Issues with Private Python Package Installation on LangGraph - dynamic package installation

[majorgilles]

I am continuing my investigations regarding the installation of a private Python package which I am forced to do because LangGraph platform does not allow me to inject an env variable in the container build process which would solve my issues.

I have been experimenting with a dynamic private Python package installation which is working locally (Python code + pytest + LangGraph Studio) but not on the LangGraph cloud platform. This is despite configuring the required environment variables for the runtime execution of the graphs in the LangGraph Platform UI.

As I am unable to inject an environment variable to authenticate our private code artifact repository inside the Docker build process on the LangGraph platform, I have been attempting to install our package dynamically. This is because it's the only way I can have control over the environment variables that are available.

I have developed a Python script that is working locally and in LangGraph Studio (Docker). The script provides utility functions to install the Lizy common package dynamically. Below is the code:

import os

import subprocess

from enum import StrEnum



class CredentialType(StrEnum):

    """Enum for the credential type (either the shared one to install common dynamically or app for the runtime)."""

    APP = "APP"

    SHARED = "SHARED"



def _configure_credentials(credential_type: CredentialType) -> None:

    # Configure command

    aws_access_key_id = os.environ[f"AWS_ACCESS_KEY_ID_{credential_type}"]

    aws_secret_access_key = os.environ[f"AWS_SECRET_ACCESS_KEY_{credential_type}"]

    aws_configure_command = f"aws configure set aws_access_key_id {aws_access_key_id}"

    subprocess.check_output(aws_configure_command, shell=True).decode("utf-8")

    aws_configure_command = f"aws configure set aws_secret_access_key {aws_secret_access_key}"

    subprocess.check_output(aws_configure_command, shell=True).decode("utf-8")



def install_common_package(version: str) -> None:

    """Install the lizy common package dynamically."""

    # We need to configure the credentials for the common package installation (in lizy-shared AWS account)

    _configure_credentials(CredentialType.SHARED)

    # Construct the AWS CLI command

    aws_command = (

        "aws codeartifact login"

        " --tool pip "

        " --domain lizy-packages-domain"

        " --domain-owner 570466319168"

        " --repository lizy-packages-repository"

    )

    # Run the AWS CLI command and capture its output

    output = subprocess.check_output(aws_command, shell=True).decode("utf-8")

    print(f"Subcommand:\n {output}")

    pip_command = f"pip install common=={version}"

    print(f"Pip command:\n {pip_command}")

    result = subprocess.check_call(pip_command, shell=True)

    print(f"Finish installing common=={version}: {result}")

    # We need to reconfigure the credentials for the runtime execution of the app (either dev / staging / prod account)

    _configure_credentials(CredentialType.APP)

I have input all the required variables in the LangGraph platform, and the Docker build phase passes. However, the "Deploy" phase does not. The error message is as follows:

File "/usr/local/lib/python3.12/subprocess.py", line 571, in run

raise CalledProcessError(retcode, process.args,

subprocess.CalledProcessError: Command 'aws codeartifact login --tool pip  --domain lizy-packages-domain --domain-owner 570466319168 --repository lizy-packages-repository' returned non-zero exit status 255.

Unfortunately, I have zero visibility over what is going wrong since the logs do not provide sufficient information. Moreover, the deploy times out after 300 seconds.

[majorgilles]

In the end, this was caused by previous AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY defined in LangGraph Cloud's UI which were interfering with aws configure's behaviour.

[majorgilles]

Still a a pain to have to resort to unorthodox ways of accessing my private artefacts. Please make it possible to configure the env variables of the langgraph cloud docker build process.

[andrewnguonly]

Please make it possible to configure the env variables of the langgraph cloud docker build process.

Yes, we will support this in the future.