mlflow-docker-azure

Install Dependencies

Requires:

• Docker
• Docker Compose
• htpasswd
• Azure CLI (optional: for creating storage account)

Install docker and docker-compose through the links above.

htpasswd can be installed by:

sudo apt install apache2-utils

Configure

Populate the variables at the start of the configure-env.sh file:

### Set these variables to be written to .bashrc/stdout

# MYSQL
MYSQL_USER=""
MYSQL_PASSWORD=""

# MLFLOW
MLFLOW_TRACKING_USERNAME=""
MLFLOW_TRACKING_HOSTNAME=""

# AZURE
AZURE_STORAGE_ACCESS_KEY=""
AZURE_STORAGE_ACCOUNT=""
AZURE_STORAGE_CONTAINER=""
AZURE_RESOURCE_GROUP=""
AZURE_RESOURCE_GROUP_LOCATION=""

Run script appropriately to output to environment. Use the following to display usage:

./configure-env.sh help

Source environment file as appropriate for shell:

source ~/.bashrc

Optional: Create Azure Storage Account

If a storage account has not yet been created, the create-azure-sa.sh file can create the resource group, storage account, and storage container through the environment variables provided through configure-env.sh.

The following environment variables must be set for storage account creation:

AZURE_STORAGE_ACCOUNT=""
AZURE_STORAGE_CONTAINER=""
AZURE_RESOURCE_GROUP=""
AZURE_RESOURCE_GROUP_LOCATION=""

Run the following command:

./create-azure-sa.sh create_sa

Deploy

Start Mlflow and relevant images:

docker-compose up --build

Access

The value of MLFLOW_TRACKING_URI is valid to access the Mlflow dashboard - simply add https:// to the provided value of MLFLOW_TRACKING_HOSTNAME.