werkit

Toolkit for encapsulating Python-based computation into deployable and distributable tasks.

Provides code that helps package things up:

• Serializing results
• Handling and serializing errors
• Deploying task workers using Redis, RQ and the Fargate CLI

They're particularly useful for providing repsonse consistency across different revisions of a service or different services.

Installation

pip install werkit

Usage

from werkit import Manager

def myfunc(param, verbose=False, handle_exceptions=True):
    with Manager(handle_exceptions=handle_exceptions, verbose=verbose) as manager:
        manager.result = do_some_computation()
    return manager.serialized_result

Parallel computation

Werkit supports parallel computation using Redis and RQ.

You must install the dependencies separately:

pip install redis rq

Requesting work

from mylib import myfunc
from werkit.parallel import invoke_for_each


items = {'a': ..., 'b': ...}
job_ids = invoke_for_each(myfunc, items, connection=Redis.from_url(...))

Performing work

pip install redis rq
rq worker --burst werkit-default --url rediss://...

Note: mylib.myfunc must be importable.

Using CloudManager

In place of the low-level API you can make your calls using CloudManager:

#!/usr/bin/env python


import click
from werkit.parallel import Config, CloudManager, invoke_for_each

manager = CloudManager(
    config=Config(
        local_repository="my-project",
        ecr_repository="123456789012.dkr.ecr.us-east-1.amazonaws.com/my-project",
        ecs_task_name="my-project",
        task_args=[
            "--cpu",
            "1024",
            "--memory",
            "2048",
            "--task-role",
            "arn:aws:iam::123456789012:role/...",
            "--security-group-id",
            "sg-...",
            "--subnet-id",
            "subnet-...",
        ],
        default_task_count=5,
    )
)


@click.group()
def cli():
    pass


@cli.command()
def login():
    manager.login()


@cli.command()
@click.argument("tag")
def build_and_push(tag):
    manager.build_and_push()


@cli.command()
def enqueue():
    from myproject import myfunc

    items = {"key1": "value1", "key2": "value2"}

    invoke_for_each(
        measure_body,
        items,
        clean=True,
        connection=manager.redis_connection,
    )


@cli.command()
@click.option(
    "--count",
    default=manager.config.default_task_count,
    type=int,
    help="Number of tasks to run",
)
@click.argument("tag")
def run(count, tag):
    manager.run(tag=tag, count=count)


@cli.command()
def dashboard():
    manager.dashboard()


@cli.command()
def ps():
    manager.ps()


@cli.command()
def get_results():
    print(manager.get_results())


@cli.command()
def clean():
    manager.clean()


if __name__ == "__main__":
    cli()

Getting results

from redis import Redis
from werkit.parallel import get_results


get_results(wait_until_done=True, connection=Redis.from_url(...))

Monitoring

You can monitor your queues using RQ Dashboard or one of the other methods outlined here.

Parallel computation on AWS lambda

Werkit also implements a parallel map on AWS lambda.

Werkit comes with a default lambda handler, that accepts an event of the form {"input":[a, b, ...],"extra_args":[c, d, ...]}. Werkit invokes a lambda function in parallel for every item in input, with an event of the form {"input": a, "extra_args":[c, d, ...]}.

The werkit default handler is configurable via the following environmnent variables:

• LAMBDA_WORKER_FUNCTION_NAME: Name of the lambda worker function to invoke
• LAMBDA_WORKER_TIMEOUT: How long to wait in seconds for the lambda worker function to return before returning a TimeoutError

Building and deploying functions to AWS Lambda

Werkit provides tools for programmatically building and deploying functions to AWS Lambda. There are two distinct steps: build and deploy.

The build process can run natively using a virtualenv, or in Docker. When either of the following cases apply, you can use the native virtualenv method:

• The function's dependencies are pure Python (no compiled extensions).
• You are building the function in Linux.

When building a function using compiled dependencies in OS X, the virtualenv method will try to pack up the OS X dependencies which of course won't work on Lambda. In that case you must use the Docker method.

Building a function natively using a virtualenv

def build_natively(build_dir="build", target_dir="build"):
    import os
    import shutil
    from werkit.aws_lambda.build import (
        collect_zipfile_contents,
        create_venv_with_dependencies,
        create_zipfile_from_dir,
    )

    shutil.rmtree(build_dir, ignore_errors=True)

    venv_dir = os.path.join(build_dir, "venv")
    create_venv_with_dependencies(
        venv_dir=venv_dir,
        # These are the defaults, which you can override if necessary.
        upgrade_pip=True,
        install_wheel=True,
        install_werkit=False,
        install_requirements_from=["requirements.txt"],
        # You can pass credentials to `pip install`.
        environment={"DEPLOY_TOKEN": DEPLOY_TOKEN},
    )

    contents_dir = os.path.join(build_dir, "contents")
    collect_zipfile_contents(
        target_dir=contents_dir,
        venv_dir=venv_dir,
        src_dirs=["mypackage", "assets"],
        # Specify additional system files to copy to `lib/` inside the zipfile.
        lib_files=[...],
    )

    os.makedirs(target_dir, exist_ok=True)
    temp_path_to_zipfile = os.path.join(target_dir, "function.zip")
    create_zipfile_from_dir(
        dir_path=contents_dir,
        path_to_zipfile=temp_path_to_zipfile
    )

