Configure logging and monitoring

Google Distributed Cloud (software only) for bare metal supports multiple options for cluster logging and monitoring, including cloud-based managed services, open source tools, and validated compatibility with third-party commercial solutions. This page explains these options and provides some basic guidance on selecting the proper solution for your environment.

This page is for Admins and architects and Operators who want to monitor the health of deployed applications or services, such as for service level objective (SLO) compliance. To learn more about common roles and example tasks that we reference in Google Cloud content, see Common GKE Enterprise user roles and tasks.

Options for Google Distributed Cloud

You have several logging and monitoring options for your cluster:

  • Cloud Logging and Cloud Monitoring, enabled by default on bare metal system components.
  • Prometheus and Grafana are available from the Cloud Marketplace.
  • Validated configurations with third-party solutions.

Cloud Logging and Cloud Monitoring

Google Cloud Observability is the built-in observability solution for Google Cloud. It offers a fully managed logging solution, metrics collection, monitoring, dashboarding, and alerting. Cloud Monitoring monitors Google Distributed Cloud clusters in a similar way as cloud-based GKE clusters.

Cloud Logging and Cloud Monitoring is enabled by default when you create clusters with the required service accounts and IAM roles. You can't disable Cloud Logging and Cloud Monitoring. For more information service accounts and the required roles, see Configure service accounts.

The agents can be configured to change the scope of logging and monitoring, as well as the level of metrics collected:

  • Scope of logging and monitoring can be set to system components only (default) or for system components and applications.
  • Level of metrics collected can be configured for an optimized set of metrics (default) or for full metrics.

See Configuring Stackdriver agents for Google Distributed Cloud on this document for more information.

Logging and Monitoring provide a single, easy-to-configure, powerful cloud-based observability solution. We highly recommend Logging and Monitoring when running workloads on Google Distributed Cloud. For applications with components running on Google Distributed Cloud and standard on-premises infrastructure, you might consider other solutions for an end-to-end view of those applications.

Prometheus and Grafana

Prometheus and Grafana are two popular open source monitoring products available in the Cloud Marketplace:

  • Prometheus collects application and system metrics.

  • Alertmanager handles sending out alerts with several different alerting mechanisms.

  • Grafana is a dashboarding tool.

We recommend that you use Google Cloud Managed Service for Prometheus, which is based in Cloud Monitoring, for all your monitoring needs. With Google Cloud Managed Service for Prometheus you can monitor system components without charge. Google Cloud Managed Service for Prometheus is also compatible with Grafana. However, if you prefer a pure local monitoring system, you can choose to install Prometheus and Grafana in your clusters.

If you installed Prometheus locally and want to collect metrics from system components, you need to give permission to your local Prometheus instance to access the metrics endpoints of system components:

  • Bind the service account for your Prometheus instance to the predefined gke-metrics-agent ClusterRole, and use service account token as credential to scrape metrics from the following system components:

    • kube-apiserver
    • kube-scheduler
    • kube-controller-manager
    • kubelet
    • node-exporter

  • Use the client key and cert stored in the kube-system/stackdriver-prometheus-etcd-scrape secret to authenticate the metric scrape from etcd.

  • Create a NetworkPolicy to allow access from your namespace to kube-state-metrics.

Third-party solutions

Google has worked with several third-party logging and monitoring solution providers to help their products work well with Google Distributed Cloud. These include Datadog, Elastic, and Splunk. Additional validated third parties will be added in the future.

The following solution guides are available for using third-party solutions with Google Distributed Cloud:

How Logging and Monitoring for Google Distributed Cloud works

Cloud Logging and Cloud Monitoring are installed and activated in each cluster when you create a new admin or user cluster.

The Stackdriver agents include several components on each cluster:

  • Stackdriver Operator (stackdriver-operator-*). Manages the lifecycle for all other Stackdriver agents deployed onto the cluster.

  • Stackdriver Custom Resource. A resource that is automatically created as part of the Google Distributed Cloud installation process.

  • GKE Metrics Agent (gke-metrics-agent-*). An OpenTelemetry Collector based DaemonSet that scrapes metrics from each node to Cloud Monitoring. A node-exporter DaemonSet and a kube-state-metrics deployment are also included to provide more metrics about the cluster.

  • Stackdriver Log Forwarder (stackdriver-log-forwarder-*). A Fluent Bit DaemonSet that forwards logs from each machine to the Cloud Logging. The log Forwarder buffers the log entries on the node locally and re-sends them for up to 4 hours. If the buffer gets full or if the Log Forwarder can't reach the Cloud Logging API for more than 4 hours, logs are dropped.

  • Metadata Agent (stackdriver-metadata-agent-). A deployment that sends metadata for Kubernetes resources such as pods, deployments, or nodes to the Config Monitoring for Ops API; this data is used to enrich metric queries by enabling you to query by deployment name, node name, or even Kubernetes service name.

