The interesting part here is actually the parameter executor_config. All using the PythonOperator to execute python callable functions, either print_stuff or use_vim_binary. You have three tasks here (we will see the last one later). You may ask, what actually does this code? Well it’s fairly easy. By the way, if you want learn more about Airflow and have a special promotion, just click right here. KubernetesExecutor, which is quite new and allows you to run your tasks using Kubernetes and so makes your Airflow cluster elastic to your workload in order to avoid wasting your precious ressources.It’s up to you to choose either Dask or Celery according to the framework fitting the most your needs. DaskExecutor, Dask in another Python distributed task processing system like Celery.It is the recommended way to go in production since you will be able to absorb the workload you need. All the distribution is managed by Celery. You basically run your tasks on multiple nodes (airflow workers) and each task is first queued through the use of a RabbitMQ for example. CeleryExecutor allows you to horizontally scale your Airflow cluster.Scale quite well (vertical scaling) and can be used in production.

LocalExecutor which runs multiple subprocesses to execute your tasks concurrently on the same host.Recommended for debugging and testing only. SequentialExecutor which is the most simple one to execute your tasks in a sequential manner.Then, according to the executor used, the execution of the task will differ.Īpache Airflow gives you 5 type of executors: Before getting executed, a task is always scheduled first and pushed into a queue implemented as an OrderedDict in order to keep them sorted by their addition order. A task corresponds to a node in your DAG where an action must be done such as, executing a bash shell command, a python script, kick off a spark job and so on. Now your memories about Kubenertes are fresh let’s move on Airflow executors.īasically, an executor defines how your tasks are going to be executed. In the context of Airflow and Kubernetes Executors, you can think of Kubernetes as a pool of ressources giving a simple but powerful API to dynamically launch complex deployments. Basically, in the most common Kubertenes use case (and in the case of Airflow), a Pod will run a single Docker container corresponding to a component of your application. A Pod is the smallest deployable object in Kubernetes.It encapsulates an application’s container (or multiple containers) as well as storage resources (shared volumes), a unique network IP, and options to set how the container(s) should run. In very simple terms, Kubernetes orchestrates the different containers composing your application so that they can work smoothly together.Ī very important concept to understand in Kubertenes is the concept of Pod. It orchestrates computing, networking and storage infrastructure and offers very nice features such as deployment, scaling, load balancing, logging and monitoring. So what is Kubernetes? Kubernetes is an open-source platform for managing containerized applications. Before starting to dive into the Kubernetes Executor, let me first give you a quick reminder about Kubernetes.

Airflow docker image for kubernetes