Building a function using Docker

Create a collect_and_zip.py script which will run in Docker, with /target mounted to a folder on the host system.

from werkit.aws_lambda.build import create_zipfile_from_dir

VERBOSE = False

collect_zipfile_contents(
    target_dir="/build/contents",
    venv_dir="/build/venv",
    src_dirs=["mypackage", "assets"],
    # Specify additional system files to copy to `lib/` inside the zipfile.
    lib_files=[...],
)

# To improve performance, zip into the container and copy to the host
# afterward.
create_zipfile_from_dir(
    dir_path="/build/contents",
    path_to_zipfile="/build/function.zip"
)
shutil.copyfile("/build/function.zip", "/target/function.zip")

Create a Dockerfile:

FROM python:3.7

WORKDIR /src

# Install any system dependencies you want to include in the zipfile.
RUN apt-get install -y --no-install-recommends ...
RUN rm -rf /var/lib/apt/lists/*

# Install werkit, along with any other dependencies needed for
# `collect_and_zip.py`.
RUN python3 -m pip install "werkit==0.10.0"

# Optionally receive credentials from the environment.
ARG DEPLOY_TOKEN

# Create the venv. As is necessary for some Docker base images, upgrade pip and
# install wheel.
RUN python3 -m venv /build/venv
RUN /build/venv/bin/pip install --upgrade pip wheel

# Install Python dependencies.
COPY requirements.txt /src/
RUN DEPLOY_TOKEN=${DEPLOY_TOKEN} /build/venv/bin/pip install -r requirements.txt

COPY mypackage/ /src/mypackage/
COPY assets/ /src/assets/
COPY collect_and_zip.py /src/

# Optionally set PYTHONPATH if `collect_and_zip.py` imports any internal source
# code.
ENV PYTHONPATH /src

CMD python3 collect_and_zip.py

Invoke the build:

DEPLOY_TOKEN = "..."

def build_in_docker(build_dir="build", target_dir="build"):
    from executor import execute

    docker_tag = "mypackage-lambda-builder"

    execute(
        "docker",
        "build",
        "-t",
        docker_tag,
        "-f",
        "Dockerfile",
        # Optionally pass credentials for the Docker build process.
        "--build-arg",
        f"DEPLOY_TOKEN={DEPLOY_TOKEN}",
        ".",
    )
    print("Build image created")

    os.makedirs("build/", exist_ok=True)
    execute(
        "docker",
        "run",
        "--volume",
        f"{os.path.abspath("build/")}:/target:Z",
        "-t",
        docker_tag,
    )
    print("Build finished")

Deploying the function

# A Lambda role is always required.
LAMBDA_ROLE = ...
# When the zipfile is larger than 50 MB, a temporary bucket is required.
S3_CODE_BUCKET = ...

def create():
    from werkit.aws_lambda.deploy import perform_create

    perform_create(
        # Region is required.
        aws_region="us-east-1",
        function_name="myfunction",
        local_path_to_zipfile="build/function.zip",
        # The importable name of your handler function, which should have the
        # signature `def handler(event, context):`.
        handler="mypackage.worker.handler",
        role=LAMBDA_ROLE,
        # Required when the zipfile is larger than 50 MB.
        s3_code_bucket=S3_CODE_BUCKET,
        # Optionally override.
        timeout=TIMEOUT,
        memory_size=WORKER_MEMORY_SIZE,
        runtime="python3.8",
        env_vars={},
    )

Updating the code for an existing function

LAMBDA_ROLE = ...
S3_CODE_BUCKET = ...

def update_code():
    from werkit.aws_lambda.deploy import perform_update_code

    perform_update_code(
        aws_region="us-east-1",
        function_name="myfunction",
        local_path_to_zipfile="build/function.zip",
        # Required when the zipfile is larger than 50 MB.
        s3_code_bucket=S3_CODE_BUCKET,
    )

Type definitions

TypeScript types are available for the Werkit job message format.

Contribute

• Issue Tracker: https://github.com/metabolize/werkit/issues
• Source Code: https://github.com/metabolize/werkit

Pull requests welcome!

Support

If you are having issues, please let us know.

License

The project is licensed under the MIT License.