2. Installation

This page outlines how to install or update Clara Deploy SDK on your system.

2.1. System Requirements

As of version 0.7.0, the Clara Deploy SDK has been tested with the following system requirements:

  • Ubuntu Linux 18.04 LTS

  • We use the NVIDIA Triton Inference Server 1.15.0 (Container Image tag 20.07-v1-py3). Release 1.15.0 is based on NVIDIA CUDA 11.0.194, which requires NVIDIA Driver release 450 or later. However, if you are running on Tesla (for example, T4 or any other Tesla board), you may use NVIDIA driver release 418.xx or 440.30. The CUDA driver’s compatibility package only supports particular drivers. For a complete list of supported drivers, see the CUDA Application Compatibility topic. For more information, see CUDA Compatibility and Upgrades.

    • Installation of CUDA Toolkit would make both CUDA and NVIDIA Display Drivers available

    • Due to AWS kernel update 5.3.0-1017-aws, nvidia-smi will fail with .. code-block:: guess

      NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver

      This issue affects AWS, Azure and GCP, and it is recommended that the NVIDIA driver is installed using CUDA 10.2 deb package.

  • NVIDIA GPU is Pascal or newer, including Pascal, Volta, Turing and Ampere families

  • Kubernetes 1.15.4

  • Docker 19.03.1

  • NVIDIA Docker 2.2.0

  • Docker configured with nvidia as the default runtime (Prerequisite of NVIDIA device plugin for k8s)

  • Helm 2.15.2

  • At least 30GB of available disk space

It is recommended that the user uses the bootstrap script to install the dependencies for Clara Deploy SDK.

The Clara installation workflow is outlined in the diagram below:

width:800 align:center

2.2. Steps to Install

  1. Download the bootstrap.zip file from NGC.

    1. Log in to NGC.
    2. Select the appropriate org and team.
    3. Navigate to Model Scripts.
    4. Find and select Clara Bootstrap.
    5. Go to the File Browser section and download the latest version of the bootstrap.zip file.
    6. Unzip the bootstrap.zip file:
    # Unzip
unzip bootstrap.zip -d bootstrap

  2. Skip this step if you are performing a new install of Clara Deploy SDK. Before upgrading a previous install of Clara Deploy SDK, run the uninstall_prereq.sh script:

    # Run the script
sudo ./uninstall_prereq.sh
    1. Use the following command to check that all Clara pods have been terminated:
    kubectl get pods

    Important

    We do not recommend manually deleting Clara Deploy components; the uninstall_prereq.sh script will delete all necessary binaries and stop all Clara Deploy containers.

  3. Install all prerequisites required to run Clara Deploy SDK using the bootstrap.sh script:

    # Run the script
sudo ./bootstrap.sh

    This will install the following required prerequisites:

    • Docker
    • NVIDIA Container Runtime for Docker
    • Kubernetes (including kubectl)
    • Helm

    Note

    If your system does not have NVIDIA CUDA Toolkit installed, you will be provided with a link to install it.

  4. Install Clara CLI.

    1. Log in to NGC.
    2. Select the appropriate org and team.
    3. Navigate to Model Scripts.
    4. Find and select Clara CLI.
    5. Go to the File Browser section and download the latest version of the cli.zip file.
    6. Extract the binaries into /usr/bin/ using following command. You should see output similar to that shown below:
    sudo unzip cli.zip -d /usr/bin/ && sudo chmod 755 /usr/bin/clara*
Archive:  cli.zip
inflating: /usr/bin/clara
inflating: /usr/bin/clara-dicom
inflating: /usr/bin/clara-monitor
inflating: /usr/bin/clara-platform
inflating: /usr/bin/clara-pull
inflating: /usr/bin/clara-render
    1. Verify that clara CLI has been successfully installed:
    clara version
    1. Generate clara completion script for Bash
    sudo bash -c 'clara completion bash > /etc/bash_completion.d/clara-completion.bash' && exec bash

  5. Configure Clara CLI to use your NGC credentials:

    If –orgteam is not specified, the orgteam defaults to ‘nvidia/clara’ which points to the publicly available Clara NGC artifacts.

    clara config --key NGC_API_KEY [--orgteam NGC_ORG/NGC_TEAM] [-y|--yes]

    Note

    To learn more about NGC API Key go to https://docs.nvidia.com/ngc/ngc-getting-started-guide/index.html#generating-api-key

  6. Pull and deploy Clara Platform with the following steps:

    1. Pull the latest Clara Platform helm charts:
    clara pull platform
    1. Start the Clara Deploy SDK:
    clara platform start

  7. Pull and deploy Clara Deploy services.

    1. Pull the latest Clara Deploy services helm charts:
    clara pull dicom
clara pull render
clara pull monitor
clara pull console
    1. Start the Clara Deploy SDK:
    clara dicom start
clara render start
clara monitor start
clara console start

Note

The console is not included in the current version of Clara Deploy, so the “console” commands will safely fail.

