Use the KubernetesPodOperator

Cloud Composer 1 | Cloud Composer 2

This page describes how to use the KubernetesPodOperator to launch Kubernetes pods from Cloud Composer into the Google Kubernetes Engine cluster that is part of your Cloud Composer environment and to ensure your environment has the appropriate resources.

KubernetesPodOperator launches Kubernetes pods in your environment's cluster. In comparison, Google Kubernetes Engine operators run Kubernetes pods in a specified cluster, which can be a separate cluster that is not related to your environment. You can also create and delete clusters using Google Kubernetes Engine operators.

The KubernetesPodOperator is a good option if you require:

  • Custom Python dependencies that are not available through the public PyPI repository.
  • Binary dependencies that are not available in the stock Cloud Composer worker image.

This page walks you through an example DAG that includes the following KubernetesPodOperator configurations:

Before you begin

Ensuring that your environment has sufficient resources

When you create a Cloud Composer environment, you specify its performance parameters, including performance parameters for environment's cluster. Launching Kubernetes pods into the environment cluster can cause competition for cluster resources, such as CPU or memory. Because the Airflow scheduler and workers are in the same GKE cluster, the schedulers and workers won't work properly if the competition results in resource starvation.

To prevent resource starvation, take one or more of the following actions:

Creating a node pool

The preferred way to prevent resource starvation in the Cloud Composer environment is to create a new node pool and configure the Kubernetes pods to execute using only resources from that pool.

Console

  1. In the Google Cloud Console, go to the Environments page.

    Go to Environments

  2. Click the name of your environment.

  3. On the Environment details page, go to Environment configuration tab.

  4. In the Resources > GKE cluster section, follow the view cluster details link.

  5. Create a node pool as described in Adding a node pool.

gcloud

  1. Determine the name of your environment's cluster:

    gcloud composer environments describe ENVIRONMENT_NAME \
      --location LOCATION \
      --format="value(config.gkeCluster)"
    

    Replace:

    • ENVIRONMENT_NAME with the name of the environment.
    • LOCATION with the Compute Engine region where the environment is located.
  2. The output contains the name of your environment's cluster. For example, this can be europe-west3-example-enviro-af810e25-gke.

  3. Create a node pool as described in Adding a node pool.

Increasing the number of nodes in your environment

Increasing the number of nodes in your Cloud Composer environment increases the computing power available to your workloads. This increase does not provide additional resources for tasks that require more CPU or RAM than the specified machine type provides.

To increase node count, update your environment.

Specifying the appropriate machine type

During Cloud Composer environment creation, you can specify a machine type. To ensure available resources, specify a machine type for the type of computing that occurs in your Cloud Composer environment.

KubernetesPodOperator configuration

To follow along with this example, put the entire kubernetes_pod_operator.py file in your environment's dags/ folder or add the relevant KubernetesPodOperator code to a DAG.

The following sections explain each KubernetesPodOperator configuration in the example. For information about each configuration variable, see the Airflow reference.

Airflow 2

import datetime

from airflow import models
from airflow.kubernetes.secret import Secret
from airflow.providers.cncf.kubernetes.operators.kubernetes_pod import KubernetesPodOperator


# A Secret is an object that contains a small amount of sensitive data such as
# a password, a token, or a key. Such information might otherwise be put in a
# Pod specification or in an image; putting it in a Secret object allows for
# more control over how it is used, and reduces the risk of accidental
# exposure.

secret_env = Secret(
    # Expose the secret as environment variable.
    deploy_type='env',
    # The name of the environment variable, since deploy_type is `env` rather
    # than `volume`.
    deploy_target='SQL_CONN',
    # Name of the Kubernetes Secret
    secret='airflow-secrets',
    # Key of a secret stored in this Secret object
    key='sql_alchemy_conn')
secret_volume = Secret(
    deploy_type='volume',
    # Path where we mount the secret as volume
    deploy_target='/var/secrets/google',
    # Name of Kubernetes Secret
    secret='service-account',
    # Key in the form of service account file name
    key='service-account.json')

# If you are running Airflow in more than one time zone
# see https://airflow.apache.org/docs/apache-airflow/stable/timezone.html
# for best practices
YESTERDAY = datetime.datetime.now() - datetime.timedelta(days=1)

# If a Pod fails to launch, or has an error occur in the container, Airflow
# will show the task as failed, as well as contain all of the task logs
# required to debug.
with models.DAG(
        dag_id='composer_sample_kubernetes_pod',
        schedule_interval=datetime.timedelta(days=1),
        start_date=YESTERDAY) as dag:
    # Only name, namespace, image, and task_id are required to create a
    # KubernetesPodOperator. In Cloud Composer, currently the operator defaults
    # to using the config file found at `/home/airflow/composer_kube_config if
    # no `config_file` parameter is specified. By default it will contain the
    # credentials for Cloud Composer's Google Kubernetes Engine cluster that is
    # created upon environment creation.

    kubernetes_min_pod = KubernetesPodOperator(
        # The ID specified for the task.
        task_id='pod-ex-minimum',
        # Name of task you want to run, used to generate Pod ID.
        name='pod-ex-minimum',
        # Entrypoint of the container, if not specified the Docker container's
        # entrypoint is used. The cmds parameter is templated.
        cmds=['echo'],
        # The namespace to run within Kubernetes, default namespace is
        # `default`. There is the potential for the resource starvation of
        # Airflow workers and scheduler within the Cloud Composer environment,
        # the recommended solution is to increase the amount of nodes in order
        # to satisfy the computing requirements. Alternatively, launching pods
        # into a custom namespace will stop fighting over resources.
        namespace='default',
        # Docker image specified. Defaults to hub.docker.com, but any fully
        # qualified URLs will point to a custom repository. Supports private
        # gcr.io images if the Composer Environment is under the same
        # project-id as the gcr.io images and the service account that Composer
        # uses has permission to access the Google Container Registry
        # (the default service account has permission)
        image='gcr.io/gcp-runtimes/ubuntu_18_0_4')
    kubenetes_template_ex = KubernetesPodOperator(
        task_id='ex-kube-templates',
        name='ex-kube-templates',
        namespace='default',
        image='bash',
        # All parameters below are able to be templated with jinja -- cmds,
        # arguments, env_vars, and config_file. For more information visit:
        # https://airflow.apache.org/docs/apache-airflow/stable/macros-ref.html

