

• Manual trigger: workflow_dispatch from GitHub UI
• Input: part = major | minor | patch

1. Prepare release - Bump version, update changelog, commit, create tag (using pypyr release-prep)
2. Build & push multi-arch images (amd64 + arm64) using docker/build-push-action
3. Push commit & tag to repository

Tags pushed:

• 18-vX.Y.Z
• 18-latest

Note:

• CI/CD always builds and pushes multi-arch images (amd64 + arm64)
• Local release workflow (pypyr release) builds single-arch images by default (faster for testing)
• Use multiarch=true flag for local multi-arch builds: pypyr release part=patch multiarch=true

This project is licensed under the MIT License.