flowmancer 0.10.1

The Python Thing-Doer

Flowmancer

Flowmancer aims to help you do things in a sequential or parallel manner. It enables you to write tasks in Python, describe their order, then execute them with as little effort as possible.

But why do I need this? Couldn't I just write my own Python code to do stuff?

You certainly could!

Though Flowmancer provides gives you a head-start to building your custom processes with optional add-ons for logging, checkpoint/restarts in the event of failures, or even custom task observers to do...things while your things do things!

Installation

Simply install the flowmancer package with:

pip3 install flowmancer

NOTE: flowmancer supports only Python 3.7 and higher.

Basic Usage

Let's assume you have a new project with a basic structure like so:

my_project
├─ job.yaml
├─ main.py
└─ tasks/
   ├─ __init__.py
   └─ mytasks.py

To use flowmancer, you'll need to provide a few things:

• Task implementations (mytasks.py)
• A job YAML file (job.yaml)
• Your main/driver code (main.py)

Tasks

By default, Flowmancer recursively searches in the ./tasks directory (relative to where Flowmancer() is initialized - in this case, main.py) for Task implementations decorated with @task. See the Advanced Usage section for details on how to add other directories or packages that contain Task implementations.

A flowmancer task is simply a class that extends the Task abstract class, which, at minimum requires that the run method be implemented:

import time
from flowmancer.task import Task, task

@task
class WaitAndSucceed(Task):
    # All variables should be given type hints and optional vars should be given default values.
    my_required_string_var: str
    my_optional_int_var: int = 5

    def run(self):
        # Store string input var in the shared dictionary accessible by other tasks.
        self.shared_dict["my_var"] = f"Hello from: {self.my_required_string_var}!"

        # Sleep for seconds defined by input var (using default of 5).
        print(f"Starting up and sleeping for {self.my_optional_int_var} seconds!")
        time.sleep(self.my_optional_int_var)
        print("Done!")

@task
class ImmediatelySucceed(Task):
    def run(self):
        # Print statements will automatically be sent to configured loggers.
        print("Success!")

@task
class FailImmediately(Task):
    def run(self):
        print(f"Printing `my_var` value: {self.shared_dict['my_var']}")
        # Raise errors to cause tasks to fail and additionally block dependent tasks, if any.
        raise RuntimeError("Let this be caught by Flowmancer")

Any print() or exceptions will write log messages to any configured loggers (zero or more loggers may be defined).

Job Definition YAML File

This file describes what code to run, in what order, as well as additional add-ons to supplement the job during execution:

version: 0.1

# This entire config block is currently optional, however, it is recommended to at least provide a unique name for each
# Job Definition YAML file, as this name is used for checkpointing jobs in the event of failures.
config:
  name: 'my-flowmancer-job'

tasks:
  # No dependency - run right away
  # Add `parameters` key-value pairs for any required and optional task variables.
  succeed-task-a:
    task: WaitAndSucceed
    parameters:
      my_required_string_var: "My First Task!"

  # No dependency - run right away
  succeed-task-b:
    task: ImmediatelySucceed

  # Only run if prior 2 tasks complete successfully
  final-fail-task:
    task: FailImmediately
    max_attempts: 3  # Retry uup to 2 times upon failure (1 initial exec + 2 retries = 3 attempts)
    dependencies:
      - succeed-task-a
      - succeed-task-b

Driver

The driver is super simple and simply requires running an instance of Flowmancer

# main.py
import sys
from flowmancer import Flowmancer

if __name__ == '__main__':
    # The `start()` method will return a non-zero integer on failure, typically equal to the number of failed tasks.
    ret = Flowmancer().start()

    # Exceptions from tasks will be captured and logged, rather than being raised up to this level. To cause this
    # driver program to fail, either explicitly raise your own error OR call sys.exit.
    if ret:
      raise RuntimeError('Flowmancer job has failed!')

    # Alternatively, instead of crashing w/ an exception, simply exit with a non-zero value.
    # sys.exit(ret)

Executing the Job

python3 main.py -j ./path/to/job.yaml

To run from point-of-failure (if any):

python3 main.py -j ./path/to/job.yaml -r

If no prior failure is detected, the job will start as if no -r flag were given.

Note that the job definition must still be provided with the -r flag.

