Flowmium

Flowmium is a workflow orchestrator that uses Kubernetes. You can define and run a YAML workflow of containers or you can run a python workflow where each function runs as a Kubernetes pod.

A python workflow would look like this

from flowmium import Flow, FlowContext
from flowmium.serializers import plain_text, json_text, pkl


flow = Flow("testing")


@flow.task(serializer=json_text)
def foo() -> str:
    return "Hallo world"


@flow.task({"input_str": foo}, serializer=plain_text)
def replace_letter_a(input_str: str, flowctx: FlowContext) -> str:
    return input_str.replace("a", "e") + str(flowctx.task_id)


@flow.task({"input_str": foo}, serializer=pkl)
def replace_letter_t(input_str: str) -> str:
    return input_str.replace("t", "d")


@flow.task(
    {"first": replace_letter_t, "second": replace_letter_a}, serializer=plain_text
)
def concat(first: str, second: str) -> str:
    return f"{first} {second}"


if __name__ == "__main__":
    flow.run()

Getting started

• Setting up on local for testing
• Deploying on production
• Example python package workflow
• Example python script workflow
• Example YAML definition workflow
• Python framework documentation
• API documentation
• Rust client docs
• Integrating into an existing Rust project

flowctl CLI

The flowctl CLI is used to monitor current status of workflows, submit new workflows and download artifacts.

Install

cargo install flowmium

Usage

Notes

Secrets are stored in the server and can be referred to set environment variable values in YAML definition or the Python workflows. This is so you don't have to commit secrets to your repository. They don't however use Kubernetes secrets, they are set as normal environment variables when workflow tasks are deployed as a Job.

YAML flow definition schema

Reference for YAML flow definition. See example.

Root

Task

Env

Input

Output

Running from source

Running python flow example from source

These instructions will allow you to run an example python flow (framework/tests/example_flow.py) all from local source without pulling from upstream (including the executor). Use this to validate your changes. Instructions assume you are at the root of the repo.

• Install sqlx CLI

  cargo install sqlx-cli
  

• Run a test kubernetes cluster, minio and container registry in local

  cd flowmium/
  make up
  

• Watch for pods running in the local cluster

  cd flowmium/
  make watch
  

• Run migrations

  cd flowmium/
  sqlx migrate run
  

• Run the flowmium server from root of this repo

  cd flowmium/
  export FLOWMIUM_POSTGRES_URL='postgres://flowmium:flowmium@localhost/flowmium'
  export FLOWMIUM_STORE_URL='http://localhost:9000'
  export FLOWMIUM_TASK_STORE_URL='http://172.16.238.4:9000'
  export FLOWMIUM_BUCKET_NAME='flowmium-test'
  export FLOWMIUM_ACCESS_KEY='minio'
  export FLOWMIUM_SECRET_KEY='password'
  export FLOWMIUM_INIT_CONTAINER_IMAGE='docker.io/shnoo28/flowmium:latest'
  export FLOWMIUM_NAMESPACE=default
  export KUBECONFIG=./kubeconfig.yaml
  cargo run --bin flowmium -- server --port 8080
  

• Watch flow status using flowctl

  cd flowmium/
  cargo build
  watch ./target/debug/flowctl list
  

• Build and push the example python flow (NOTE: You might want to use a different image name if you running the test for the second time or prune docker images on your machine)

  cd framework/
  docker build . -t py-flow-test
  docker tag py-flow-test localhost:5180/py-flow-test:latest
  docker push localhost:5180/py-flow-test:latest
  

• Submit the flow to the executor server

  python3 -m tests --image registry:5000/py-flow-test:latest --cmd 'python3 -m tests' --flowmium-server http://localhost:8080
  

Running e2e tests

• For running e2e tests with init container from upstream

  make test
  

• For running e2e tests with init container from source

  FLOWMIUM_INIT_CONTAINER_IMAGE_FROM_SOURCE=true make test
  

Running unit tests for python framework

Run make test from framework/ path.