2.3. Verify Installation

The deploy script will automatically start the following components:

  • The Clara Platform
  • The DICOM Adapter
  • The Render Server
  • The Monitor Server

To verify that the installation is successful, run the following command:

helm ls

The following three helm charts should be returned:

  • clara
  • clara-dicom-adapter
  • clara-monitor-server
  • clara-render-server
  • clara-console

To verify that the helm charts are up and running, run the following command:

kubectl get pods

The command should return pods with the below prefixes:

  • clara-clara-platformapiserver-
  • clara-dicom-adapter-
  • clara-monitor-server-fluentd-elasticsearch-
  • clara-monitor-server-grafana-
  • clara-monitor-server-monitor-server-
  • clara-render-server-clara-renderer-
  • clara-resultsservice-
  • clara-ui-
  • clara-console-
  • clara-console-mongodb-
  • clara-workflow-controller-
  • elasticsearch-master-0
  • elasticsearch-master-1

They should be all in a Running state

2.4. Ansible Installation

Clara can also be installed with a set of Ansible playbooks. This has the added benefit of being able to be deployed across multiple hosts at one time.

2.4.1. Steps to Install

Note

This section assumes you will be using Ansible to install Clara on a remote machine.

  1. Download Ansible.

  2. Download the ansible.zip file from NGC.

    1. Log in to NGC,
    2. Select the appropriate org and team.
    3. Navigate to Resources.
    4. Find and select Clara Deploy Ansible Installation.
    5. Go to the File Browser section and download the latest version of the ansible.zip file.
    6. Unzip the ansible.zip file:
    ```bash
# Unzip
ansible.zip -d ansible
```

  3. Configure your Ansible hosts details in the file playbooks/clara_hosts. See Ansible’s inventory documentation for more information on how to work with inventory files.

    Note

    You may choose to update /etc/ansible/hosts instead of updating this file and can leave off the -i flag from the commands below.

  4. Install Clara Prerequisites To install Clara prerequisites (Docker, NVIDIA Driver, NVIDIA Docker, Openshify Python Libraries, Kubernetes and Helm) run:

    ansible-playbook -K install-clara-prereqs.yml -i clara_hosts

  5. Installing Clara Components To install basic Clara components run:

    ansible-playbook -K install-clara.yml -i clara_hosts

    By default only platform and dicom are installed, however, the user may override the default variable clara_components as:

    ansible-playbook -K install-clara.yml -i clara_hosts -e 'clara_components=["platform", "dicom", "render"]'

    or by updating the default value of clara_components in ansible/playbooks/roles/clara-components/default/main.yml.

2.4.2. Setting Up Hosts/Groups for installing Clara locally

To run the playbooks locally, within the playbooks/clara_hosts, instead of passing the IP address of a remote system, add localhost ansible_connection=local. This will allow you to run the playbook locally. You will also need to pass the flag -ask-become-pass with all ansible-playbook commands to give Ansible the needed sudo permissions to install packages. For example: ansible-playbook -K install-clara.yml --ask-become-pass

Note

The playbook`install-nvidia-driver.yml` forces a reboot after installation to ensure the driver is set up correctly. If installing locally, after the system reboot, continue the installation with the next playbook install-clara-prereqs.yml and install-clara.yml

2.5. Installation on a Cloud Service Provider (CSP)

Researchers and data scientists who don’t have access to a GPU server can still get started with Clara Deploy SDK without needing to become a Docker and Kubernetes expert. Clara Deploy SDK has been validated on the following services:

  • Amazon Web Services (AWS)
  • Microsoft Azure Cloud Services (Azure)
  • Google Cloud Platform Services (GCP)

The following subsections describe the configuration for each CSP. Once the VM is provisioned according to the documentation, you can follow the Steps to Install section above to install Clara Deploy SDK.

2.5.1. AWS Virtual Machine Configuration

The AWS VM configuration used for testing can be found below:

  • Location : US East (Ohio)
  • Operating System : Ubuntu 18.04
  • Amazon machine image : Ubuntu Server 18.04 LTS (HVM), SSD Volume Type (64-bit)
  • Instance type : p3.8xlarge (32 vcpus, 244 GB memory, 4 NVIDIA GPUs in the Pascal, Volta and Turing families)
  • Storage: General Purpose SSD (100 GB)
  • Ports Open : SSH, HTTP, HTTPS

2.5.2. Azure Virtual Machine Configuration

The Azure VM configuration used for testing can be found below:

  • Location : West US2
  • Operating System: Ubuntu 18.04
  • Size : Standard NC6s_v2 (6 vcpus, 112 GB memory, 1 GPU-NVIDIA Tesla P100)
  • OS Disk Size : Premium SSD, 300GB (mounted on root)
  • Ports Open : SSH, HTTP, HTTPS

2.5.3. GCP Virtual Machine Configuration

The GCP VM configuration used for testing can be found below:

  • Location :
    • Region: us-central1 (Iowa)
    • Zone: us-central1-c
  • Operating System : Ubuntu 18.04 LTS
  • Machine type: 8vCPU, 32GB, 1 GPU (NVIDIA Tesla P4), Turn on display device
  • Disk Size: SSD 100GB
  • Ports Open : SSH, HTTP, HTTPS