Charmed Spark K8s

Enable and configuring Apache Spark monitoring

Charmed Apache Spark supports native integration with the Canonical Observability Stack (COS). If you want to enable monitoring on top of Charmed Apache Spark, make sure that you have a Juju model with COS correctly deployed.

To deploy COS on MicroK8s, follow the step-by-step tutorial. For more information about Charmed Apache Spark and COS integration, refer to the COS documentation and the monitoring explanation section.

Once COS is correctly deployed, to enable monitoring it is necessary to:

1. Integrate and configure the COS bundle with Charmed Apache Spark
2. Configure the Apache Spark service account

Integrating/Configuring with COS

The Charmed Apache Spark solution already bundles all the components required to integrate COS as well as to configure the monitoring artifacts.

The deployments of these resources can be enabled/disabled using either overlays (for Juju bundles) or input variables (for Terraform bundles). Please refer to the how-to deploy guide for more information.

After the deployment settles on an active/idle state, you can make sure that Grafana is correctly setup with dedicated dashboards. To do so, retrieve the credentials for logging into the Grafana dashboard, by switching to the COS model (using juju switch <cos_model>) and using the following action:

juju run grafana/leader get-admin-password

The action will also show you the grafana endpoint where you will be able to log in with the provided credentials for the admin user.

After the login, Grafana will show a new dashboard: Spark Dashboard where the Apache Spark metrics will be displayed.

Customize Charmed Apache Spark dashboards

The cos-configuration-k8s charm included in the bundle can be used to customise the Grafana dashboard. If needed, we provide a default dashboard as part of the bundle.

The cos-configuration-k8s charm syncs the content of a repository with the resources provided to Grafana, thus enabling versioning of the resource. You can specify the repository, branch and folder using the config options, i.e.

# deploy cos configuration charm to import the grafana dashboard
juju config cos-configuration \
  git_repo=<your-repo> \
  git_branch=<your-branch> \
  git_depth=1 \
  grafana_dashboards_path=<path-to-dashboard-folder>

After updating the configuration or whenever the repository is updated, contents is synced either upon an update-status event or after running the action:

juju run cos-configuration-k8s/leader sync-now

For more information, refer to the cos-configuration-k8s charm docs.

Configuring Scraping intervals

The prometheus-scrape-config-k8s charm included in the bundle can be used to configure the prometheus scraping jobs.

In particular, it is crucial to configure the scraping interval to make sure data points have proper sampling frequency, e.g.:

juju config scrape-config --config scrape_interval=<SCRAPE_INTERVAL>

For more information about the properties that can be set using prometheus-scrape-config-k8s, please refer to its documentation.

Enable Log Forwarding to Loki

Logs from each driver or executor can be enabled using two Spark configuration options:

• spark.executorEnv.LOKI_URL
• spark.kubernetes.driverEnv.LOKI_URL

They are used to forward executor and driver logs respectively to a Loki server.

There are two ways to provide the LOKI_URL variables:

1. Manually, via Spark configuration:

   • spark.executorEnv.LOKI_URL - for executors
   • spark.kubernetes.driverEnv.LOKI_URL - for drivers

2. Using the logging relation in an Integration Hub for Apache Spark charm either with the Grafana-agent charm (recommended) or directly with the Loki charm, for example:

   juju integrate spark-integration-hub-k8s:logging grafana-agent-k8s:logging-provider
   

Configure Apache Spark service account

Charmed Apache Spark service account created by spark-client snap and spark8t Python library are automatically configured to use monitoring by the spark-integration-hub-k8s charm, that is deployed as part of the Charmed Apache Spark bundle.

Just make sure that the spark-integration-hub-k8s charm is correctly related to the prometheus-pushgateway charm on the pushgateway interface.

You can also double-check that the configuration done by the spark-integration-hub-k8s was effective by inspecting the Charmed Apache Spark service account properties using the snap

spark-client.service-account-registry get-config --username <username> --namespace <namespace>

and check that the following property

spark.metrics.conf.driver.sink.prometheus.pushgateway-address=<PROMETHEUS_GATEWAY_ADDRESS>:<PROMETHEUS_PORT>

is configured with the correct values. The Prometheus Pushgateway address and port should be can be consistent with what is exposed by Juju, e.g.

PROMETHEUS_GATEWAY=$(juju status --format=yaml | yq ".applications.prometheus-pushgateway-k8s.address")

---