PyFlink 1.15 development on Apple Silicon Mac

🚨 This limitation only applies to Flink 1.15, and only for local development on Apple Silicon macOS machines.
If you are using Flink 1.18 or later, you can install PyFlink libraries on your macOS machine as any other Python library.
See the Getting Started for more details.

Problem

There are known issues with installing PyFlink dependencies on Mac using Apple Silicon chips (see FLINK-26981). These issues, resolved in Flink 1.16.x, make it difficult to develop PyFlink 1.15.x applications on new Mac machines, using IDE like PyCharm or Visual Studio Code.

Solution overview

Use a Docker image built for Intel architecture (amd64) with all required Python dependencies.
Enable Rosetta emulation in Docker, to run on Apple Silicon Mac images for Intel architecture.
Leverage the ability of advanced IDE to run the Python interpreter inside a Docker container.

Step-by-step instructions

Requirements

• An IDE supporting the execution of Python interpreter inside a Docker container:
  • PyCharm Professional (Community Edition does not support this, directly)
  • Visual Studio Code
• Docker running on the Mac machine; the instructions will refer to Docker Desktop
• Install IDE plugins to integrate with Docker and AWS
  • PyCharm Docker Plugin
  • AWS Toolkit for PyCharm
  • AWS Toolkit for Visual Studio Code

Docker configuration

1. Enable Rosetta emulation

   In Docker Desktop: Settings > Feature in development > Beta features > Use Rosetta for x86/amd64 emulation on Apple Silicon: enable

2. Enable Docker socket, required for IDE integration

   In Docker Desktop: Settings > Advanced > Allow the default docker socket to be used : enable

Dockerfile

The provided Dockerfile is based on what recommended in Flink 1.15 documentation.

We install Python 3.8, boto3, that is often required if you connect to AWS API directly, and AWS CLI for troubleshooting.

FROM flink:1.15.4

ARG PYTHON_VERSION=3.8.9

RUN apt-get update -y && \
apt-get install -y build-essential libssl-dev zlib1g-dev libbz2-dev libffi-dev && \
wget https://www.python.org/ftp/python/${PYTHON_VERSION}/Python-${PYTHON_VERSION}.tgz && \
tar -xvf Python-${PYTHON_VERSION}.tgz && \
cd Python-${PYTHON_VERSION} && \
./configure --without-tests --enable-shared && \
make -j6 && \
make install && \
ldconfig /usr/local/lib && \
cd .. && rm -f Python-${PYTHON_VERSION}.tgz && rm -rf Python-${PYTHON_VERSION} && \
ln -s /usr/local/bin/python3 /usr/local/bin/python && \
apt-get clean && \
rm -rf /var/lib/apt/lists/*

RUN pip3 install apache-flink==1.15.4 boto3 awscli

The docker image must be built for the linux/amd64 platform not to face the Python dependency issues.

There are two options:

• a) Build the image manually and instruct the IDE to use the specific image. In this case the image can be pulled from an Image Repository.
• b) Provide the Dockerfile to the IDE and instruct the IDE to build the image for linux/amd64.

For option (b) you need to copy the Dockerfile in your project folder and skip the next step.

Build the Docker image for "Intel" architecture

This step is required for option (a), to build the docker image.

From the directory containing the Dockerfile, run the following command:

docker build --platform linux/amd64 --tag pyflink-115:latest .

This can take a few minutes and will create an image, named pyflink-115 (you can choose a different name).

PyCharm: Configure to run the Python interpreter in Docker

1. Create Docker execution environment

   Settings > Build, Execution, Deployment > Docker

   • Add new Docker environment
     • Select Docker for Mac
     • Verify “connection successful” is reported at the bottom of the window,
       If it cannot connect, verify you have enabled the default docker socket and restarted docker daemon

   

2. Set the interpreter for the project

   Project Settings > Python Interpreter : Add Interpreter : On Docker

   • Select: Existing Docker environment

   • Image - depending on the option you chose, above

     • a) Pull or use existing

       • type the name of the image (pyflink-115:latest)
       • 

     • b) Build

       • Dockerfile: select the dockerfile you previously copied in the project folder
       • Image tag: pyflink-115:latest (you can choose a different tag)
       • Build options: --platform linux/amd64
       • Uncheck Rebuild image automatically every time before running code (the image takes a few minutes to build and there is no reason of rebuilding on every run, unless you modify the Dockerfile)
       • 

   • Next (Pull Docker Image or Build Image)

   • Next (Project directory and language runtime configuration)

     • Select: System Interpreter (/usr/local/bion/python3)

   • Create

Configure AWS Toolkit plugin to provide the AWS credential to the running application

Configure the AWS Toolkit to get a valid AWS authentication with access to all resources the Flink application must access (e.g. Kinesis Data Streams, S3 buckets...)
The configuration depends on the way you provide the credentials (refer to the AWS Toolkit documentation).

Also configure the script Run Configuration accordingly, e.g. to “Use the currently selected credential profile/region”.

Configuring Visual Studio Code

Refer to Visual Studio Code documentation to develop with Python inside a container.

Refer to AWS Toolkit documentation to pass the AWS authentication to the PyFlink application.

Additional notes

While experimenting with the docker image, you may end up with many dangling images. As each PyFlink image is 2.1 GB you may quickly run out of space.
Periodically check the images

docker image ls --all

...and prune all dangling ones

docker image prune

Conclusions

Following these steps you can set up a local development environment for PyFlink 1.15 on Mac using Apple Silicon Chips and use IDE like PyCharm or Visual Studio Code to run and debug the application on your Mac development machine.

No change to the application is required.

These steps will not affect the way you package your application for the environment where you deploy your application, whether it is a self-managed Apache Flink cluster or Amazon Managed Service for Apache Flink.

📌 출처: https://github.com/aws-samples/amazon-managed-service-for-apache-flink-examples.git