AWS Autoscaler for Apache Airflow Mesos Provider

This AWS autoscaler will start EC2 instances if Airflow does not get matched offers from mesos.

Issues

To open an issue, please use this place: https://github.com/m3scluster/mesos-airflow-autoscaler-aws/issues

Requirements

• Airflow min 2.1.x
• Apache Mesos min 1.6.x
• Airflow Mesos Provider min 2.0
• AWS

Configuration

ENV	Default	Description
AIRFLOW_MESOS_SCHEDULER	127.0.0.1:11000	IP Address and port of the Apache Airflow Mesos provider
LOGLEVEL	debug	Loglevel (info, warn, debug)
WAIT_TIME	2m	The time in minutes the autoscaler have to wait until it will create a mesos instance in AWS
WAIT_TIME_OVERWRITE_INSTANCE	15m	If the DAG is still after 15m in the queue, the autoscaler will ignore the custom instance type to start a new ec2 instance.
REDIS_SERVER	127.0.0.1:6480	Redis server and port
REDIS_PASSWORD		Redis DB password
REDIS_DB	2	Redis DB Number
REDIS_PREFIX	asg	Prefix for every Redis key
API_USERNAME	user	Username to authenticate against the Apache Airflow Mesos provider
API_PASSWORD	password	Password to authenticate against the Apache Airflow Mesos provider
AWS_SECRET		AWS Secret
AWS_REGION	eu-central-1	AWS Region
AWS_WAIT_TIME	10m	The time the autoscaler have to wait until it check if the EC2 intance can be terminated.
AWS_LAUNCH_TEMPLATE_ID		The AWS Launche Template ID
AWS_INSTANCE_FALLBACK	t3a.2xlarge	Fallback instance type will be used if there are no more ec2 resources in AWS.
AWS_INSTANCE_DEFAULT_ARCHITECTURE	x86_64	Default architecture of ec2 instance.
AWS_INSTANCE_ALLOW	"t3.,r5.,c5."	Only these instances are allowed.
AWS_INSTANCE_MAX_AGE	6h	Max age on an instance. That should prevent instances to run endless.
MESOS_AGENT_SSL	false	Enable SSL for the communication to the Mesos agent
MESOS_AGENT_USERNAME		Username of the Mesos Agent
MESOS_AGENT_PASSWORD		Password of the Mesos Agent
MESOS_AGENT_PORT	5051	Port of the Mesos Agent
MESOS_AGENT_TIMEOUT	10m	Mesos agent timeout
SSL	false	Enable SSL for the communication to the Airflow Scheduler API
DAG_TTL	6h	Set the TTL for DAG'keys in Redis.

Airflow Executor Config

These autoscaler support the following executor config parameters:

  executor_config={
    "mem_limit": 2048,
    "instance_type": "t2.xlarge"
 }

• With the help if the mem_limit parameter, the autoscaler will determin which AWS instance is the right one.
• instance_type will overwrite the mem_limit and just crate a instance of that type.

Documentation

AWS - https://docs.aws.amazon.com/code-samples/latest/catalog/go-ec2-create_instance.go.html