Creating alerting policies

This page shows how to create metric-based alerting policies for Google Distributed Cloud clusters. We've provided several downloadable samples to help you set up alerting policies for common scenarios. For more information about metric-based alerting policies, see Create metric-threshold alerting policies in Google Cloud Observability documentation.

Before you begin

You must have the following permissions to create alerting policies:

  • monitoring.alertPolicies.create
  • monitoring.alertPolicies.delete
  • monitoring.alertPolicies.update

You have these permissions if you have any one of the following roles:

  • monitoring.alertPolicyEditor
  • monitoring.editor
  • Project Editor
  • Project Owner

If you want to create log-based alerting policies by using the Google Cloud CLI, then you must also have the serviceusage.serviceUsageConsumer role. For instructions to set up log-based alerting policies, see Configure log-based alerts in Google Cloud Observability documentation.

To check your roles, go to the IAM page in the Google Cloud console.

Creating an example policy: API server unavailable

In this exercise, you create an alerting policy for Kubernetes API servers of clusters. With this policy in place, you can arrange to be notified whenever the API server of a cluster is unavailable.

  1. Download the policy configuration file: apiserver-unavailable.json

  2. Create the policy:

    gcloud alpha monitoring policies create --policy-from-file=POLICY_CONFIG

    Replace POLICY_CONFIG with the path of the configuration file you just downloaded.

  3. View your alerting policies:


    1. In the Google Cloud console, go to the Monitoring page.

      Go to Monitoring

    2. On the left, select Alerting.

    3. Under Policies, you can see a list of your alerting policies.

      In the list, select Anthos cluster API server unavailable (critical) to see details about your new policy. Under Conditions, you can see a description of the policy. For example:

      Policy violates when ANY condition is met
      Anthos cluster API server uptime is absent for 5m


    gcloud alpha monitoring policies list

    The output shows detailed information about the policy. For example:

    combiner: OR
    - conditionAbsent:
        - alignmentPeriod: 60s
          crossSeriesReducer: REDUCE_MEAN
          - resource.label.project_id
          - resource.label.location
          - resource.label.cluster_name
          - resource.label.namespace_name
          - resource.label.container_name
          - resource.label.pod_name
          perSeriesAligner: ALIGN_MAX
        duration: 300s
        filter: resource.type = "k8s_container" AND metric.type = ""
          AND resource.label."container_name"=monitoring.regex.full_match("kube-apiserver")
          count: 1
      displayName: Anthos cluster API server uptime is absent for 5m
      name: projects/…/alertPolicies/…/conditions/…
    displayName: Anthos cluster API server unavailable (critical)
    enabled: true
      mutateTime: …
      mutatedBy: …
    name: projects/…/alertPolicies/…

Creating additional alerting policies

This section provides descriptions and configuration files for a set of recommended alerting policies.

To create a policy, follow the same steps that you used in the preceding exercise:

  1. To download the configuration file, click the link in the right column.

  2. Optionally, tune the conditions to better fit your specific needs, for example, you can add additional filters for a subset of clusters, or adjust the threshold values to balance between noisiness and criticality.

  3. To create the policy, run gcloud alpha monitoring policies create.

You can download and install all of the alert policy ssamples described in this document with the following script:

# 1. Create a directory named alert_samples:

mkdir alert_samples && cd alert_samples
declare -a alerts=("apiserver-unavailable.json" "controller-manager-unavailable.json" "scheduler-unavailable.json" \
  "pod-crash-looping.json" "pod-not-ready-1h.json" "container-cpu-usage-high-reaching-limit.json" \
  "container-memory-usage-high-reaching-limit.json" "persistent-volume-usage-high.json" "node-cpu-usage-high.json" \
  "node-disk-usage-high.json" "node-memory-usage-high.json" "node-not-ready-1h.json" "apiserver-error-ratio-high.json" \
  "etcd-leader-changes-or-proposal-failures-frequent.json" "etcd-server-not-in-quorum.yaml" "etcd-storage-usage-high.json")

# 2. Download all alert samples into the alert_samples/ directory:

for x in "${alerts[@]}"

# 3. (optional) Uncomment and provide your project ID to set the default project
# for gcloud commands:

# gcloud config set project <PROJECT_ID>

# 4. Create alert policies for each of the downloaded samples:

for x in "${alerts[@]}"
  gcloud alpha monitoring policies create --policy-from-file=${x}

Control plane components availability

Alert name Description Alerting policy definition in Cloud Monitoring
API server unavailable (critical) API server uptime metric is unavailable apiserver-unavailable.json
Scheduler unavailable (critical) Scheduler uptime metric is unavailable scheduler-unavailable.json
Controller manager unavailable (critical) Controller manager uptime metric is unavailable controller-manager-unavailable.json

Kubernetes system

Alert name Description Alerting policy definition in Cloud Monitoring
Pod crash looping (warning) Pod keeps restarting and might be in a crash loop status pod-crash-looping.json
Pod not ready for more than one hour (critical) Pod is in a non-ready state for more than one hour pod-not-ready-1h.json
Container cpu usage exceeds 80 percent (warning) Container cpu usage is over 80% of limit container-cpu-usage-high-reaching-limit.json
Container memory usage exceeds 85 percent (warning) Container memory usage is over 85% of limit container-memory-usage-high-reaching-limit.json
Persistent volume high usage (critical) Claimed persistent volume has less than 3 percent of free space persistent-volume-usage-high.json
Node cpu usage exceeds 80 percent (warning) Node cpu usage is over 80% of total allocatable for 5m node-cpu-usage-high.json
Node disk usage exceeds 85 percent (warning) Less than 15 percent is free per disk mountpoint for 10 mins node-disk-usage-high.json
Node memory usage exceeds 80 percent (warning) Node memory usage is over 80% of total allocatable for 5m node-memory-usage-high.json
Node not ready for more than one hour (critical) Node is in a non-ready state for more than one hour node-not-ready-1h.json

Kubernetes performance

Alert name Description Alerting policy definition in Cloud Monitoring
API server error ratio exceeds 20 percent (critical) API server gives 5xx or 429 errors on more than 20% of all requests per verb for 15m apiserver-error-ratio-high.json
ETCD leader change or proposal failure too frequent (warning) The etcd leader changes or proposal failures happen too frequently etcd-leader-changes-or-proposal-failures-frequent.json
ETCD server is not in quorum (critical) No etcd server proposals committed for 5 min, so they might have lost quorum etcd-server-not-in-quorum.yaml
ETCD storage exceeds 90 percent limit (warning) The etcd storage usage is more than 90% of limit etcd-storage-usage-high.json

Alert Policies with PromQL

The queries in alert policies can also be expressed in PromQL instead of MQL. For example, the PromQL version of the API server error ratio exceeds 20 percent (critical) policy is available to download: apiserver-error-ratio-high-promql.json.

For more information, refer to the Use Managed Service for Prometheus for Google Distributed Cloud documentation and the Alerting policies with PromQL for Cloud Monitoring documentation.

Getting notified

After you create an alerting policy, you can define one or more notification channels for the policy. There are several kinds of notification channels. For example, you can be notified by email, a Slack channel, or a mobile app. You can choose the channels that suit your needs.

For instructions about how to configure notification channels, see Managing notification channels.