ayeaye-fossa 0.0.33

Execution engine for Aye-Aye ETL models

fossa

Execution engine for Aye-Aye ETL models

Overview

Fossa runs Aye-aye models and their subtasks across a distributed environment.

Aye-aye models can be run without Fossa (they're just Python code) but when a model grows too big to execute on a single computer the task needs to be spread across multiple computers. A 'distributed environment' is one with multiple compute node that are networked so messages can be passed between nodes.

An instance of Fossa runs on each compute node where it facilitates the communication of messages between nodes.

A node could be a docker or full computer instance.

Getting Started

Setup your virtual environment (venv, pipenv, poetry etc.)

pip install ayeaye-fossa

Make a python module to use this-

import ayeaye

from fossa import run_fossa, BaseConfig

class NothingEtl(ayeaye.Model):
    def build(self):
        pass

class FossaConfig(BaseConfig):
    ACCEPTED_MODEL_CLASSES = [NothingEtl]

run_fossa(FossaConfig)

Run it and point your browser to http://0.0.0.0:2345/. You have a fossa node that will only run the NothingEtl model locally. This is useless as that model doesn't do anything but it's a start.

The examples directory has that example plus a few more.

Running Fossa locally

For local development (so an IDE and debugging tools etc. can be used) the Flask built in server can be used. run_fossa (demonstrated above) uses gunicorn.

Ensure your working directory is the same directory as this README file.

Then install dependencies and run the tests-

cp local_env_example .env
pipenv shell
pipenv install --dev
python -m unittest discover tests

The .env file is used by pipenv.

For all python commands below you will need to be in this pipenv shell.

In a distributed environment one instance of Fossa would run on each compute node. To experiment with Fossa just run one or more instances on a local computer.

Fossa runs a small web-server app which can be used to submit tasks and query the progress and status of tasks. In production, jobs are more likely to be fetched from a message queue.

Copy the example config file into your own person config; sym-link to local_config.py so the run_local_app() function in fossa.app can find it.

e.g.

cd fossa/settings
# replace xxxxx with your name or a more useful identifier for your environment
cp local_config_example.py local_config_xxxxx.py
# have a look in your config file. Is there anything you'd like to change to fit with your system?
ln -s local_config_xxxxx.py local_config.py

In the virtual env (provided by pipenv) from above and with the current working directory being the project's root directory-

python fossa/app.py

You'll now have a locally running web app. It will output IP addresses it is accepting connections from. Typically just point a browser at `http://0.0.0.0:2345/'

## Running Fossa locally using gunicorn

export PYTHONPATH=`pwd`:`pwd`/lib
DEPLOYMENT_ENVIRONMENT=local python lib/fossa/main.py

Posting a task

In a production environment tasks are more likely to arrive through a message queue. But it's also possible to use an HTTP POST to submit a task.

If you used the local_config_example.py file as a starting point for your local config it will have a single model already in the ACCEPTED_MODEL_CLASSES parameter. This is a tiny example ETL model.

Fossa will only run models that have been pre-defined before start-up. The ACCEPTED_MODEL_CLASSES config variable is the simplest way to set this.

POST to your local instance of Fossa a task specification. This example runs the complete SimpleExampleEtl model-

curl --header "Content-Type: application/json" \
     --data '{"model_class":"NothingEtl"}'  \
     --request POST http://0.0.0.0:2345/api/0.01/task

Distributed (but still local) processing

TODO

curl --header "Content-Type: application/json" \
     --data '{"model_class":"PartitionedExampleEtl"}'  \
     --request POST http://0.0.0.0:2345/api/0.01/task

Tests

The normal unittests will run with the python -m unittest discover tests command detailed above. There are also some integration tests in an early stage of development. They are messy and output a lot of rubbish to the console. Enable them by supplying a RabbitMQ broker through an environmental variable. e.g. export RABBITMQ_URL=amqp://guest:guest@localhost.