You can see the agents installed by Stackdriver by running the following command:

kubectl -n kube-system get pods -l "managed-by=stackdriver"

The output of this command is similar to the following:

kube-system   gke-metrics-agent-4th8r                                     1/1     Running   1 (40h ago)   40h
kube-system   gke-metrics-agent-8lt4s                                     1/1     Running   1 (40h ago)   40h
kube-system   gke-metrics-agent-dhxld                                     1/1     Running   1 (40h ago)   40h
kube-system   gke-metrics-agent-lbkl2                                     1/1     Running   1 (40h ago)   40h
kube-system   gke-metrics-agent-pblfk                                     1/1     Running   1 (40h ago)   40h
kube-system   gke-metrics-agent-qfwft                                     1/1     Running   1 (40h ago)   40h
kube-system   kube-state-metrics-9948b86dd-6chhh                          1/1     Running   1 (40h ago)   40h
kube-system   node-exporter-5s4pg                                         1/1     Running   1 (40h ago)   40h
kube-system   node-exporter-d9gwv                                         1/1     Running   2 (40h ago)   40h
kube-system   node-exporter-fhbql                                         1/1     Running   1 (40h ago)   40h
kube-system   node-exporter-gzf8t                                         1/1     Running   1 (40h ago)   40h
kube-system   node-exporter-tsrpp                                         1/1     Running   1 (40h ago)   40h
kube-system   node-exporter-xzww7                                         1/1     Running   1 (40h ago)   40h
kube-system   stackdriver-log-forwarder-8lwxh                             1/1     Running   1 (40h ago)   40h
kube-system   stackdriver-log-forwarder-f7cgf                             1/1     Running   2 (40h ago)   40h
kube-system   stackdriver-log-forwarder-fl5gf                             1/1     Running   1 (40h ago)   40h
kube-system   stackdriver-log-forwarder-q5lq8                             1/1     Running   2 (40h ago)   40h
kube-system   stackdriver-log-forwarder-www4b                             1/1     Running   1 (40h ago)   40h
kube-system   stackdriver-log-forwarder-xqgjc                             1/1     Running   1 (40h ago)   40h
kube-system   stackdriver-metadata-agent-cluster-level-5bb5b6d6bc-z9rx7   1/1     Running   1 (40h ago)   40h

Cloud Monitoring metrics

For a list of metrics collected by Cloud Monitoring, see View Google Distributed Cloud metrics.

Configuring Stackdriver agents for Google Distributed Cloud

The Stackdriver agents installed with Google Distributed Cloud collect data about system components for the purposes of maintaining and troubleshooting issues with your clusters. The following sections describe Stackdriver configuration and operating modes.

System Components Only (Default Mode)

Upon installation, Stackdriver agents are configured by default to collect logs and metrics, including performance details (for example, CPU and memory utilization), and similar metadata, for Google-provided system components. These include all workloads in the admin cluster, and for user clusters, workloads in the kube-system, gke-system, gke-connect, istio-system, and config-management- system namespaces.

System Components and Applications

To enable application logging and monitoring on top of the default mode, follow the steps in Enable application logging and monitoring.

Optimized metrics (Default metrics)

By default, the kube-state-metrics deployments running in the cluster collect and report an optimized set of kube metrics to Google Cloud Observability (formerly Stackdriver).

Fewer resources are needed to collect this optimized set of metrics, which improves overall performance and scalability.

To disable optimized metrics (not recommended), override the default setting in your Stackdriver custom resource.

Use Google Cloud Managed Service for Prometheus for selected system components

Google Cloud Managed Service for Prometheus is part of Cloud Monitoring and is available as an option for system components. The benefits of Google Cloud Managed Service for Prometheus include the following:

  • You can continue to use your existing Prometheus-based monitoring without altering your alerts and Grafana dashboards.

  • If you use both GKE and Google Distributed Cloud, you can use the same Prometheus Query Language (PromQL) for metrics on all your clusters. You can also use the PromQL tab in Metrics Explorer in the Google Cloud console.

Enable and disable Google Cloud Managed Service for Prometheus

Google Cloud Managed Service for Prometheus is enabled by default in Google Distributed Cloud.

