Workflow Orchestration

The Lightweight Workflow Orchestration with fewer dependencies the was created for easy to make a simple metadata driven data workflow. It can use for data operator by a .yaml template.

Warning

This package provide only orchestration workload. That mean you should not use the workflow stage to process any large volume data which use a lot of compute resource 😰.

---

📌 Rules of This Workflow engine:

1. The Minimum frequency unit of built-in scheduling is 1 Minute 🕘
2. Can not re-run only failed stage and its pending downstream ↩️
3. All parallel tasks inside workflow core engine use Multi-Threading pool (Python 3.13 unlock GIL 🐍🔓)

---

📝 Workflow Diagrams:

This diagram show where is this application run on the production infrastructure. You will see that this application do only running code with stress-less which mean you should to set the data layer separate this core program before run this application.

flowchart LR
    A((fa:fa-user User))

    subgraph Docker Container
        direction TB
        G@{ shape: rounded, label: "📡Observe<br>Application" }
    end

    subgraph Docker Container
        direction TB
        B@{ shape: rounded, label: "🏃Workflow<br>Application" }
    end

    A <-->|action &<br>response| B
    B -...-> |response| G
    G -...-> |request| B

    subgraph Data Context
        D@{ shape: processes, label: "Logs" }
        E@{ shape: lin-cyl, label: "Audit<br>Logs" }
    end

    subgraph Config Context
        F@{ shape: tag-rect, label: "YAML<br>files" }
    end

    A ---> |push| H(Repo)
    H -.-> |pull| F

    B <-->|disable &<br>read| F

    B <-->|read &<br>write| E

    B -->|write| D

    D -.->|read| G
    E -.->|read| G

Loading

Warning

Disclaimer: I inspire the dynamic YAML statement from the GitHub Action, and my experience of data framework configs pattern. 😬

Other workflow orchestration services that I interest and pick them to be this project inspiration:

• Google Workflows
• AWS Step Functions

📦 Installation

This project need ddeutil and ddeutil-io extension namespace packages to be the base deps. If you want to install this package with application add-ons, you should add app in installation;

Use-case	Install Optional	Support
Python	ddeutil-workflow	✔️
FastAPI Server	ddeutil-workflow[api]	✔️

🎯 Usage

This is examples that use workflow file for running common Data Engineering use-case.

Important

I recommend you to use the call stage for all actions that you want to do with workflow activity that you want to orchestrate. Because it is able to dynamic an input argument with the same call function that make you use less time to maintenance your data workflows.

run-py-local:

   # Validate model that use to parsing exists for template file
   type: Workflow
   on:
      # If workflow deploy to schedule, it will run every 5 minutes
      # with Asia/Bangkok timezone.
      - cronjob: '*/5 * * * *'
        timezone: "Asia/Bangkok"
   params:
      # Incoming execution parameters will validate with this type. It allows
      # to set default value or templating.
      source-extract: str
      run-date: datetime
   jobs:
      getting-api-data:
         runs-on:
            type: local
         stages:
            - name: "Retrieve API Data"
              id: retrieve-api
              uses: tasks/get-api-with-oauth-to-s3@requests
              with:
                 # Arguments of source data that want to retrieve.
                 method: post
                 url: https://finances/open-data/currency-pairs/
                 body:
                    resource: ${{ params.source-extract }}

                    # You can use filtering like Jinja template but this
                    # package does not use it.
                    filter: ${{ params.run-date | fmt(fmt='%Y%m%d') }}
                 auth:
                    type: bearer
                    keys: ${API_ACCESS_REFRESH_TOKEN}

                 # Arguments of target data that want to land.
                 writing_mode: flatten
                 aws:
                    path: my-data/open-data/${{ params.source-extract }}

                    # This Authentication code should implement with your custom call
                    # function. The template allow you to use environment variable.
                    access_client_id: ${AWS_ACCESS_CLIENT_ID}
                    access_client_secret: ${AWS_ACCESS_CLIENT_SECRET}

Before execute this workflow, you should implement caller function first.

registry-caller/
  ╰─ tasks.py

This function will store as module that will import from WORKFLOW_CORE_REGISTRY_CALLER value (This config can override by extra parameters with registry_caller key).

from ddeutil.workflow import Result, tag
from ddeutil.workflow.exceptions import StageException
from pydantic import BaseModel, SecretStr

class AwsCredential(BaseModel):
    path: str
    access_client_id: str
    access_client_secret: SecretStr

class RestAuth(BaseModel):
    type: str
    keys: SecretStr

@tag("requests", alias="get-api-with-oauth-to-s3")
def get_api_with_oauth_to_s3(
    method: str,
    url: str,
    body: dict[str, str],
    auth: RestAuth,
    writing_node: str,
    aws: AwsCredential,
    result: Result,
) -> dict[str, int]:
    result.trace.info("[CALLER]: Start get data via RestAPI to S3.")
    result.trace.info(f"... {method}: {url}")
    if method != "post":
       raise StageException(f"RestAPI does not support for {method} action.")
    return {"records": 1000}

