4.3. Using wc_env_manager build, version, and sharing computing environments for WC modeling

WC modeling requires complex computing environments with numerous dependencies. wc_env_manager helps modelers and software developers setup customizable computing environments for developing, testing, and running whole-cell (WC) models and WC modeling software. This makes it easy for modelers and software developers to install and configure the numerous dependencies required for WC modeling. This helps modelers and software developers focus on developing WC models and software tools, rather than on installing, configuring, and maintaining complicated dependencies.

In addition, wc_env_manager facilitates collaboration by helping WC modelers and software developers share a common base computing environment with third party dependencies. Furthermore, wc_env_manager helps software developers anticipate and debug issues in deployment by enabling developers to replicate similar environments to those used to test and deploy WC models and tools in systems such as Amazon EC2, CircleCI, and Heroku.

wc_env_manager uses Docker to setup a customizable computing environment that contains all of the software packages needed to run WC models and WC modeling software. This includes

  • Required non-Python packages

  • Required Python packages from PyPI and other sources

  • WC models and WC modeling tools

  • Optionally, local packages on the user’s machine such as clones of these WC models and WC modeling tools

wc_env_manager supports both the development and deployment of WC models and WC modeling tools:

  • Development: wc_env_manager can run WC models and WC modeling software that is located on the user’s machine. This is useful for testing WC models and WC modeling software before committing it to GitHub.

  • Deployment: wc_env_manager can run WC models and WC modeling software from external sources such as GitHub.

4.3.1. How wc_env_manager works

wc_env_manager is based on Docker images and containers which enable virtual environments within all major operating systems including Linux, Mac OSX, and Windows, and the DockerHub repository for versioning and sharing virtual environments.

  1. wc_env_manager creates a Docker image, wc_env_dependencies with the third-party dependencies needed for WC modeling or pulls this image from DockerHub. This image represents an Ubuntu Linux machine.

  2. wc_env_manager uses this Docker image to create another Docker image, wc_env with the WC models, WC modeling tools, and the configuration files and authentication keys needed for WC modeling.

  3. wc_env_manager uses this image to create a Docker container to run WC models and WC modeling tools. Optionally, the container can have volumes mounted from the host to run code on the host inside the Docker container, which is helpful for using the container to test and debug WC models and tools.

The images and containers created by wc_env_manager can be customized using a configuration file.

4.3.2. Installing wc_env_manager

First, install the following requirements. See Section 6 for detailed instructions.

Second, run the following command to install the latest version of wc_env_manager from GitHub:

pip install git+https://github.com/KarrLab/wc_env_manager.git#egg=wc_env_manager

4.3.3. Using wc_env_manager to build and share images for WC modeling

Administrators should follow these steps to build and disseminate the wc_env and wc_env_dependencies images.

  1. Create contexts for building the wc_env and wc_env_dependencies Docker images.

  2. Create Dockerfile templates for the wc_env and wc_env_dependencies Docker images.

  3. Set the configuration for wc_env_manager.

  4. Use wc_env_manager to build the wc_env and wc_env_dependencies Docker images.

  5. Use wc_env_manager to push the wc_env and wc_env_dependencies Docker images to DockerHub.

4.3.3.1. Creating contexts for building the wc_env and wc_env_dependencies images

First, create contexts for building the images. This can include licenses and installers for proprietary software packages.

  1. Prepare CPLEX installation

    1. Download CPLEX installer from https://ibm.onthehub.com

    2. Save the installer to the base image context

    3. Set the execution bit for the installer by running chmod ugo+x /path/to/installer

  2. Prepare Gurobi installation

    1. Create license at http://www.gurobi.com/downloads/licenses/license-center

    2. Copy the license to the gurobi_license build argument for the base image in the wc_env_manager configuration

  3. Prepare Mosek installation

    1. Request an academic license at https://license.mosek.com/academic/

    2. Receive a license by email

    3. Save the license to the context for the base image as mosek.lic

  4. Prepare XPRESS installation

    1. Install the XPRESS license server on another machine

      1. Download XPRESS from https://clientarea.xpress.fico.com

      2. Use the xphostid utility to get your host id

      3. Use the host id to create a floating license at https://app.xpress.fico.com

      4. Save the license file to the context for the base image as xpauth.xpr

      5. Run the installation program and follow the onscreen instructions

    2. Copy the IP address or hostname of the license server to the xpress_license_server build argument for the base image in the wc_env_manager configuration.

    3. Save the license file to the context for the base image as xpauth.xpr.

    4. Edit the server property in the first line of xpauth.xpr in the context for the base image. Set the property to the IP address or hostname of the license server.

4.3.3.2. Creating Dockerfile templates for wc_env and wc_env_dependencies

Second, create templates for the Dockerfiles to be rendered by Jinja, and save the Dockerfiles within the contexts for the images. The default templates illustrate how to create the Dockerfile templates.

