Kubernetes and Google Cloud Monitoring Integration

Input and output integration overview

This plugin captures metrics for Kubernetes pods and containers by communicating with the Kubelet API.

The Stackdriver plugin allows users to send metrics directly to a specified project in Google Cloud Monitoring, facilitating robust monitoring capabilities across their cloud resources.

Integration details

Kubernetes

The Kubernetes input plugin interfaces with the Kubelet API to gather metrics for running pods and containers on a single host, ideally as part of a daemonset in a Kubernetes installation. By operating on each node within the cluster, it collects metrics from the locally running kubelet, ensuring that the data reflects the real-time state of the environment. Being a rapidly evolving project, Kubernetes sees frequent updates, and this plugin adheres to the major cloud providers’ supported versions, maintaining compatibility across multiple releases within a limited time span. Significant consideration is given to the potential high series cardinality, which can burden the database; thus, users are advised to implement filtering techniques and retention policies to manage this load effectively. Configuration options provide flexible customization of the plugin’s behavior to integrate seamlessly into different setups, enhancing its utility in monitoring Kubernetes environments.

Google Cloud Monitoring

This plugin writes metrics to a project in Google Cloud Monitoring, which used to be known as Stackdriver. Authentication is a prerequisite and can be achieved via service accounts or user credentials. The plugin is designed to group metrics by a namespace variable and metric key, facilitating organized data management. However, users are encouraged to use the official naming format for enhanced query efficiency. The plugin supports additional configurations for managing metric representation and allows tags to be treated as resource labels. Notably, it imposes certain restrictions on the data it can accept, such as not allowing string values or points that are out of chronological order.

Configuration

Kubernetes

[[inputs.kubernetes]]
  ## URL for the kubelet, if empty read metrics from all nodes in the cluster
  url = "http://127.0.0.1:10255"

  ## Use bearer token for authorization. ('bearer_token' takes priority)
  ## If both of these are empty, we'll use the default serviceaccount:
  ## at: /var/run/secrets/kubernetes.io/serviceaccount/token
  ##
  ## To re-read the token at each interval, please use a file with the
  ## bearer_token option. If given a string, Telegraf will always use that
  ## token.
  # bearer_token = "/var/run/secrets/kubernetes.io/serviceaccount/token"
  ## OR
  # bearer_token_string = "abc_123"

  ## Kubernetes Node Metric Name
  ## The default Kubernetes node metric name (i.e. kubernetes_node) is the same
  ## for the kubernetes and kube_inventory plugins. To avoid conflicts, set this
  ## option to a different value.
  # node_metric_name = "kubernetes_node"

  ## Pod labels to be added as tags.  An empty array for both include and
  ## exclude will include all labels.
  # label_include = []
  # label_exclude = ["*"]

  ## Set response_timeout (default 5 seconds)
  # response_timeout = "5s"

  ## Optional TLS Config
  # tls_ca = /path/to/cafile
  # tls_cert = /path/to/certfile
  # tls_key = /path/to/keyfile
  ## Use TLS but skip chain & host verification
  # insecure_skip_verify = false

Google Cloud Monitoring

[[outputs.stackdriver]]
  ## GCP Project
  project = "project-id"

  ## Quota Project
  ## Specifies the Google Cloud project that should be billed for metric ingestion.
  ## If omitted, the quota is charged to the service account’s default project.
  ## This is useful when sending metrics to multiple projects using a single service account.
  ## The caller must have the `serviceusage.services.use` permission on the specified project.
  # quota_project = ""

  ## The namespace for the metric descriptor
  ## This is optional and users are encouraged to set the namespace as a
  ## resource label instead. If omitted it is not included in the metric name.
  namespace = "telegraf"

  ## Metric Type Prefix
  ## The DNS name used with the metric type as a prefix.
  # metric_type_prefix = "custom.googleapis.com"

  ## Metric Name Format
  ## Specifies the layout of the metric name, choose from:
  ##  * path: 'metric_type_prefix_namespace_name_key'
  ##  * official: 'metric_type_prefix/namespace_name_key/kind'
  # metric_name_format = "path"