        # Entrypoint of the container, if not specified the Docker container's
        # entrypoint is used. The cmds parameter is templated.
        cmds=['echo'],
        # DS in jinja is the execution date as YYYY-MM-DD, this docker image
        # will echo the execution date. Arguments to the entrypoint. The docker
        # image's CMD is used if this is not provided. The arguments parameter
        # is templated.
        arguments=['{{ ds }}'],
        # The var template variable allows you to access variables defined in
        # Airflow UI. In this case we are getting the value of my_value and
        # setting the environment variable `MY_VALUE`. The pod will fail if
        # `my_value` is not set in the Airflow UI.
        env_vars={'MY_VALUE': '{{ var.value.my_value }}'},
        # Sets the config file to a kubernetes config file specified in
        # airflow.cfg. If the configuration file does not exist or does
        # not provide validcredentials the pod will fail to launch. If not
        # specified, config_file defaults to ~/.kube/config
        config_file="{{ conf.get('core', 'kube_config') }}")
    kubernetes_secret_vars_ex = KubernetesPodOperator(
        task_id='ex-kube-secrets',
        name='ex-kube-secrets',
        namespace='default',
        image='ubuntu',
        startup_timeout_seconds=300,
        # The secrets to pass to Pod, the Pod will fail to create if the
        # secrets you specify in a Secret object do not exist in Kubernetes.
        secrets=[secret_env, secret_volume],
        # env_vars allows you to specify environment variables for your
        # container to use. env_vars is templated.
        env_vars={
            'EXAMPLE_VAR': '/example/value',
            'GOOGLE_APPLICATION_CREDENTIALS': '/var/secrets/google/service-account.json '})
    # Pod affinity with the KubernetesPodOperator
    # is not supported with Composer 2
    # instead, create a cluster and use the GKEStartPodOperator
    # https://cloud.google.com/composer/docs/using-gke-operator
    kubernetes_affinity_ex = KubernetesPodOperator(
        task_id='ex-pod-affinity',
        name='ex-pod-affinity',
        namespace='default',
        image='perl',
        cmds=['perl'],
        arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
        # affinity allows you to constrain which nodes your pod is eligible to
        # be scheduled on, based on labels on the node. In this case, if the
        # label 'cloud.google.com/gke-nodepool' with value
        # 'nodepool-label-value' or 'nodepool-label-value2' is not found on any
        # nodes, it will fail to schedule.
        affinity={
            'nodeAffinity': {
                # requiredDuringSchedulingIgnoredDuringExecution means in order
                # for a pod to be scheduled on a node, the node must have the
                # specified labels. However, if labels on a node change at
                # runtime such that the affinity rules on a pod are no longer
                # met, the pod will still continue to run on the node.
                'requiredDuringSchedulingIgnoredDuringExecution': {
                    'nodeSelectorTerms': [{
                        'matchExpressions': [{
                            # When nodepools are created in Google Kubernetes
                            # Engine, the nodes inside of that nodepool are
                            # automatically assigned the label
                            # 'cloud.google.com/gke-nodepool' with the value of
                            # the nodepool's name.
                            'key': 'cloud.google.com/gke-nodepool',
                            'operator': 'In',
                            # The label key's value that pods can be scheduled
                            # on.
                            'values': [
                                'pool-0',
                                'pool-1',
                            ]
                        }]
                    }]
                }
            }
        })
    kubernetes_full_pod = KubernetesPodOperator(
        task_id='ex-all-configs',
        name='pi',
        namespace='default',
        image='perl',
        # Entrypoint of the container, if not specified the Docker container's
        # entrypoint is used. The cmds parameter is templated.
        cmds=['perl'],
        # Arguments to the entrypoint. The docker image's CMD is used if this
        # is not provided. The arguments parameter is templated.
        arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
        # The secrets to pass to Pod, the Pod will fail to create if the
        # secrets you specify in a Secret object do not exist in Kubernetes.
        secrets=[],
        # Labels to apply to the Pod.
        labels={'pod-label': 'label-name'},
        # Timeout to start up the Pod, default is 120.
        startup_timeout_seconds=120,
        # The environment variables to be initialized in the container
        # env_vars are templated.
        env_vars={'EXAMPLE_VAR': '/example/value'},
        # If true, logs stdout output of container. Defaults to True.
        get_logs=True,
        # Determines when to pull a fresh image, if 'IfNotPresent' will cause
        # the Kubelet to skip pulling an image if it already exists. If you
        # want to always pull a new image, set it to 'Always'.
        image_pull_policy='Always',
        # Annotations are non-identifying metadata you can attach to the Pod.
        # Can be a large range of data, and can include characters that are not
        # permitted by labels.
        annotations={'key1': 'value1'},
        # Resource specifications for Pod, this will allow you to set both cpu
        # and memory limits and requirements.
        # Prior to Airflow 1.10.4, resource specifications were
        # passed as a Pod Resources Class object,
        # If using this example on a version of Airflow prior to 1.10.4,
        # import the "pod" package from airflow.contrib.kubernetes and use
        # resources = pod.Resources() instead passing a dict
        # For more info see:
        # https://github.com/apache/airflow/pull/4551
        resources={'limit_memory': "250M", 'limit_cpu': "100m"},
        # Specifies path to kubernetes config. If no config is specified will
        # default to '~/.kube/config'. The config_file is templated.
        config_file='/home/airflow/composer_kube_config',
        # If true, the content of /airflow/xcom/return.json from container will
        # also be pushed to an XCom when the container ends.
        do_xcom_push=False,
        # List of Volume objects to pass to the Pod.
        volumes=[],
        # List of VolumeMount objects to pass to the Pod.
        volume_mounts=[],
        # Affinity determines which nodes the Pod can run on based on the
        # config. For more information see:
        # https://kubernetes.io/docs/concepts/configuration/assign-pod-node/
        # Pod affinity with the KubernetesPodOperator
        # is not supported with Composer 2
        # instead, create a cluster and use the GKEStartPodOperator
        # https://cloud.google.com/composer/docs/using-gke-operator
        affinity={})

