Context: TruEra Model Execution Environments

Executing Python models in TruEra infrastructure requires (1) the creation of a virtual environment (via conda), and (2) populating that virtual environment with your model's dependencies. 

These dependencies are then used execute the model on the configured data, generating predictions and feature influences that underpin many of TruEra Diagnostics' analytics. 

Default PyPi and Conda repositories

To satisfy your model's dependencies, The TruEra service that executes your model reaches out to public-facing pip & conda repos on the internet:

Pip: https://pypi.python.org/simple

Conda: https://repo.anaconda.com/pkgs

If you cannot successfully ping these URLs from the host that is running your deployment of TruEra, you may need to modify the default pip & conda configuration files on the host. 

These modifications can include the use of proxy servers, internal channels (mirrors) of these repositories, and/or additional security configurations related to SSL communication.

Configuring custom PyPi or Conda Repo

By default TruEra ships with pip.conf and .condarc files pointing to the default public repositories. In many enterprise installations, access to public repositories is blocked. In these cases, we can configure the python model runner to use local (on-premise / within the same virtual network) conda and pip repositories.

Required Information: pip & conda

1. A private pip server index URL, which is used to locate the index for the pip packages.
2. Privately hosted conda channel URLs.

Particular deployments might need reconfiguring other configurations as well. So we allow users to update the whole pip.conf and .condarc based on their requirements.

Pip & Conda reference documentation

Pip.conf

For more details on pip.conf please follow:

Configuration - pip documentation v21.2.4

.Condarc

For more details on .condarc please follow:

Using the .condarc conda configuration file - conda 4.10.3.post40+c1579681 documentation

With these two pieces of information, we can update the pip.conf and .condarc in the deployment to point to the correct (and accessible) repositories. 

Configuring in Singlebox Mode

The updated/environment-specific pip.conf and .condarc should be copied to the config.override directory of the installation.

Configuring in Kubernetes

pip.conf and .condarc are configured via K8s configmap model-runner-python-configmap.yaml. When using helm , .Values.modelrunner.python.config.pipConf and .Values.modelrunner.python.config.condarc are used as the configuration files verbatim.

Additional Configuration Examples: SSL & proxy use

SSL issues while fetching packages?

Mitigation:

For pip.conf configure the trusted-host appropriately, example:

[install]
trusted-host =
    pypi.org
    pypi.python.org
    files.pythonhosted.org

For .condarc configure ssl_verify: false.

Configuring proxy to contact to internet

For pip.conf

[global]
proxy = <http://user>:password@proxy_name:port

For .condarc

proxy_servers:
    http: <http://user:pass@corp.com:8080>
    https: <https://user:pass@corp.com:8080>