Advanced Usage

Optional Configurations

In the config block of the Job Definition, the following optional parameters may be given:

Parameter	Type	Default Value	Description
name	str	'flowmancer'	Name/identifier for Job Definition. Used for saving checkpoints used for job restarts in the event of a failure.
max_concurrency	int	0	Maximum number tasks that can run in parallel. If 0 or less, then there is no limit.
extension_directories	List[str]	[]	List of paths, either absolute or relative to driver .py file, that contain any @task, @logger, or @extension decorated classes to make accessible to Flowmancer. The ./task, ./extensions, and ./loggers directories are ALWAYS checked by default.
extension_packages	List[str]	[]	List of installed Python packages that contain @task, @logger, or @extension decorated classes to make accessible to Flowmancer.
synchro_interval_seconds	float	0.25	Core execution loop interval for waking and checking status of tasks and whether loggers/extensions/checkpointer should trigger.
loggers_interval_seconds	float	0.25	Interval in seconds to wait before emitting log messages to configured Logger instances.
extensions_interval_seconds	float	0.25	Interval in seconds to wait before emitting state change information to configured Extension instances.
checkpointer_interval_seconds	float	10.0	Interval in seconds to wait before writing checkpoint information to the configured Checkpointer.

For example:

config:
  name: 'most-important-job'
  max_concurrency: 20
  extension_directories:
    - ./client_implementations
    - /opt/flowmancer/tasks
  extension_packages:
    - internal_flowmancer_package

Include YAML Files

An optional include block may be defined in the Job Definition in order to merge multiple Job Definition YAML files. YAML files are provided in a list and processed in the order given, with the containing YAML being processed last.

For example:

# <app_root_dir>/jobdefs/template.yaml
config:
  name: generic-template

tasks:
  do-something:
    task: DoSomething
    parameters:
      some_required_param: I am a required string parameter

# <app_root_dir>/jobdefs/cleanup_addon.yaml
include:
  - ./jobdefs/template.yaml

tasks:
  cleanup:
    task: Cleanup
    dependencies:
      - do-something

# <app_root_dir>/jobdefs/complete.yaml
config:
  name: complete-job

include:
  - ./jobdefs/cleanup_addon.yaml

tasks:
  do-something:
    task: Do Something
    parameters:
      added_optional_param: 99

Loading the complete.yaml job definition will result in a YAML equivalent to:

config:
  name: complete-job

tasks:
  do-something:
    task: Do Something
    parameters:
      some_required_param: I am a required string parameter
      added_optional_param: 99

:warning: Array values are NOT merged like dictionaries are. Any array values (and therfore any nested structures) within them will be replaced if modified in a later YAML.

Additionally, the above example could have all include values in the complete.yaml file and the include block removed from cleanup_addon.yaml:

# <app_root_dir>/jobdefs/complete.yaml
config:
  name: complete-job

# As with most paths in Job Definition, paths to `include` YAML files are relative to `.py` file where the `.start()`
# method for Flowmancer is invoked.
include:
  - ./jobdefs/template.yaml
  - ./jobdefs/cleanup_addon.yaml

tasks:
  do-something:
    task: Do Something
    parameters:
      added_optional_param: 99

The include values are processed in order and results in the same outcome as the original example.

Changing Default File Logger Directory

The Job Definition accepts an optional loggers section, which if left empty will default to using a FileLogger with default settings. To utilize the default FileLogger, but with a different configuration, explicitly provide the loggers block:

loggers:
  my-file-logger:
    logger: FileLogger
    parameters:
      # NOTE: this path is relative to the `.py` file where `Flowmancer().start()` is invoked.
      base_log_dir: ./my_custom_log_dir  # ./logs is the default, if omitted.
      retention_days: 3  # 10 is the default, if omitted.

Complex Parameters

While this is mostly used for Task implementations, the details outlined here apply for any built-in and custom Extension and Logger implementations.

Flowmancer makes heavy use of Pydantic to validate parameters and ensure that values loaded from the Job Definition are of the appropriate type.

This means that a Task can have complex types (including custom models) like:

from enum import Enum
from flowmancer.task import Task, task
from pydantic import BaseModel
from typing import Dict, List

class Protocol(Enum):
    HTTP: 'HTTP'
    HTTPS: 'HTTPS'

class APIDetails(BaseModel):
    protocol: Protocol = Protocol.HTTPS
    base_url: str
    endpoint: str

@task
class DownloadDataFromRestApi(Task):
    api_details: APIDetails
    target_dir: str
    target_filename: str = 'data.json'

    def run(self) -> None:
        url = f'{self.api_details.protocol}://{self.api_details.base_url}/{self.api_details.endpoint}'
        # Continued implementation...