Airflow 1

import datetime

from airflow import models
from airflow.contrib.kubernetes import secret
from airflow.contrib.operators import kubernetes_pod_operator


# A Secret is an object that contains a small amount of sensitive data such as
# a password, a token, or a key. Such information might otherwise be put in a
# Pod specification or in an image; putting it in a Secret object allows for
# more control over how it is used, and reduces the risk of accidental
# exposure.

secret_env = secret.Secret(
    # Expose the secret as environment variable.
    deploy_type='env',
    # The name of the environment variable, since deploy_type is `env` rather
    # than `volume`.
    deploy_target='SQL_CONN',
    # Name of the Kubernetes Secret
    secret='airflow-secrets',
    # Key of a secret stored in this Secret object
    key='sql_alchemy_conn')
secret_volume = secret.Secret(
    deploy_type='volume',
    # Path where we mount the secret as volume
    deploy_target='/var/secrets/google',
    # Name of Kubernetes Secret
    secret='service-account',
    # Key in the form of service account file name
    key='service-account.json')

# If you are running Airflow in more than one time zone
# see https://airflow.apache.org/docs/apache-airflow/stable/timezone.html
# for best practices
YESTERDAY = datetime.datetime.now() - datetime.timedelta(days=1)

# If a Pod fails to launch, or has an error occur in the container, Airflow
# will show the task as failed, as well as contain all of the task logs
# required to debug.
with models.DAG(
        dag_id='composer_sample_kubernetes_pod',
        schedule_interval=datetime.timedelta(days=1),
        start_date=YESTERDAY) as dag:
    # Only name, namespace, image, and task_id are required to create a
    # KubernetesPodOperator. In Cloud Composer, currently the operator defaults
    # to using the config file found at `/home/airflow/composer_kube_config if
    # no `config_file` parameter is specified. By default it will contain the
    # credentials for Cloud Composer's Google Kubernetes Engine cluster that is
    # created upon environment creation.

    kubernetes_min_pod = kubernetes_pod_operator.KubernetesPodOperator(
        # The ID specified for the task.
        task_id='pod-ex-minimum',
        # Name of task you want to run, used to generate Pod ID.
        name='pod-ex-minimum',
        # Entrypoint of the container, if not specified the Docker container's
        # entrypoint is used. The cmds parameter is templated.
        cmds=['echo'],
        # The namespace to run within Kubernetes, default namespace is
        # `default`. There is the potential for the resource starvation of
        # Airflow workers and scheduler within the Cloud Composer environment,
        # the recommended solution is to increase the amount of nodes in order
        # to satisfy the computing requirements. Alternatively, launching pods
        # into a custom namespace will stop fighting over resources.
        namespace='default',
        # Docker image specified. Defaults to hub.docker.com, but any fully
        # qualified URLs will point to a custom repository. Supports private
        # gcr.io images if the Composer Environment is under the same
        # project-id as the gcr.io images and the service account that Composer
        # uses has permission to access the Google Container Registry
        # (the default service account has permission)
        image='gcr.io/gcp-runtimes/ubuntu_18_0_4')
    kubenetes_template_ex = kubernetes_pod_operator.KubernetesPodOperator(
        task_id='ex-kube-templates',
        name='ex-kube-templates',
        namespace='default',
        image='bash',
        # All parameters below are able to be templated with jinja -- cmds,
        # arguments, env_vars, and config_file. For more information visit:
        # https://airflow.apache.org/docs/apache-airflow/stable/macros-ref.html