  • /path/to/wc_env_manager/wc_env_manager/assets/base-image/Dockerfile.template

  • /path/to/wc_env_manager/wc_env_manager/assets/image/Dockerfile.template

4.3.3.3. Setting the configuration for wc_env_manager

Third, Set the configuration for wc_env_manager by creating a configuration file ./wc_env_manager.cfg following the schema outlined in /path/to/wc_env_manager/wc_env_manager/config/core.schema.cfg and the defaults in /path/to/wc_env_manager/wc_env_manager/config/core.default.cfg.

  • Set the repository and tags for wc_env and wc_env_dependencies.

  • Set the paths for the Dockerfile templates.

  • Set the contexts for building the Docker images and the files that should be copied into the images.

  • Set the build arguments for building the Docker images. This can include licenses for proprietary software packages.

  • Set the WC modeling packages that should be installed into wc_env.

  • Set your DockerHub username and password.

4.3.3.4. Building the wc_env and wc_env_dependencies Docker images

Use the following command to build the wc_env and wc_env_dependencies images:

wc-env-manager build

4.3.3.5. Pushing the wc_env and wc_env_dependencies Docker images to DockerHub

Use the following command to push the wc_env and wc_env_dependencies images to GitHub:

wc-env-manager push

4.3.4. Using wc_env_manager to create and run Docker containers for WC modeling

Developers should follow these steps to build and use WC modeling computing environments (Docker images and containers) to test, debug, and run WC models and WC modeling tools.

  1. Use wc_env_manager to pull existing WC modeling Docker images

  2. Use wc_env_manager to create Docker containers with volumes mounted from the host and installations of software packages contained on the house

  3. Run models and tools inside the Docker containers created by wc_env_manager

4.3.4.1. Pulling existing Docker images

First, use the following command to pull existing WC modeling Docker images. This will pull both the base image with third part dependencies, wc_env_dependencies, and the image with WC models and modeling tools, wc_env.:

wc-env-manager pull

The following commands can also be used to pull the individual images.:

wc-env-manager base-image pull
wc-env-manager image pull

4.3.4.2. Building containers for WC modeling

Second, set the configuration for the containers created by wc_env_manager by creating a configuration file ./wc_env_manager.cfg following the schema outlined in /path/to/wc_env_manager/wc_env_manager/config/core.schema.cfg and the defaults in /path/to/wc_env_manager/wc_env_manager/config/core.default.cfg.

  • Set the host paths that should be mounted into the containers. This should include the root directory of your clones of WC models and WC modeling tools (e.g. map host:~/Documents to container:/root/Documents-Host).

  • Set the WC modeling packages that should be installed into wc_env. This should be specified in the pip requirements.txt format and should be specified in terms of paths within the container. The following example illustrates how install clones of wc_lang and wc_utils mounted from the host into the container.:

    /root/Documents-Host/wc_lang
    /root/Documents-Host/wc_utils
    

Third, use the following command to use wc_env to construct a Docker container.:

wc-env-manager container build

This will print out the id of the created container.

4.3.4.3. Using containers to run WC models and WC modeling tools

Fourth, use the following command to log in the container.:

cd /path/to/wc_env_manager
docker-compose up -d
docker-compose exec wc_env bash

Fifth, use the integrated WC modeling command line program, *wc_cli*, to run WC models and WC modeling tools. For example, the following command illustrates how to get help for the wc_cli program. See the *wc_cli* documentation for more information.:

container >> wc-cli --help

4.3.5. Using WC modeling computing environments with an external IDE such as PyCharm

The Docker images created with wc_env_manager can be used with external integrated development environments (IDEs) such as PyCharm. See the links below for instructions on how to use these tools with Docker images created with wc_env_manager.

4.3.6. Caveats and troubleshooting

  • Code run in containers created by wc_env_manager can create host files and overwrite existing host files. This is because wc_env_manager mounts host directories into containers.

  • Containers created by wc_env_manager can be used to run code located on your host machine. However, using different versions of Python between your host and the Docker containers can create Python caches and compiled Python files that are incompatible between your host and the Docker containers. Before switching between running code on your host your and the Docker containers, you may need to remove all __pycache__ subdirectories and *.pyc files from host packages mounted into the containers.

  • Code run in Docker containers will not have access to the absolute paths of your host and vice-versa. Consequently, arguments that represent absolute host paths or which contain absolute host paths must be mapped from absolute host paths to the equivalent container path. Similarly, outputs which represent or contain absolute container paths must be mapped to the equivalent host paths.

  • Running code in containers created with wc_env_manager will be slower than running the same code on your host. This is because wc_env_manager is based on Docker containers, which add an additional layer of abstraction between your code and your processor.