Azure VM scale sets can be configured to autoscale by a variety of metrics but in some scenarios a custom metric is needed. For example, an Azure VMSS might be running a set of workers consuming jobs from several ServiceBus subscriptions and should scale according to the total number of messages in all subscriptions. This repo includes python reference code for a custom autoscaling service for Azure VM scale sets. The code can be run as a time triggered Azure Function or can be run as a service running in a Docker container.
Although this code implements custom scaling according to a specific scenario (total number of messages in a number of subscriptions), it can be used as a reference for other similar scenarios.
The scenario addressed in this repo is described by the following diagram:
All of the scaler configurations can be found in config.json. Values should be changed to fit the VMSS to be scaled and ServiceBus
to be polled.
Also, the service uses an Azure service prinicipal that should be created and given a Contributor role to be able to scale the VMSS (see this article). The service uses the following environment variables for Azure authentication:
- AZURE_CLIENT_ID
- AZURE_CLIENT_SECRET
- AZURE_TENANT_ID
- SUBSCRIPTION_ID
The values for the above variables are provided by the service principal.
Additional environment variables can be defined to override config.json values:
- LOOP
- INTERVAL_IN_SEC
- SCALING_MIN_CAPACITY
- SCALING_MAX_CAPACITY
- SCALING_LOW_THRESHOLD
- SCALING_HIGH_THRESHOLD
- SCALING_UP_FACTOR
- SCALING_DOWN_FACTOR
A Docker image of this application can be created using the included Dockerfile. To define the credentials inside the container an environment variables file can be used when running the container.
Example docker run command:
docker run -d --name scaler --env-file ./env.list orizohar/autoscalerExample of env.list:
AZURE_CLIENT_ID=XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
AZURE_CLIENT_SECRET=1111111111111111111111111111111111111111111=
AZURE_TENANT_ID=YYYYYYYY-YYYY-YYYY-YYYY-YYYYYYYYYYYY
SUBSCRIPTION_ID=ZZZZZZZZ-ZZZZ-ZZZZ-ZZZZ-ZZZZZZZZZZZZ
SCALING_MIN_CAPACITY=4
SCALING_MAX_CAPACITY=20
The autoscaling service can be also be deployed using Azure Functions. This repository is structured in a way which allows continuous deployment with local git or GitHub deployment sources (for more info, please see this article). Note that function.json defines a function with a time trigger.
As mentioned above, the service expects credential information to be available via environment variables. Environment variables can be configured by going to Function app settings --> Configure app settings from the Function blade and configuring the variables like so:
The repo also includes an example message client which can be used for sending and receiving messages to/from ServiceBus for testing. See usage information by running:
python msg_client.py --help