        # Entrypoint of the container, if not specified the Docker container's
        # entrypoint is used. The cmds parameter is templated.
        cmds=['echo'],
        # DS in jinja is the execution date as YYYY-MM-DD, this docker image
        # will echo the execution date. Arguments to the entrypoint. The docker
        # image's CMD is used if this is not provided. The arguments parameter
        # is templated.
        arguments=['{{ ds }}'],
        # The var template variable allows you to access variables defined in
        # Airflow UI. In this case we are getting the value of my_value and
        # setting the environment variable `MY_VALUE`. The pod will fail if
        # `my_value` is not set in the Airflow UI.
        env_vars={'MY_VALUE': '{{ var.value.my_value }}'},
        # Sets the config file to a kubernetes config file specified in
        # airflow.cfg. If the configuration file does not exist or does
        # not provide validcredentials the pod will fail to launch. If not
        # specified, config_file defaults to ~/.kube/config
        config_file="{{ conf.get('core', 'kube_config') }}")
    kubernetes_secret_vars_ex = kubernetes_pod_operator.KubernetesPodOperator(
        task_id='ex-kube-secrets',
        name='ex-kube-secrets',
        namespace='default',
        image='ubuntu',
        startup_timeout_seconds=300,
        # The secrets to pass to Pod, the Pod will fail to create if the
        # secrets you specify in a Secret object do not exist in Kubernetes.
        secrets=[secret_env, secret_volume],
        # env_vars allows you to specify environment variables for your
        # container to use. env_vars is templated.
        env_vars={
            'EXAMPLE_VAR': '/example/value',
            'GOOGLE_APPLICATION_CREDENTIALS': '/var/secrets/google/service-account.json '})
    kubernetes_affinity_ex = kubernetes_pod_operator.KubernetesPodOperator(
        task_id='ex-pod-affinity',
        name='ex-pod-affinity',
        namespace='default',
        image='perl',
        cmds=['perl'],
        arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
        # affinity allows you to constrain which nodes your pod is eligible to
        # be scheduled on, based on labels on the node. In this case, if the
        # label 'cloud.google.com/gke-nodepool' with value
        # 'nodepool-label-value' or 'nodepool-label-value2' is not found on any
        # nodes, it will fail to schedule.
        affinity={
            'nodeAffinity': {
                # requiredDuringSchedulingIgnoredDuringExecution means in order
                # for a pod to be scheduled on a node, the node must have the
                # specified labels. However, if labels on a node change at
                # runtime such that the affinity rules on a pod are no longer
                # met, the pod will still continue to run on the node.
                'requiredDuringSchedulingIgnoredDuringExecution': {
                    'nodeSelectorTerms': [{
                        'matchExpressions': [{
                            # When nodepools are created in Google Kubernetes
                            # Engine, the nodes inside of that nodepool are
                            # automatically assigned the label
                            # 'cloud.google.com/gke-nodepool' with the value of
                            # the nodepool's name.
                            'key': 'cloud.google.com/gke-nodepool',
                            'operator': 'In',
                            # The label key's value that pods can be scheduled
                            # on.
                            'values': [
                                'pool-0',
                                'pool-1',
                            ]
                        }]
                    }]
                }
            }
        })
    kubernetes_full_pod = kubernetes_pod_operator.KubernetesPodOperator(
        task_id='ex-all-configs',
        name='pi',
        namespace='default',
        image='perl',
        # Entrypoint of the container, if not specified the Docker container's
        # entrypoint is used. The cmds parameter is templated.
        cmds=['perl'],
        # Arguments to the entrypoint. The docker image's CMD is used if this
        # is not provided. The arguments parameter is templated.
        arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
        # The secrets to pass to Pod, the Pod will fail to create if the
        # secrets you specify in a Secret object do not exist in Kubernetes.
        secrets=[],
        # Labels to apply to the Pod.
        labels={'pod-label': 'label-name'},
        # Timeout to start up the Pod, default is 120.
        startup_timeout_seconds=120,
        # The environment variables to be initialized in the container
        # env_vars are templated.
        env_vars={'EXAMPLE_VAR': '/example/value'},
        # If true, logs stdout output of container. Defaults to True.
        get_logs=True,
        # Determines when to pull a fresh image, if 'IfNotPresent' will cause
        # the Kubelet to skip pulling an image if it already exists. If you
        # want to always pull a new image, set it to 'Always'.
        image_pull_policy='Always',
        # Annotations are non-identifying metadata you can attach to the Pod.
        # Can be a large range of data, and can include characters that are not
        # permitted by labels.
        annotations={'key1': 'value1'},
        # Resource specifications for Pod, this will allow you to set both cpu
        # and memory limits and requirements.
        # Prior to Airflow 1.10.4, resource specifications were
        # passed as a Pod Resources Class object,
        # If using this example on a version of Airflow prior to 1.10.4,
        # import the "pod" package from airflow.contrib.kubernetes and use
        # resources = pod.Resources() instead passing a dict
        # For more info see:
        # https://github.com/apache/airflow/pull/4551
        resources={'limit_memory': "250M", 'limit_cpu': "100m"},
        # Specifies path to kubernetes config. If no config is specified will
        # default to '~/.kube/config'. The config_file is templated.
        config_file='/home/airflow/composer_kube_config',
        # If true, the content of /airflow/xcom/return.json from container will
        # also be pushed to an XCom when the container ends.
        do_xcom_push=False,
        # List of Volume objects to pass to the Pod.
        volumes=[],
        # List of VolumeMount objects to pass to the Pod.
        volume_mounts=[],
        # Affinity determines which nodes the Pod can run on based on the
        # config. For more information see:
        # https://kubernetes.io/docs/concepts/configuration/assign-pod-node/
        affinity={})

Minimal configuration

To create a KubernetesPodOperator, only name, namespace, image, and task_id are required.

When you place the following code snippet in a DAG, the configuration uses the defaults in /home/airflow/composer_kube_config. You don't need to modify the code for the pod-ex-minimum task to succeed.

Airflow 2

kubernetes_min_pod = KubernetesPodOperator(
    # The ID specified for the task.
    task_id='pod-ex-minimum',
    # Name of task you want to run, used to generate Pod ID.
    name='pod-ex-minimum',
    # Entrypoint of the container, if not specified the Docker container's
    # entrypoint is used. The cmds parameter is templated.
    cmds=['echo'],
    # The namespace to run within Kubernetes, default namespace is
    # `default`. There is the potential for the resource starvation of
    # Airflow workers and scheduler within the Cloud Composer environment,
    # the recommended solution is to increase the amount of nodes in order
    # to satisfy the computing requirements. Alternatively, launching pods
    # into a custom namespace will stop fighting over resources.
    namespace='default',
    # Docker image specified. Defaults to hub.docker.com, but any fully
    # qualified URLs will point to a custom repository. Supports private
    # gcr.io images if the Composer Environment is under the same
    # project-id as the gcr.io images and the service account that Composer
    # uses has permission to access the Google Container Registry
    # (the default service account has permission)
    image='gcr.io/gcp-runtimes/ubuntu_18_0_4')

Airflow 1

kubernetes_min_pod = kubernetes_pod_operator.KubernetesPodOperator(
    # The ID specified for the task.
    task_id='pod-ex-minimum',
    # Name of task you want to run, used to generate Pod ID.
    name='pod-ex-minimum',
    # Entrypoint of the container, if not specified the Docker container's
    # entrypoint is used. The cmds parameter is templated.
    cmds=['echo'],
    # The namespace to run within Kubernetes, default namespace is
    # `default`. There is the potential for the resource starvation of
    # Airflow workers and scheduler within the Cloud Composer environment,
    # the recommended solution is to increase the amount of nodes in order
    # to satisfy the computing requirements. Alternatively, launching pods
    # into a custom namespace will stop fighting over resources.
    namespace='default',
    # Docker image specified. Defaults to hub.docker.com, but any fully
    # qualified URLs will point to a custom repository. Supports private
    # gcr.io images if the Composer Environment is under the same
    # project-id as the gcr.io images and the service account that Composer
    # uses has permission to access the Google Container Registry
    # (the default service account has permission)
    image='gcr.io/gcp-runtimes/ubuntu_18_0_4')

Template configuration

Airflow supports using Jinja Templating. You must declare the required variables (task_id, name, namespace, and image) with the operator. As shown in the following example, you can template all other parameters with Jinja, including cmds, arguments, env_vars, and config_file.