The above workflow template is main executor pipeline that you want to do. If you want to schedule this workflow, you want to dynamic its parameters change base on execution time such as run-date should change base on that workflow running date.

from ddeutil.workflow import Workflow, Result

workflow: Workflow = Workflow.from_conf('run-py-local')
result: Result = workflow.execute(
   params={"source-extract": "USD-THB", "run-date": "2024-01-01"}
)

Note

So, this package provide the Schedule template for this action, and you can pass the parameters dynamically for changing align with that running time by the release prefix.

schedule-run-local-wf:

   # Validate model that use to parsing exists for template file
   type: Schedule
   workflows:

      # Map existing workflow that want to deploy with scheduler application.
      # It allows you to pass release parameter that dynamic change depend on the
      # current context of this scheduler application releasing that time.
      - name: run-py-local
        params:
          source-extract: "USD-THB"
          run-date: "${{ release.logical_date }}"

The main method of the Schedule model that use to running is pending. If you do not pass the stop date on this method, it will use config with WORKFLOW_APP_STOP_BOUNDARY_DELTA key for generate this stop date.

from ddeutil.workflow import Schedule

(
   Schedule
   .from_conf("schedule-run-local-wf")
   .pending(stop=None)
)

Warning

The scheduler feature is the expensive feature of this project. You should avoid to use it and find a scheduler tool instead.

🍪 Configuration

The main configuration that use to dynamic changing this workflow engine for your objective use environment variable only. If any configuration values do not set yet, it will use default value and do not raise any error to you.

Important

The config value that you will set on the environment should combine with prefix, component, and name which is WORKFLOW_{component}_{name} (Upper case).

Name	Component	Default	Description
REGISTRY_CALLER	Core	.	List of importable string for the call stage.
REGISTRY_FILTER	Core	ddeutil.workflow.templates	List of importable string for the filter template.
CONF_PATH	Core	./conf	The config path that keep all template .yaml files.
TIMEZONE	Core	Asia/Bangkok	A Timezone string value that will pass to ZoneInfo object.
STAGE_DEFAULT_ID	Core	false	A flag that enable default stage ID that use for catch an execution output.
STAGE_RAISE_ERROR	Core	false	A flag that all stage raise StageException from stage execution.
MAX_CRON_PER_WORKFLOW	Core	5
MAX_QUEUE_COMPLETE_HIST	Core	16
GENERATE_ID_SIMPLE_MODE	Core	true	A flog that enable generating ID with md5 algorithm.
DEBUG_MODE	Log	true	A flag that enable logging with debug level mode.
FORMAT	Log	%(asctime)s.%(msecs)03d (%(name)-10s, %(process)-5d,%(thread)-5d) [%(levelname)-7s] %(message)-120s (%(filename)s:%(lineno)s)
FORMAT_FILE	Log	{datetime} ({process:5d}, {thread:5d}) {message:120s} ({filename}:{lineno})
DATETIME_FORMAT	Log	%Y-%m-%d %H:%M:%S
TRACE_PATH	Log	./logs	The log path of the workflow saving log.
TRACE_ENABLE_WRITE	Log	false
AUDIT_PATH	Log	./audits
AUDIT_ENABLE_WRITE	Log	true	A flag that enable logging object saving log to its destination.
MAX_PROCESS	App	2	The maximum process worker number that run in scheduler app module.
MAX_SCHEDULE_PER_PROCESS	App	100	A schedule per process that run parallel.
STOP_BOUNDARY_DELTA	App	'{"minutes": 5, "seconds": 20}'	A time delta value that use to stop scheduler app in json string format.

API Application:

This config part use for the workflow application that build from the FastAPI only.

Environment	Component	Default	Description
ENABLE_ROUTE_WORKFLOW	API	true	A flag that enable workflow route to manage execute manually and workflow logging.
ENABLE_ROUTE_SCHEDULE	API	true	A flag that enable run scheduler.

🚀 Deployment

This package able to run as an application service for receive manual trigger from any node via RestAPI or use to be Scheduler background application like crontab job but via Python API or FastAPI app.

API Server

This server use FastAPI package to be the base application.

(.venv) $ uvicorn ddeutil.workflow.api:app \
  --host 127.0.0.1 \
  --port 80 \
  --no-access-log

Note

If this package already deploy, it is able to use multiprocess; uvicorn ddeutil.workflow.api:app --host 127.0.0.1 --port 80 --workers 4

Local Schedule

Warning

This CLI does not implement yet.

(.venv) $ ddeutil-workflow schedule

Docker Container

Build a Docker container from this package.

$ docker build -t ddeutil-workflow:latest -f .container/Dockerfile .
$ docker run -i ddeutil-workflow:latest ddeutil-workflow

💬 Contribute

I do not think this project will go around the world because it has specific propose, and you can create by your coding without this project dependency for long term solution. So, on this time, you can open the GitHub issue on this project 🙌 for fix bug or request new feature if you want it.