To disable Google Cloud Managed Service for Prometheus:

  1. Open the Stackdriver object named stackdriver for editing:

    kubectl --kubeconfig CLUSTER_KUBECONFIG --namespace kube-system \
    edit stackdriver stackdriver

  2. Add the enableGMPForSystemMetrics feature gate, and set it to false:

    apiVersion: addons.gke.io/v1alpha1
kind: Stackdriver
metadata:
  name: stackdriver
  namespace: kube-system
spec:
  featureGates:
    enableGMPForSystemMetrics: false

  3. Close your editing session.

View metric data

When enableGMPForSystemMetrics is set to true, metrics for the following components have a different format for how they are stored and queried in Cloud Monitoring:

  • kube-apiserver
  • kube-scheduler
  • kube-controller-manager
  • kubelet and cadvisor
  • kube-state-metrics
  • node-exporter

In the new format, you can query the preceding metrics by using either PromQL or Monitoring Query Language (MQL):

PromQL

Example PromQL query:

histogram_quantile(0.95, sum(rate(apiserver_request_duration_seconds_bucket[5m])) by (le))

MQL

To use MQL, set the monitored resource to prometheus_target, use the metric name with kubernetes.io/anthos prefix, and add the Prometheus type as a suffix to the metric name.

fetch prometheus_target
| metric 'kubernetes.io/anthos/apiserver_request_duration_seconds/histogram'
| align delta(5m)
| every 5m
| group_by [], [value_histogram_percentile: percentile(value.histogram, 95)]

Configuring Grafana dashboards with Google Cloud Managed Service for Prometheus

To use Grafana with metrics data from Google Cloud Managed Service for Prometheus, you must first configure and authenticate the Grafana data source. To configure and authenticate the data source, you use the data source syncer (datasource-syncer) to generate OAuth2 credentials and sync them to Grafana through the Grafana data source API. The data source syncer sets the Cloud Monitoring API as the Prometheus server URL (the URL value starts with https://monitoring.googleapis.com) under the data source in Grafana.

Follow the steps in Query using Grafana to authenticate and configure a Grafana data source to query data from Google Cloud Managed Service for Prometheus.

A set of sample Grafana dashboards are provided in the anthos-samples repository on GitHub. To install the sample dashboards, do the following:

  1. Download the sample JSON files:

    git clone https://github.com/GoogleCloudPlatform/anthos-samples.git
cd anthos-samples/gmp-grafana-dashboards

  2. If your Grafana data source was created with a name different with Managed Service for Prometheus, change the datasource field in all the JSON files:

    sed -i "s/Managed Service for Prometheus/[DATASOURCE_NAME]/g" ./*.json

    Replace [DATASOURCE_NAME] with the name of the data source in your Grafana that was pointed to the Prometheus frontend service.

  3. Access Grafana UI from your browser, and select + Import under the Dashboards menu.

    Navigating to dashboard import in Grafana.

  4. Either upload the JSON file, or copy and paste the file content and select Load. Once the file content is successfully loaded, select Import. Optionally you can also change the dashboard name and UID before importing.

    Importing dashboard in Grafana.

  5. The imported dashboard should load successfully if your Google Distributed Cloud and the data source are configured correctly. For example, the following screenshot shows the dashboard configured by cluster-capacity.json.

    Cluster capacity dashboard in Grafana.

Additional resources

For more information about Google Cloud Managed Service for Prometheus, see the following:

Configuring Stackdriver component resources

When you create a cluster, Google Distributed Cloud automatically creates a Stackdriver custom resource. You can edit the spec in the custom resource to override the default values for CPU and memory requests and limits for a Stackdriver component, and you can separately override the default optimized metrics setting.

Overriding the default CPU and memory requests and limits for a Stackdriver component

Clusters with high pod density introduce higher logging and monitoring overhead. In extreme cases, Stackdriver components may report close to the CPU and memory utilization limit or even may be subject to constant restarts due to resource limits. In this case, to override the default values for CPU and memory requests and limits for a Stackdriver component, use the following steps:

  1. Run the following command to open your Stackdriver custom resource in a command line editor:

    kubectl -n kube-system edit stackdriver stackdriver

  2. In the Stackdriver custom resource, add the resourceAttrOverridesection under the spec field:

    resourceAttrOverride:
      DAEMONSET_OR_DEPLOYMENT_NAME/CONTAINER_NAME:
        LIMITS_OR_REQUESTS:
          RESOURCE: RESOURCE_QUANTITY

    Note that the resourceAttrOverride section overrides all existing default limits and requests for the component you specify. The following components are supported by resourceAttrOverride:

    • gke-metrics-agent/gke-metrics-agent
    • stackdriver-log-forwarder/stackdriver-log-forwarder
    • stackdriver-metadata-agent-cluster-level/metadata-agent
    • node-exporter/node-exporter
    • kube-state-metrics/kube-state-metrics

    An example file looks like the following:

    apiVersion: addons.gke.io/v1alpha1
kind: Stackdriver
metadata:
  name: stackdriver
  namespace: kube-system
spec:
  anthosDistribution: baremetal
  projectID: my-project
  clusterName: my-cluster
  clusterLocation: us-west-1a
  resourceAttrOverride:
    gke-metrics-agent/gke-metrics-agent:
      requests:
        cpu: 110m
        memory: 240Mi
      limits:
        cpu: 200m
        memory: 4.5Gi

  3. To save changes to the Stackdriver custom resource, save and quit your command line editor.

  4. Check the health of your Pod:

    kubectl -n kube-system get pods -l "managed-by=stackdriver"

    A response for a healthy Pod looks like the following:

    gke-metrics-agent-4th8r                1/1     Running   1   40h

  5. Check the Pod spec of the component to make sure the resources are set correctly.

    kubectl -n kube-system describe pod POD_NAME

    Replace POD_NAME with the name of the Pod you just changed. For example, gke-metrics-agent-4th8r.