Airflow 2

kubenetes_template_ex = KubernetesPodOperator(
    task_id='ex-kube-templates',
    name='ex-kube-templates',
    namespace='default',
    image='bash',
    # All parameters below are able to be templated with jinja -- cmds,
    # arguments, env_vars, and config_file. For more information visit:
    # https://airflow.apache.org/docs/apache-airflow/stable/macros-ref.html

    # Entrypoint of the container, if not specified the Docker container's
    # entrypoint is used. The cmds parameter is templated.
    cmds=['echo'],
    # DS in jinja is the execution date as YYYY-MM-DD, this docker image
    # will echo the execution date. Arguments to the entrypoint. The docker
    # image's CMD is used if this is not provided. The arguments parameter
    # is templated.
    arguments=['{{ ds }}'],
    # The var template variable allows you to access variables defined in
    # Airflow UI. In this case we are getting the value of my_value and
    # setting the environment variable `MY_VALUE`. The pod will fail if
    # `my_value` is not set in the Airflow UI.
    env_vars={'MY_VALUE': '{{ var.value.my_value }}'},
    # Sets the config file to a kubernetes config file specified in
    # airflow.cfg. If the configuration file does not exist or does
    # not provide validcredentials the pod will fail to launch. If not
    # specified, config_file defaults to ~/.kube/config
    config_file="{{ conf.get('core', 'kube_config') }}")

Airflow 1

kubenetes_template_ex = kubernetes_pod_operator.KubernetesPodOperator(
    task_id='ex-kube-templates',
    name='ex-kube-templates',
    namespace='default',
    image='bash',
    # All parameters below are able to be templated with jinja -- cmds,
    # arguments, env_vars, and config_file. For more information visit:
    # https://airflow.apache.org/docs/apache-airflow/stable/macros-ref.html

    # Entrypoint of the container, if not specified the Docker container's
    # entrypoint is used. The cmds parameter is templated.
    cmds=['echo'],
    # DS in jinja is the execution date as YYYY-MM-DD, this docker image
    # will echo the execution date. Arguments to the entrypoint. The docker
    # image's CMD is used if this is not provided. The arguments parameter
    # is templated.
    arguments=['{{ ds }}'],
    # The var template variable allows you to access variables defined in
    # Airflow UI. In this case we are getting the value of my_value and
    # setting the environment variable `MY_VALUE`. The pod will fail if
    # `my_value` is not set in the Airflow UI.
    env_vars={'MY_VALUE': '{{ var.value.my_value }}'},
    # Sets the config file to a kubernetes config file specified in
    # airflow.cfg. If the configuration file does not exist or does
    # not provide validcredentials the pod will fail to launch. If not
    # specified, config_file defaults to ~/.kube/config
    config_file="{{ conf.get('core', 'kube_config') }}")

Without changing the DAG or your environment, the ex-kube-templates task fails because of two errors. The logs show this task is failing because the appropriate variable does not exist (my_value). The second error, which you can get after fixing the first error, shows that the task fails because core/kube_config is not found in config.

To fix both errors, follow the steps outlined further.

To set my_value with gcloud or the Airflow UI:

Airflow UI

In the Airflow 2 UI:

  1. Go to the Airflow UI.

  2. In the toolbar, select Admin > Variables.

  3. On the List Variable page, click Add a new record.

  4. On the Add Variable page, enter the following information:

    • Key:my_value
    • Val: example_value
  5. Click Save.

In the Airflow 1 UI:

  1. Go to the Airflow UI.

  2. In the toolbar, select Admin > Variables.

  3. On the Variables page, click the Create tab.

  4. On the Variable page, enter the following information:

    • Key:my_value
    • Val: example_value
  5. Click Save.

gcloud

For Airflow 2, enter the following command:

gcloud composer environments run ENVIRONMENT \
    --location LOCATION \
    variables set -- \
    my_value example_value

For Airflow 1, enter the following command:

gcloud composer environments run ENVIRONMENT \
    --location LOCATION \
    variables -- \
    --set my_value example_value

Replace:

  • ENVIRONMENT with the name of the environment.
  • LOCATION with the Compute Engine region where the environment is located.

To refer to a custom config_file (a Kubernetes configuration file), override the kube_config Airflow configuration option to a valid Kubernetes configuration:

Section Key Value
core kube_config /home/airflow/composer_kube_config

Wait a few minutes for your environment to update. Then run the ex-kube-templates task again and verify that the ex-kube-templates task succeeds.

Secret Variables Configuration

A Kubernetes secret is an object that contains sensitive data. You can pass secrets to the Kubernetes pods by using the KubernetesPodOperator. Secrets must be defined in Kubernetes, or the pod fails to launch.

This example shows two ways of using Kubernetes Secrets: as an environment variable, and as a volume mounted by the pod.

The first secret, airflow-secrets, is set to a Kubernetes environment variable named SQL_CONN (as opposed to an Airflow or Cloud Composer environment variable).

The second secret, service-account, mounts service-account.json, a file with a service account token, to /var/secrets/google.

