time-based-trigger

Trigger Function on time-based schedule

Overview

This sample demonstrates how to trigger Kyma Function on your custom schedule. It is based on the hello-timer SAP Faas Runtime example and can serve the purpose of showing how a similar use case can be migrated from deprecated SAP Faas Runtime into the Kyma runtime.

Steps

Inspect the Function files

Go to the time-based-trigger/hello-timer folder and inspect the code (handler.js), dependencies (package.json) and the Function configuration file, which manifests the features of the Function (config.yaml). As you see, the timer is not configurable in the manifest. The time-based triggering can be realised using the Cron Job feature that is built in Kubernetes.

Deploy the Function using Kyma CLI

NOTE: If you prefer to deploy the scenario using kubectl CLI, use the attached hello-timer-resources.yaml with the kubectl apply command and skip to the testing part.

Run the following command to deploy the Function:

$ kyma apply function

To verify if the Function was built run:

$ kubectl get functions   
NAME              CONFIGURED   BUILT   RUNNING   RUNTIME    VERSION   AGE
hello-timer       True         True    True      nodejs14   1         41s

At this point the Function is iddle. Now you can deploy the Cron Job that triggers it, based on a desired schedule.

Create Cron Job

Collect the trigger configuration:

Export the Function name and Namespace as environment variables.

export TRIGGER_FN_NAME={FUNCTION NAME}
export TRIGGER_FN_NAMESPACE={FUNCTION NAMESPACE}

For example:

export TRIGGER_FN_NAME=hello-timer
export TRIGGER_FN_NAMESPACE=default

Export the Function UID:

export TRIGGER_FN_UID=$(kubectl get function -n ${TRIGGER_FN_NAMESPACE} ${TRIGGER_FN_NAME} -o=jsonpath={".metadata.uid"})

All this data allows you to define the Function as the Cron Job's owner. If the Function is deleted from the cluster the Cron Jobs are deleted as well.

Export the schedule of the trigger in cron format.

export TRIGGER_SCHEDULE={TRIGGER SCHEDULE}

For example */5 * * * * to schedule execution every 5 minutes.

Create Cron Job

Run the following kubectl command to create Cron Job:

cat <<EOF | kubectl create -f -
apiVersion: batch/v1
kind: CronJob
metadata:
  generateName: trigger-${TRIGGER_FN_NAME}-
  namespace: ${TRIGGER_FN_NAMESPACE}
  ownerReferences:
    - apiVersion: serverless.kyma-project.io/v1alpha1
      kind: Function
      name: ${TRIGGER_FN_NAME}
      uid: ${TRIGGER_FN_UID}
spec:
  schedule: "${TRIGGER_SCHEDULE}"
  concurrencyPolicy: "Forbid"
  failedJobsHistoryLimit: 1
  successfulJobsHistoryLimit: 0
  jobTemplate:
    spec:
      template:
        spec:
          containers:
            - name: cron-trigger
              image: eu.gcr.io/kyma-project/external/curlimages/curl:7.70.0
              imagePullPolicy: IfNotPresent
              command:
                - sh
                - -c
                - |
                  curl ${TRIGGER_FN_NAME}.${TRIGGER_FN_NAMESPACE};
                  x=\$?; curl -fsI -X POST http://localhost:15020/quitquitquit && exit \$x
          restartPolicy: Never
EOF

Test the setup

Inspect the Function logs. This can be done by a command similar to:

➜  ~ k logs hello-timer-mthpc-d6c879fb9-76cz8 function -f
run at 2022-07-13T10:05:06.959Z
run at 2022-07-13T10:10:09.077Z
run at 2022-07-13T10:15:05.686Z

Considerations

• This approach is not recommended if you need high accuracy in execution times. Kubernetes starts Cron Job (deploys a container), and there is a few seconds delay before your Function gets executed.
• You can consider running your logic in the Cron Job container, instead of having Cron Job calling your Function using its internal address.
• This approach is not suitable for very short intervals (every 30 seconds or shorter). It takes time to deploy the Cron Job's container and clean it up once it's done. With short execution intervals (for example, every 10 seconds) there might be a resource problem (cluttering the cluster).