The ecsspawner (also known as JupyterHub ECS/AWS Spawner) enables JupyterHub to spawn single-user notebook servers on a ECS cluster in AWS.
Amazon ECS (Elastic Container Service) is a highly scalable, high-performance container orchestration service which allows us to easily run and scale containerized applications on AWS.
Jupyter Notebooks are instantiated using ECS Tasks. A Task is the mechanism offered by Amazon ECS to encapsulate containers allowing us to specificy its charateristics using Task definitions.
A task definition includes:
- The Docker image to use
- CPU and memory requirements
- AWS Network, logging, volumens and IAM roles
- Launch types: EC2 or Fargate. Read more about
When using EC2 launch type, we have some alternatives for starting ECS tasks:
- Let Amazon ECS to run the Task based on the task definition using the current cluster capacity.
- Create a EC2 instance on the cluster, and start the ECS task on the created EC2 instance.
- Combined: Let amazon to run the task (1) and as a fallback scenario use (2) when no cluster capacity is available.
## Spawner Configuration
c.JupyterHub.spawner_class = 'awsspawner.EcsTaskSpawner'
strategy: Defines the spawning mechanism to be used. Possible values are
- ECSSpawnerHandler: Let Amazon ECS to run the Task based on the task definition
- ECSxEC2SpawnerHandler: Create a EC2 instance on the cluster, and start the ECS task on the created EC2 instance
cluster_name: ECS cluster name.
ecs_task_definition: When using ECS spawning mechanim, a task definition string is mandatory. The task definition must point to a definition where a Jupyter notebook is specified.
ec2_instance_template: When using EC2/ECSxEC2 mechanism, an AWS ECS instance template is required to determine EC2 instantiation.
c.Spawner.strategy = 'ECSxEC2SpawnerHandler'
c.Spawner.strategy_parms = {
'cluster_name': 'notebook-cluster',
'ec2_instance_template': 'ec2-demo-template',
'ecs_task_definition': 'jupyter-notebook-template',
'port': 8888
}