Here's what the secrets look like:

Airflow 2

secret_env = Secret(
    # Expose the secret as environment variable.
    deploy_type='env',
    # The name of the environment variable, since deploy_type is `env` rather
    # than `volume`.
    deploy_target='SQL_CONN',
    # Name of the Kubernetes Secret
    secret='airflow-secrets',
    # Key of a secret stored in this Secret object
    key='sql_alchemy_conn')
secret_volume = Secret(
    deploy_type='volume',
    # Path where we mount the secret as volume
    deploy_target='/var/secrets/google',
    # Name of Kubernetes Secret
    secret='service-account',
    # Key in the form of service account file name
    key='service-account.json')

Airflow 1

secret_env = secret.Secret(
    # Expose the secret as environment variable.
    deploy_type='env',
    # The name of the environment variable, since deploy_type is `env` rather
    # than `volume`.
    deploy_target='SQL_CONN',
    # Name of the Kubernetes Secret
    secret='airflow-secrets',
    # Key of a secret stored in this Secret object
    key='sql_alchemy_conn')
secret_volume = secret.Secret(
    deploy_type='volume',
    # Path where we mount the secret as volume
    deploy_target='/var/secrets/google',
    # Name of Kubernetes Secret
    secret='service-account',
    # Key in the form of service account file name
    key='service-account.json')

The name of the first Kubernetes secret is defined in the secret variable. This particular secret is named airflow-secrets. It is exposed as an environment variable, as dictated by the deploy_type. The environment variable it sets to, deploy_target, is SQL_CONN. Finally, the key of the secret that is stored in the deploy_target is sql_alchemy_conn.

The name of the second Kubernetes secret is defined in the secret variable. This particular secret is named service-account. It is exposed as an volume, as dictated by the deploy_type. The path of the file to mount, deploy_target, is /var/secrets/google. Finally, the key of the secret that is stored in the deploy_target is service-account.json.

Here's what the operator configuration looks like:

Airflow 2

kubernetes_secret_vars_ex = KubernetesPodOperator(
    task_id='ex-kube-secrets',
    name='ex-kube-secrets',
    namespace='default',
    image='ubuntu',
    startup_timeout_seconds=300,
    # The secrets to pass to Pod, the Pod will fail to create if the
    # secrets you specify in a Secret object do not exist in Kubernetes.
    secrets=[secret_env, secret_volume],
    # env_vars allows you to specify environment variables for your
    # container to use. env_vars is templated.
    env_vars={
        'EXAMPLE_VAR': '/example/value',
        'GOOGLE_APPLICATION_CREDENTIALS': '/var/secrets/google/service-account.json '})

Airflow 1

kubernetes_secret_vars_ex = kubernetes_pod_operator.KubernetesPodOperator(
    task_id='ex-kube-secrets',
    name='ex-kube-secrets',
    namespace='default',
    image='ubuntu',
    startup_timeout_seconds=300,
    # The secrets to pass to Pod, the Pod will fail to create if the
    # secrets you specify in a Secret object do not exist in Kubernetes.
    secrets=[secret_env, secret_volume],
    # env_vars allows you to specify environment variables for your
    # container to use. env_vars is templated.
    env_vars={
        'EXAMPLE_VAR': '/example/value',
        'GOOGLE_APPLICATION_CREDENTIALS': '/var/secrets/google/service-account.json '})

Without making any changes to the DAG or your environment, the ex-kube-secrets task fails. If you look at the logs, the task fails because of a Pod took too long to start error. This error occurs because Airflow cannot find the secret specified in the configuration, secret_env.

gcloud

To set the secret using gcloud:

  1. Get information about your Cloud Composer environment cluster.

    1. Run the following command:

      gcloud composer environments describe ENVIRONMENT \
          --location LOCATION \
          --format="value(config.gkeCluster)"
      

      Replace:

      • ENVIRONMENT with the name of your environment.
      • LOCATION with the region where the Cloud Composer environment is located.

      The output of this command uses the following format: projects/<your-project-id>/zones/<zone-of-composer-env>/clusters/<your-cluster-id>.

    2. To get the GKE cluster ID, copy the output after /clusters/ (ends in -gke).

    3. To get the zone, copy the output after /zones/.

  2. Connect to your GKE cluster by running the following command:

    gcloud container clusters get-credentials CLUSTER_ID \
      --project PROJECT \
      --zone ZONE
    

    Replace:

    • CLUSTER_ID with your GKE cluster ID.
    • PROJECT with the ID of your Google Cloud project.
    • ZONE with the zone where your GKE is located.
  3. Create Kubernetes secrets.

    1. Create a Kubernetes secret that sets the value of sql_alchemy_conn to test_value by running the following command:

      kubectl create secret generic airflow-secrets \
        --from-literal sql_alchemy_conn=test_value
      
    2. Create a Kubernetes secret that sets the value of service-account.json to a local path of a service account key file called key.json by running the following command:

      kubectl create secret generic service-account \
        --from-file service-account.json=./key.json
      
  4. After you set the secrets, run the ex-kube-secrets task again in the Airflow UI.

  5. Verify the ex-kube-secrets task succeeds.

Pod Affinity Configuration

When you configure the affinity parameter in the KubernetesPodOperator, you control what nodes to schedule pods on, such as nodes only in a particular node pool. In this example, the operator runs only on node pools named pool-0 and pool-1. Your Cloud Composer 1 environment nodes are in the default-pool, so your pods do not run on the nodes in your environment.

Airflow 2

# Pod affinity with the KubernetesPodOperator
# is not supported with Composer 2
# instead, create a cluster and use the GKEStartPodOperator
# https://cloud.google.com/composer/docs/using-gke-operator
kubernetes_affinity_ex = KubernetesPodOperator(
    task_id='ex-pod-affinity',
    name='ex-pod-affinity',
    namespace='default',
    image='perl',
    cmds=['perl'],
    arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
    # affinity allows you to constrain which nodes your pod is eligible to
    # be scheduled on, based on labels on the node. In this case, if the
    # label 'cloud.google.com/gke-nodepool' with value
    # 'nodepool-label-value' or 'nodepool-label-value2' is not found on any
    # nodes, it will fail to schedule.
    affinity={
        'nodeAffinity': {
            # requiredDuringSchedulingIgnoredDuringExecution means in order
            # for a pod to be scheduled on a node, the node must have the
            # specified labels. However, if labels on a node change at
            # runtime such that the affinity rules on a pod are no longer
            # met, the pod will still continue to run on the node.
            'requiredDuringSchedulingIgnoredDuringExecution': {
                'nodeSelectorTerms': [{
                    'matchExpressions': [{
                        # When nodepools are created in Google Kubernetes
                        # Engine, the nodes inside of that nodepool are
                        # automatically assigned the label
                        # 'cloud.google.com/gke-nodepool' with the value of
                        # the nodepool's name.
                        'key': 'cloud.google.com/gke-nodepool',
                        'operator': 'In',
                        # The label key's value that pods can be scheduled
                        # on.
                        'values': [
                            'pool-0',
                            'pool-1',
                        ]
                    }]
                }]
            }
        }
    })