    The response looks like the following:

      Name:         gke-metrics-agent-4th8r
  Namespace:    kube-system
  ...
  Containers:
    gke-metrics-agent:
      Limits:
        cpu: 200m
        memory: 4.5Gi
      Requests:
        cpu: 110m
        memory: 240Mi
      ...

Disable optimized metrics

By default, the kube-state-metrics deployments running in the cluster collect and report an optimized set of kube metrics to Stackdriver. If you require additional metrics, we recommend that you find a replacement from the list of Google Distributed Cloud metrics.

Here are some examples of replacements you might use:

Disabled metric Replacements
kube_pod_start_time container/uptime
kube_pod_container_resource_requests container/cpu/request_cores
container/memory/request_bytes
kube_pod_container_resource_limits container/cpu/limit_cores
container/memory/limit_bytes

To disable the optimized metrics default setting (not recommended), do the following:

  1. Open your Stackdriver custom resource in a command line editor:

    kubectl -n kube-system edit stackdriver stackdriver

  2. Set the optimizedMetrics field to false:

    apiVersion: addons.gke.io/v1alpha1
kind: Stackdriver
metadata:
name: stackdriver
namespace: kube-system
spec:
anthosDistribution: baremetal
projectID: my-project
clusterName: my-cluster
clusterLocation: us-west-1a
optimizedMetrics: false

  3. Save changes, and quit your command line editor.

Metrics Server

Metrics Server is the source of the container resource metrics for various autoscaling pipelines. Metrics Server retrieves metrics from kubelets and exposes them through the Kubernetes Metrics API. HPA and VPA then use these metrics to determine when to trigger autoscaling. Metrics server is scaled using addon- resizer.

In extreme cases where high pod density creates too much logging and monitoring overhead, Metrics Server might be stopped and restarted due to resource limitations. In this case, you can allocate more resources to metrics server by editing the metrics-server-config configmap in gke-managed-metrics-server namespace, and changing the value for cpuPerNode and memoryPerNode.

kubectl edit cm metrics-server-config -n gke-managed-metrics-server

The example content of the ConfigMap is:

apiVersion: v1
data:
  NannyConfiguration: |-
    apiVersion: nannyconfig/v1alpha1
    kind: NannyConfiguration
    cpuPerNode: 3m
    memoryPerNode: 20Mi
kind: ConfigMap

After updating the ConfigMap, recreate the metrics-server pods with the following command:

kubectl delete pod -l k8s-app=metrics-server -n gke-managed-metrics-server

Configuration requirements for Logging and Monitoring

There are several configuration requirements to enable Cloud Logging and Cloud Monitoring with Google Distributed Cloud. These steps are included in Configuring a service account for use with Logging and Monitoring on the Enabling Google services page, and in the following list:

  1. A Cloud Monitoring Workspace must be created within the Google Cloud project. This is accomplished by clicking Monitoring in Google Cloud console and following the workflow.

  2. You need to enable the following Stackdriver APIs:

  3. You need to assign the following IAM roles to the service account used by the Stackdriver agents:

    • logging.logWriter
    • monitoring.metricWriter
    • stackdriver.resourceMetadata.writer
    • monitoring.dashboardEditor
    • opsconfigmonitoring.resourceMetadata.writer

Log tags

Many Google Distributed Cloud logs have a tag of F:

logtag: "F"

This tag means that the log entry is complete or full. To learn more about this tag, see Log format in the Kubernetes design proposals on GitHub.

Pricing

There is no charge for Google Kubernetes Engine (GKE) Enterprise edition system logs and metrics.

In a Google Distributed Cloud cluster, Google Kubernetes Engine (GKE) Enterprise edition system logs and metrics include the following:

  • Logs and metrics from all components in an admin cluster.
  • Logs and metrics from components in these namespaces in a user cluster: kube-system, gke-system, gke-connect, knative-serving, istio-system, monitoring-system, config-management-system, gatekeeper-system, cnrm-system.

For more information, see Pricing for Google Cloud Observability.

To learn about credit for Cloud Logging metrics, contact sales for pricing.