lkshminarayanan / omnigres

Postgres as a Platform

Documentation | Bounties

---

Omnigres makes Postgres a developer-first application platform. You can deploy a single database instance and it can host your entire application, scaling as needed.

• Running application logic inside or next to the database instance
• Deployment provisioning (Git, containers, etc.)
• Database instance serves HTTP, WebSocket and other protocols
• In-memory and volatile on-disk caching
• Routine application building blocks (authentication, authorization, payments, etc.)
• Database-modeled application logic via reactive queries
• Automagic remote APIs and form handling
• Live data updates

Blogs and Publications

• Omnigres maintainer's blog

🏃 Quick start

The fastest way to try Omnigres out is by using its container image:

docker volume create omnigres
docker run --name omnigres --mount source=omnigres,target=/var/lib/postgresql/data \
           -p 127.0.0.1:5432:5432 -p 127.0.0.1:8080:8080 --rm ghcr.io/omnigres/omnigres:latest
# Now you can connect to it:
psql -h localhost -p 5432 -U omnigres omnigres # password is `omnigres`

Postgres parameters such as database, user or password can be overridden as per the "Environment Variables" section in postgres image instructions

You can access the HTTP server at localhost:8080

Building your own image

If you can't use the pre-built image (for example, you are running a fork or made changes), you can build the image yourself:

# Build the image
DOCKER_BUILDKIT=1 docker build . -t ghcr.io/omnigres/omnigres

👋 "Hello, world"

Here we expect you are running the container image, which has omni_httpd and omni_web extensions provisioned by default.

Let's start with a traditional example. Here we will instruct the handler that is provisioned by omni_httpd by default to use the enclosed query to greet the world.

Below, we'll show examples in Python and plain SQL (or PL/pgSQL). Support for more languages is coming!

$ curl localhost:8080
Hello, world!

Python (Flask) implementation

from omni_python import pg
from flask import Flask
from omni_http.omni_httpd import flask

app = Flask('myapp')


@app.route('/')
def hello():
    return "Hello, world!"


handle = pg(flask.Adapter(app))

To connect the endpoint:

update omni_httpd.handlers
set
    query =
        $$select handle(request.*) from request$$;

NB: Please note that you will need to follow Python setup steps for the time being before our CLI tooling is ready.

Plain SQL

You can also achieve the same using plain SQL with very little setup.

update omni_httpd.handlers
set
    query =
        $$select omni_httpd.http_response('Hello, world!') from request$$;

Now, let's make it more personal and let it greet the requester by name.

$ curl "localhost:8080?name=John"
Hello, John!

Python (Flask) implementation

from flask import request  # we need to access `request`


@app.route('/')
def hello():
    return f"Hello, {request.args.get('name', 'world')}!"

Plain SQL

update omni_httpd.handlers
set
    query =
        $$select omni_httpd.http_response('Hello, ' || 
                   coalesce(omni_web.param_get(request.query_string, 'name'), 'world') || '!')
          from request$$;

This, of course, only barely scratches the surface, but it may give you a very high-level concept of how Omnigres web services can be built.

For a more complex example, that uses the underlying database and employs more real-world layout, check out this MOTD service example.

🏗️ Component Roadmap

Below is the current list of components being worked on, experimented with and discussed. This list will change (and grow) over time.

Name	Status	Description
omni_schema	✅ First release candidate	Application schema management
omni_json	✅ First release candidate	JSON toolkit
omni_xml	✅ First release candidate	XML toolkit
omni_http	✅ First release candidate	Common HTTP types library
omni_httpd and omni_web	✅ First release candidate	Serving HTTP in Postgres and building services in SQL
omni_mimetypes	✅ First release candidate	MIME types and file extensions
omni_httpc	✅ First release candidate	HTTP client
omni_sql	🚧 Extremely limited API surface	Programmatic SQL manipulation
omni_vfs	☑️ Initial prototype	Virtual File System interface
omni_containers	☑️ Initial prototype	Managing containers
omni_ext and Dynpgext interface	☑️ Getting ready to become first release candidate	Advanced Postgres extension loader
omni_types	✅ First release candidate	Advanced Postgres typing techniques (sum types, etc.)
omni_seq	✅ First release candidate	Extended Postgres sequence tooling
omni_txn	✅ First release candidate	Transaction management
omni_python	☑️ Initial prototype	First-class Python Development Experience
omni_git	🥼 Early experiments (unpublished)	Postgres Git client
omni_reactive	🗓️ Haven't started yet	Reactive queries

⌨️ Hacking

Building & using extensions

To build and run Omnigres, you would need:

• a recent C compiler
• OpenSSL 3.1+
• cmake >= 3.25.1
• (optionally, to use omni_containers or run a full set of tests) a recent version of Docker

cmake -S . -B build
cmake --build build --parallel
make psql_<COMPONENT_NAME> # for example, `psql_omni_containers`

Troubleshooting

cmake not picking up Python version you want?

To use a specific Python build use the cmake flag Python3_EXECUTABLE:

cmake -S . -B build -DPython3_EXECUTABLE=/path/to/python

Build fails for whatever other reason?

Remove build and .pg directories for a clean rebuild:

rm -rf .pg build

Running tests

# in the build directory
CTEST_PARALLEL_LEVEL=$(nproc) make -j $(nproc) all test

Devenv.sh-based local development environment

Initial setup

Follow these guides:

1. https://devenv.sh/getting-started/
2. https://devenv.sh/automatic-shell-activation/
3. Run direnv allow in omnigres repo

Day-to-day development

1. cd into the repo. This brings in all dependencies.
2. To bring up development stack (Postgres with all extensions, etc.), run: devenv up

Once the development environment is running, you can connect to it by issuing:

• pg -> this connects to Postgres through a UNIX socket, for maximum performance. CLI args forwarded.
• pgclear -> removes the PGDATA folder contents. You want to restart devenv up after this so Postgres can reinitialize as per devenv.nix.