How to start airflow-scheduler pod for Google Cloud Composer?

Question

A Composer cluster went down because its airflow-worker pods needed a Docker image that was not accessible.

Now access to the Docker image was restore, but the airflow-scheduler pod has disappeared.

I tried updating the Composer Environment by setting a new Environment Variable, with the following error :

UPDATE operation on this environment failed X minutes ago with the following error message: (404)
Reason: Not Found
HTTP response headers: HTTPHeaderDict({
  "Date": "recently",
  "Audit-Id": "my-own-audit-id",
  "Content-Length": "236",
  "Content-Type": "application/json",
  "Cache-Control": "no-cache, private"
})
HTTP response body: {
  "kind": "Status",
  "apiVersion": "v1",
  "metadata": {},
  "status": "Failure",
  "message": "deployments.apps \"airflow-scheduler\" not found",
  "reason": "NotFound",
  "details": {
    "name": "airflow-scheduler",
    "group": "apps",
    "kind": "deployments"
  },
  "code": 404
}

Error in Composer Agent

How can I launch a airflow-scheduler pod on my Composer cluster ?

What is the .yaml configuration file I need to apply ?

I tried launching the scheduler from inside another pod with airflow scheduler, and while it effectively starts a scheduler, it's not a Kubernetes pod and will not integrate well with the managed airflow cluster.

Ismail Ismail · Accepted Answer · 2021-01-20T20:59:00

To restart the airflow-scheduler, run the following

# Fetch the old deployment, and pipe it into the replace command.
COMPOSER_WORKSPACE=$(kubectl get namespace | egrep -i 'composer|airflow' | awk '{ print $1 }')
kubectl get deployment airflow-scheduler --output yaml \
    --namespace=${COMPOSER_WORKSPACE}| kubectl replace --force -f -

How to start airflow-scheduler pod for Google Cloud Composer?

1 Answers