Airflow 1

kubernetes_affinity_ex = kubernetes_pod_operator.KubernetesPodOperator(
    task_id='ex-pod-affinity',
    name='ex-pod-affinity',
    namespace='default',
    image='perl',
    cmds=['perl'],
    arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
    # affinity allows you to constrain which nodes your pod is eligible to
    # be scheduled on, based on labels on the node. In this case, if the
    # label 'cloud.google.com/gke-nodepool' with value
    # 'nodepool-label-value' or 'nodepool-label-value2' is not found on any
    # nodes, it will fail to schedule.
    affinity={
        'nodeAffinity': {
            # requiredDuringSchedulingIgnoredDuringExecution means in order
            # for a pod to be scheduled on a node, the node must have the
            # specified labels. However, if labels on a node change at
            # runtime such that the affinity rules on a pod are no longer
            # met, the pod will still continue to run on the node.
            'requiredDuringSchedulingIgnoredDuringExecution': {
                'nodeSelectorTerms': [{
                    'matchExpressions': [{
                        # When nodepools are created in Google Kubernetes
                        # Engine, the nodes inside of that nodepool are
                        # automatically assigned the label
                        # 'cloud.google.com/gke-nodepool' with the value of
                        # the nodepool's name.
                        'key': 'cloud.google.com/gke-nodepool',
                        'operator': 'In',
                        # The label key's value that pods can be scheduled
                        # on.
                        'values': [
                            'pool-0',
                            'pool-1',
                        ]
                    }]
                }]
            }
        }
    })

As the example is currently configured, the task fails. If you look at the logs, the task fails because node pools pool-0 and pool-1 do not exist.

To make sure the node pools in values exist, make any of the following configuration changes:

  • If you created a node pool earlier, replace pool-0 and pool-1 with the names of your node pools and upload your DAG again.

  • Create a node pool named pool-0 or pool-1. You can create both, but the task needs only one to succeed.

  • Replace pool-0 and pool-1 with default-pool, which is the default pool that Airflow uses. Then, upload your DAG again.

After you make the changes, wait a few minutes for your environment to update. Then run the ex-pod-affinity task again and verify that the ex-pod-affinity task succeeds.

Full Configuration

This example shows all the variables that you can configure in the KubernetesPodOperator. You don't need to modify the code for the the ex-all-configs task to succeed.

For details on each variable, see the Airflow KubernetesPodOperator reference.

Airflow 2

kubernetes_full_pod = KubernetesPodOperator(
    task_id='ex-all-configs',
    name='pi',
    namespace='default',
    image='perl',
    # Entrypoint of the container, if not specified the Docker container's
    # entrypoint is used. The cmds parameter is templated.
    cmds=['perl'],
    # Arguments to the entrypoint. The docker image's CMD is used if this
    # is not provided. The arguments parameter is templated.
    arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
    # The secrets to pass to Pod, the Pod will fail to create if the
    # secrets you specify in a Secret object do not exist in Kubernetes.
    secrets=[],
    # Labels to apply to the Pod.
    labels={'pod-label': 'label-name'},
    # Timeout to start up the Pod, default is 120.
    startup_timeout_seconds=120,
    # The environment variables to be initialized in the container
    # env_vars are templated.
    env_vars={'EXAMPLE_VAR': '/example/value'},
    # If true, logs stdout output of container. Defaults to True.
    get_logs=True,
    # Determines when to pull a fresh image, if 'IfNotPresent' will cause
    # the Kubelet to skip pulling an image if it already exists. If you
    # want to always pull a new image, set it to 'Always'.
    image_pull_policy='Always',
    # Annotations are non-identifying metadata you can attach to the Pod.
    # Can be a large range of data, and can include characters that are not
    # permitted by labels.
    annotations={'key1': 'value1'},
    # Resource specifications for Pod, this will allow you to set both cpu
    # and memory limits and requirements.
    # Prior to Airflow 1.10.4, resource specifications were
    # passed as a Pod Resources Class object,
    # If using this example on a version of Airflow prior to 1.10.4,
    # import the "pod" package from airflow.contrib.kubernetes and use
    # resources = pod.Resources() instead passing a dict
    # For more info see:
    # https://github.com/apache/airflow/pull/4551
    resources={'limit_memory': "250M", 'limit_cpu': "100m"},
    # Specifies path to kubernetes config. If no config is specified will
    # default to '~/.kube/config'. The config_file is templated.
    config_file='/home/airflow/composer_kube_config',
    # If true, the content of /airflow/xcom/return.json from container will
    # also be pushed to an XCom when the container ends.
    do_xcom_push=False,
    # List of Volume objects to pass to the Pod.
    volumes=[],
    # List of VolumeMount objects to pass to the Pod.
    volume_mounts=[],
    # Affinity determines which nodes the Pod can run on based on the
    # config. For more information see:
    # https://kubernetes.io/docs/concepts/configuration/assign-pod-node/
    # Pod affinity with the KubernetesPodOperator
    # is not supported with Composer 2
    # instead, create a cluster and use the GKEStartPodOperator
    # https://cloud.google.com/composer/docs/using-gke-operator
    affinity={})

