Skip to content

Kowalski: a multi-survey data archive and alert broker for time-domain astronomy

License

Notifications You must be signed in to change notification settings

ZwickyTransientFacility/kowalski

 
 

Repository files navigation

Kowalski: a multi-survey data archive and alert broker for time-domain astronomy

Kowalski is an API-driven multi-survey data archive and alert broker. Its main focus is the Zwicky Transient Facility.

Technical details

A schematic overview of the functional aspects of Kowalski and how they interact is shown below:

data/img/kowalski.jpg

  • A non-relational (NoSQL) database MongoDB powers the data archive, the alert stream sink, and the alert handling service.
  • An API layer provides an interface for the interaction with the backend: it is built using a python asynchronous web framework, aiohttp, and the standard python async event loop serves as a simple, fast, and robust job queue. Multiple instances of the API service are maintained using the Gunicorn WSGI HTTP Server.
  • A programmatic python client is also available to interact with Kowalski's API.
  • Incoming and outgoing traffic can be routed through traefik, which acts as a simple and performant reverse proxy/load balancer.
  • An alert brokering layer listens to Kafka alert streams and uses a dask.distributed cluster for distributed alert packet processing, which includes data preprocessing, execution of machine learning models, catalog cross-matching, and ingestion into MongoDB. It also executes user-defined filters based on the augmented alert data and posts the filtering results to a SkyPortal instance.
  • Kowalski is containerized using Docker software and orchestrated with docker-compose allowing for simple and efficient deployment in the cloud and/or on-premise.

Interacting with a Kowalski instance

Kowalski is an API-first system. The full OpenAPI specs can be found here. Most users will only need the queries section of the specs.

The easiest way to interact with a Kowalski instance is by using a python client penquins.

Spin up your own kowalski

Cloning and Environment configuration

Start off by cloning the repo, then cd into the cloned directory:

git clone https://github.com/dmitryduev/kowalski.git
cd kowalski

Make sure you have a python environment that meets the requirements to run Kowalski:

pip install -r requirements.txt

You can then use the kowalski.py utility to manage Kowalski.

Setting up config files

You need config files in order to run Kowalski. You can start off by copying the default config/secrets over:

cp config.defaults.yaml config.yaml
cp docker-compose.defaults.yaml docker-compose.yaml

config.yaml contains the API and ingester configs, the supevisord config for the API and ingester containers, together with all the secrets, so be careful when committing code / pushing docker images.

However, if you want to run in a production setting, be sure to modify config.yaml and choose strong passwords!

docker-compose.yaml serves as a config file for docker-compose, and can be used for different Kowalski deployment modes. Kowalski comes with several template docker-compose configs (see below for more info).

Building Kowalski

Finally, once you've set the config files, you can build an instance of Kowalski. You can do this with the following command:

./kowalski.py up --build

You have now successfully built a Kowalski instance! Any time you want to rebuild kowalski, you need to re-run this command.

Interacting with a Kowalski build

If you want to just interact with a Kowalski instance that has already been built, you can drop the --build flag:

  • ./kowalski.py up to start up a pre-built Kowalski instance
  • ./koiwalski.py downto shut down a pre-built Kowalski instance

Run tests

You can check that a running Kowalski instance is working by using the Kowalski test suite:

./kowalski.py test

Different Deployment scenarios

Kowalski uses docker-compose under the hood and requires a docker-compose.yaml file. There are several available deployment scenarios:

  • Bare-bones
  • Bare-bones + broker for SkyPortal / Fritz
  • Behind traefik

Bare-bones

Use docker-compose.defaults.yaml as a template for docker-compose.yaml. Note that the environment variables for the mongo service must match admin_* under kowalski.database in config.yaml.

Bare-bones + broker for SkyPortal / Fritz

Use docker-compose.fritz.defaults.yaml as a template for docker-compose.yaml. If you want the alert ingester to post (filtered) alerts to SkyPortal, make sure {"misc": {"broker": true}} in config.yaml.

Behind traefik

Use docker-compose.traefik.defaults.yaml as a template for docker-compose.yaml.

If you have a publicly accessible host allowing connections on port 443 and a DNS record with the domain you want to expose pointing to this host, you can deploy kowalski behind traefik, which will act as the edge router -- it can do many things including load-balancing and getting a TLS certificate from letsencrypt.

In docker-compose.yaml:

  • Replace [email protected] with your email.
  • Replace private.caltech.edu with your domain.

Shut down Kowalski

./kowalski.py down

Docs

OpenAPI specs are to be found under /docs/api once Kowalski is up and running.

Developer guidelines

How to contribute

Contributions to Kowalski are made through GitHub Pull Requests, a set of proposed commits (or patches).

To prepare, you should:

Then, for each feature you wish to contribute, create a pull request:

  1. Download the latest version of Kowalski, and create a new branch for your work.

    Here, let's say we want to contribute some documentation fixes; we'll call our branch rewrite-contributor-guide.

    git checkout master
    git pull upstream master
    git checkout -b rewrite-contributor-guide
  2. Make modifications to Kowalski and commit your changes using git add and git commit. Each commit message should consist of a summary line and a longer description, e.g.:

    Rewrite the contributor guide
    
    While reading through the contributor guide, I noticed several places
    in which instructions were out of order. I therefore reorganized all
    sections to follow logically, and fixed several grammar mistakes along
    the way.
    
  3. When ready, push your branch to GitHub:

    git push origin rewrite-contributor-guide

    Once the branch is uploaded, GitHub should print a URL for turning your branch into a pull request. Open that URL in your browser, write an informative title and description for your pull request, and submit it. There, you can also request a review from a team member and link your PR with an existing issue.

  4. The team will now review your contribution, and suggest changes. To simplify review, please limit pull requests to one logical set of changes. To incorporate changes recommended by the reviewers, commit edits to your branch, and push to the branch again (there is no need to re-create the pull request, it will automatically track modifications to your branch).

  5. Sometimes, while you were working on your feature, the master branch is updated with new commits, potentially resulting in conflicts with your feature branch. To fix this, please merge in the latest upstream/master branch:

    git merge rewrite-contributor-guide upstream/master

Developers may merge master into their branch as many times as they want to.

  1. Once the pull request has been reviewed and approved by at least two team members, it will be merged into Kowalski.

Pre-commit hook

Install our pre-commit hook as follows:

pip install pre-commit
pre-commit install

This will check your changes before each commit to ensure that they conform with our code style standards. We use black to reformat Python code and flake8 to verify that code complies with PEP8.

About

Kowalski: a multi-survey data archive and alert broker for time-domain astronomy

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.0%
  • Dockerfile 1.0%