And the Job Definition snippet for this task might be:

tasks:
  download-file-one:
    task: DownloadDataFromRestApi
    parameters:
      api_details:
        # We leave out `protocol` because we want to just use the default `HTTPS` value.
        base_url: www.some_data_api.com
        endpoint: /v1/data/weather/today
      target_dir: /data/todays_weather
      # Override the default `target_filename` value given in the class implementation.
      target_filename: weather.json

Task Lifecycle Methods

In addition to the required run method, an implementation of Task may optionally include the following methods:

Method	Required	Order	Description
on_create	No	1	First method executed when a task is released for execution. Note that a task is not considered "created" until it enters the RUNNING state.
on_restart	No	2	Executed only if a task is running from the result of a recovery from FAILED state. If a task was failed in DEFAULTED state, this method will not be executed.
run	Yes	3	Always required and always executed once task is in RUNNING state, unless prior lifecycle methods have failed.
on_success	No	4	Executed only if run method ends in success.
on_failure	No	5	Executed only if run method ends in failure/exception.
on_destroy	No	6	Always executed after all other lifecycle methods.
on_abort	No	-	Executed when SIGINT signal is sent to tasks/Flowmancer.

Just as with run, all lifecycle methods have access to self.shared_dict and any parameters.

Custom Loggers

Custom implementations of the Logger may be provided to Flowmancer to either replace OR write to in addition to the default FileLogger.

A custom implementation must extend the Logger class, be decorated with the logger decorator, and implement the async update method at minimum:

@logger
import json
import requests
from flowmancer.loggers.logger import Logger, logger

class SlackMessageLogger(Logger):
    webhook: str

    def _post_to_slack(self, msg: str) -> None:
        requests.post(
            self.webhook,
            data=json.dumps({'text': title, 'attachments': [{'text': msg}]}),
            headers={'Content-Type': 'application/json'},
        )

    async def update(self, evt: SerializableLogEvent) -> None:
        # The `LogStartEvent` and `LogEndEvent` events only have a `name` property.
        if isinstance(evt, LogStartEvent):
            self._post_to_slack(f'[{evt.name}] START: Logging is beginning')
        elif isinstance(evt, LogEndEvent):
            self._post_to_slack(f'[{evt.name}] END: Logging is ending')
        # The `LogWriteEvent` additionally has `severity` and `message` properties.
        elif isinstance(evt, LogWriteEvent):
            self._post_to_slack(f'[{evt.name}] {evt.severity.value}: {evt.message}')

The Logger implementation may also have the following optional async lifecycle methods:

• on_create
• on_restart
• on_success
• on_failure
• on_destroy
• on_abort

To incorporate your custom Logger into Flowmancer, ensure that it exists in a module either in ./loggers or in a module listed in config.extension_directories in the Job Definition.

This allows it to be provided in the loggers section of the Job Definition.

:warning: Providing the loggers section will remove the default logger (FileLogger) from your job's configuration. If you want to add your custom logger alongside the default logger, the FileLogger must explicitly be configured.

loggers:
  # Load the default logger with default parameters
  default-logger:
    logger: FileLogger

  # Custom logger implementation
  slack-logger:
    logger: SlackMessageLogger
    parameters:
      webhook: https://some.webhook.url

Custom Extensions

Coming soon.

Custom Checkpointers

Custom implementations of the Checkpointer may be provided to Flowmancer to replace the default FileCheckpointer.

:warning: Unlike loggers and extensions, only one checkpointer can be configured per Job Definition.

A custom implementation must extend the Checkpointer class, be decorated with the checkpointer decorator, and implement the async write_checkpoint, read_checkpoint, and clear_checkpoints methods at minimum. It may also optinoally implement async lifecycle methods, similar to Custom Loggers:

from .checkpointer import CheckpointContents, Checkpointer, NoCheckpointAvailableError, checkpointer

@checkpointer
class DatabaseCheckpointer(Checkpointer):
    host: str
    port: int
    username: str
    password: str

    def write_checkpoint(self, name: str, content: CheckpointContents) -> None:
        # Store checkpoint state - must be able to store contents of
        # `CheckpointContents` in a way that it can be reconstructed later.

    def read_checkpoint(self, name: str) -> CheckpointContents:
        # Recall checkpoint state - reconstruct and return `CheckpointContents`
        # if exists for `name`. Otherwise raise `NoCheckpointAvailableError`
        # to indicate no valid checkpoint exists to restart from.

    def clear_checkpoint(self, name: str) -> None:
        # Remove checkpoint state for `name`.

To incorporate your custom Checkpointer into Flowmancer, ensure that it exists in a module either in ./extensions or in a module listed in config.extension_directories in the Job Definition.

This allows it to be provided in the checkpointer section of the Job Definition:

checkpointer:
  checkpointer: DatabaseCheckpointer
  parameters:
    host: something
    port: 9999
    username: user
    password: 1234