Airflow 1

kubernetes_full_pod = kubernetes_pod_operator.KubernetesPodOperator(
    task_id='ex-all-configs',
    name='pi',
    namespace='default',
    image='perl',
    # Entrypoint of the container, if not specified the Docker container's
    # entrypoint is used. The cmds parameter is templated.
    cmds=['perl'],
    # Arguments to the entrypoint. The docker image's CMD is used if this
    # is not provided. The arguments parameter is templated.
    arguments=['-Mbignum=bpi', '-wle', 'print bpi(2000)'],
    # The secrets to pass to Pod, the Pod will fail to create if the
    # secrets you specify in a Secret object do not exist in Kubernetes.
    secrets=[],
    # Labels to apply to the Pod.
    labels={'pod-label': 'label-name'},
    # Timeout to start up the Pod, default is 120.
    startup_timeout_seconds=120,
    # The environment variables to be initialized in the container
    # env_vars are templated.
    env_vars={'EXAMPLE_VAR': '/example/value'},
    # If true, logs stdout output of container. Defaults to True.
    get_logs=True,
    # Determines when to pull a fresh image, if 'IfNotPresent' will cause
    # the Kubelet to skip pulling an image if it already exists. If you
    # want to always pull a new image, set it to 'Always'.
    image_pull_policy='Always',
    # Annotations are non-identifying metadata you can attach to the Pod.
    # Can be a large range of data, and can include characters that are not
    # permitted by labels.
    annotations={'key1': 'value1'},
    # Resource specifications for Pod, this will allow you to set both cpu
    # and memory limits and requirements.
    # Prior to Airflow 1.10.4, resource specifications were
    # passed as a Pod Resources Class object,
    # If using this example on a version of Airflow prior to 1.10.4,
    # import the "pod" package from airflow.contrib.kubernetes and use
    # resources = pod.Resources() instead passing a dict
    # For more info see:
    # https://github.com/apache/airflow/pull/4551
    resources={'limit_memory': "250M", 'limit_cpu': "100m"},
    # Specifies path to kubernetes config. If no config is specified will
    # default to '~/.kube/config'. The config_file is templated.
    config_file='/home/airflow/composer_kube_config',
    # If true, the content of /airflow/xcom/return.json from container will
    # also be pushed to an XCom when the container ends.
    do_xcom_push=False,
    # List of Volume objects to pass to the Pod.
    volumes=[],
    # List of VolumeMount objects to pass to the Pod.
    volume_mounts=[],
    # Affinity determines which nodes the Pod can run on based on the
    # config. For more information see:
    # https://kubernetes.io/docs/concepts/configuration/assign-pod-node/
    affinity={})

Troubleshooting

Tips for troubleshooting Pod failures

In addition to checking the task logs in the Airflow UI, also check the following logs:

  • Output of the Airflow scheduler and workers:

    1. In the Google Cloud Console, go to the Environments page.

      Go to Environments

    2. Follow the DAGs link for your environment.

    3. In the bucket of your environment, go up one level.

    4. Review the logs in the logs/<DAG_NAME>/<TASK_ID>/<EXECUTION_DATE> folder.

  • Detailed pod logs in the Cloud Console under GKE workloads. These logs include the pod definition YAML file, pod events, and pod details.

Non-zero return codes when also using the GKEStartPodOperator

When using the KubernetesPodOperator and the GKEStartPodOperator, the return code of the container's entrypoint determines whether the task is considered successful or not. Non-zero return codes indicate failure.

A common pattern when using the KubernetesPodOperator and the GKEStartPodOperator is to execute a shell script as the container entrypoint to group together multiple operations within the container.

If you are writing such a script, we recommended that you include the set -e command at the top of the script so that failed commands in the script terminate the script and propagate the failure to the Airflow task instance.

Pod timeouts

The default timeout for KubernetesPodOperator is 120 seconds, which can result in timeouts occurring before larger images download. You can increase the timeout by altering the startup_timeout_seconds parameter when you create the KubernetesPodOperator.

When a pod times out, the task specific log is available in the Airflow UI. For example:

Executing <Task(KubernetesPodOperator): ex-all-configs> on 2018-07-23 19:06:58.133811
Running: ['bash', '-c', u'airflow run kubernetes-pod-example ex-all-configs 2018-07-23T19:06:58.133811 --job_id 726 --raw -sd DAGS_FOLDER/kubernetes_pod_operator_sample.py']
Event: pod-name-9a8e9d06 had an event of type Pending
...
...
Event: pod-name-9a8e9d06 had an event of type Pending
Traceback (most recent call last):
  File "/usr/local/bin/airflow", line 27, in <module>
    args.func(args)
  File "/usr/local/lib/python2.7/site-packages/airflow/bin/cli.py", line 392, in run
    pool=args.pool,
  File "/usr/local/lib/python2.7/site-packages/airflow/utils/db.py", line 50, in wrapper
    result = func(*args, **kwargs)
  File "/usr/local/lib/python2.7/site-packages/airflow/models.py", line 1492, in _run_raw_task
    result = task_copy.execute(context=context)
  File "/usr/local/lib/python2.7/site-packages/airflow/contrib/operators/kubernetes_pod_operator.py", line 123, in execute
    raise AirflowException('Pod Launching failed: {error}'.format(error=ex))
airflow.exceptions.AirflowException: Pod Launching failed: Pod took too long to start

Pod Timeouts can also occur when the Cloud Composer Service Account lacks the necessary IAM permissions to perform the task at hand. To verify this, look at pod-level errors using the GKE Dashboards to look at the logs for your particular Workload, or use Cloud Logging.

Failed to Establish a New Connection

Auto-upgrade is enabled by default in GKE clusters. If a node pool is in a cluster that is upgrading, you might see the following error:

<Task(KubernetesPodOperator): gke-upgrade> Failed to establish a new
connection: [Errno 111] Connection refused

To check if your cluster is upgrading, in Google Cloud Console, go to the Kubernetes clusters page and look for the loading icon next to your environment's cluster name.

What's next