  ## Metric Data Type
  ## By default, telegraf will use whatever type the metric comes in as.
  ## However, for some use cases, forcing int64, may be preferred for values:
  ##   * source: use whatever was passed in
  ##   * double: preferred datatype to allow queries by PromQL.
  # metric_data_type = "source"

  ## Tags as resource labels
  ## Tags defined in this option, when they exist, are added as a resource
  ## label and not included as a metric label. The values from tags override
  ## the values defined under the resource_labels config options.
  # tags_as_resource_label = []

  ## Custom resource type
  # resource_type = "generic_node"

  ## Override metric type by metric name
  ## Metric names matching the values here, globbing supported, will have the
  ## metric type set to the corresponding type.
  # metric_counter = []
  # metric_gauge = []
  # metric_histogram = []

  ## NOTE: Due to the way TOML is parsed, tables must be at the END of the
  ## plugin definition, otherwise additional config options are read as part of
  ## the table

  ## Additional resource labels
  # [outputs.stackdriver.resource_labels]
  #   node_id = "$HOSTNAME"
  #   namespace = "myapp"
  #   location = "eu-north0"

Input and output integration examples

Kubernetes

1. Dynamic Resource Allocation Monitoring: By utilizing the Kubernetes plugin, teams can set up alerts for resource usage patterns across various pods and containers. This proactive monitoring approach enables automatic scaling of resources in response to specific thresholds—helping to optimize performance while minimizing costs during peak usage.

2. Multi-tenancy Resource Isolation Analysis: Organizations using Kubernetes can leverage this plugin to track resource consumption per namespace. In a multi-tenant scenario, understanding the resource allocations and usages across different teams becomes critical for ensuring fair access and performance guarantees, leading to better resource management strategies.

3. Real-time Health Dashboards: Integrate the data captured by the Kubernetes plugin into visualization tools like Grafana to create real-time dashboards. These dashboards provide insights into the overall health and performance of the Kubernetes environment, allowing teams to quickly identify and rectify issues across clusters, pods, and containers.

4. Automated Incident Response Workflows: By combining the Kubernetes plugin with alert management systems, teams can automate incident response procedures based on real-time metrics. If a pod’s resource usage exceeds predefined limits, an automated workflow can trigger remediation actions, such as restarting the pod or reallocating resources—all of which can help improve system resilience.

Google Cloud Monitoring

1. Multi-Project Metric Aggregation: Use this plugin to send aggregated metrics from various applications across different projects into a single Google Cloud Monitoring project. This use case helps centralize metrics for teams managing multiple applications, providing a unified view for performance monitoring and enhancing decision-making. By configuring different quota projects for billing, organizations can ensure proper cost management while benefiting from a consolidated monitoring strategy.

2. Anomaly Detection Setup: Integrate the plugin with a machine learning-based analytics tool that identifies anomalies in the collected metrics. Using the historical data provided by the plugin, the tool can learn normal baseline behavior and promptly alert the operations team when unusual patterns arise, enabling proactive troubleshooting and minimizing service disruptions.

3. Dynamic Resource Labeling: Implement dynamic tagging by utilizing the tags_as_resource_label option to adaptively attach resource labels based on runtime conditions. This setup allows metrics to provide context-sensitive information, such as varying environmental parameters or operational states, enhancing the granularity of monitoring and reporting without changing the fundamental metric structure.

4. Custom Metric Visualization Dashboards: Leverage the data collected by the Google Cloud Monitoring output plugin to feed a custom metrics visualization dashboard using a third-party framework. By visualizing metrics in real-time, teams can achieve better situational awareness, notably by correlating different metrics, improving operational decision-making, and streamlining performance management workflows.

Feedback

Thank you for being part of our community! If you have any general feedback or found any bugs on these pages, we welcome and encourage your input. Please submit your feedback in the InfluxDB community Slack.