Scraping additional Prometheus sources and importing those metrics - Amazon CloudWatch
Prometheus scrape configuration CloudWatch agent configuration for PrometheusTutorial for adding a new Prometheus scrape target: Prometheus API Server metrics

Scraping additional Prometheus sources and importing those metrics

The CloudWatch agent with Prometheus monitoring needs two configurations to scrape the Prometheus metrics. One is for the standard Prometheus configurations as documented in <scrape_config> in the Prometheus documentation. The other is for the CloudWatch agent configuration.

For Amazon EKS clusters, the configurations are defined in prometheus-eks.yaml (for the EC2 launch type) or prometheus-eks-fargate.yaml (for the Fargate launch type) as two config maps:

  • The name: prometheus-config section contains the settings for Prometheus scraping.

  • The name: prometheus-cwagentconfig section contains the configuration for the CloudWatch agent. You can use this section to configure how the Prometheus metrics are collected by CloudWatch. For example, you specify which metrics are to be imported into CloudWatch, and define their dimensions.

For Kubernetes clusters running on Amazon EC2 instances, the configurations are defined in the prometheus-k8s.yaml YAML file as two config maps:

  • The name: prometheus-config section contains the settings for Prometheus scraping.

  • The name: prometheus-cwagentconfig section contains the configuration for the CloudWatch agent.

To scrape additional Prometheus metrics sources and import those metrics to CloudWatch, you modify both the Prometheus scrape configuration and the CloudWatch agent configuration, and then re-deploy the agent with the updated configuration.

VPC security group requirements

The ingress rules of the security groups for the Prometheus workloads must open the Prometheus ports to the CloudWatch agent for scraping the Prometheus metrics by the private IP.

The egress rules of the security group for the CloudWatch agent must allow the CloudWatch agent to connect to the Prometheus workloads' port by private IP.

Prometheus scrape configuration

The CloudWatch agent supports the standard Prometheus scrape configurations as documented in <scrape_config> in the Prometheus documentation. You can edit this section to update the configurations that are already in this file, and add additional Prometheus scraping targets. By default, the sample configuration file contains the following global configuration lines:

global:
  scrape_interval: 1m
  scrape_timeout: 10s

  • scrape_interval— Defines how frequently to scrape targets.

  • scrape_timeout— Defines how long to wait before a scrape request times out.

You can also define different values for these settings at the job level, to override the global configurations.

Prometheus scraping jobs

The CloudWatch agent YAML files already have some default scraping jobs configured. For example, in prometheus-eks.yaml, the default scraping jobs are configured in the job_name lines in the scrape_configs section. In this file, the following default kubernetes-pod-jmx section scrapes JMX exporter metrics.

   - job_name: 'kubernetes-pod-jmx'
      sample_limit: 10000
      metrics_path: /metrics
      kubernetes_sd_configs:
      - role: pod
      relabel_configs:
      - source_labels: [__address__]
        action: keep
        regex: '.*:9404$'
      - action: labelmap
        regex: __meta_kubernetes_pod_label_(.+)
      - action: replace
        source_labels:
        - __meta_kubernetes_namespace
        target_label: Namespace
      - source_labels: [__meta_kubernetes_pod_name]
        action: replace
        target_label: pod_name
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_container_name
        target_label: container_name
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_controller_name
        target_label: pod_controller_name
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_controller_kind
        target_label: pod_controller_kind
      - action: replace
        source_labels:
        - __meta_kubernetes_pod_phase
        target_label: pod_phase

Each of these default targets are scraped, and the metrics are sent to CloudWatch in log events using embedded metric format. For more information, see Embedding metrics within logs.

Log events from Amazon EKS and Kubernetes clusters are stored in the /aws/containerinsights/cluster_name/prometheus log group in CloudWatch Logs. Log events from Amazon ECS clusters are stored in the /aws/ecs/containerinsights/cluster_name/prometheus log group.

Each scraping job is contained in a different log stream in this log group. For example, the Prometheus scraping job kubernetes-pod-appmesh-envoy is defined for App Mesh. All App Mesh Prometheus metrics from Amazon EKS and Kubernetes clusters are sent to the log stream named /aws/containerinsights/cluster_name>prometheus/kubernetes-pod-appmesh-envoy/.

To add a new scraping target, you add a new job_name section to the scrape_configs section of the YAML file, and restart the agent. For an example of this process, see Tutorial for adding a new Prometheus scrape target: Prometheus API Server metrics.

CloudWatch agent configuration for Prometheus

The CloudWatch agent configuration file has a prometheus section under metrics_collected for the Prometheus scraping configuration. It includes the following configuration options:

  • cluster_name— specifies the cluster name to be added as a label in the log event. This field is optional. If you omit it, the agent can detect the Amazon EKS or Kubernetes cluster name.

  • log_group_name— specifies the log group name for the scraped Prometheus metrics. This field is optional. If you omit it, CloudWatch uses /aws/containerinsights/cluster_name/prometheus for logs from Amazon EKS and Kubernetes clusters.

  • prometheus_config_path— specifies the Prometheus scrape configuration file path. If the value of this field starts with env: the Prometheus scrape configuration file contents will be retrieved from the container's environment variable. Do not change this field.

  • ecs_service_discovery— is the section to specify the configuration for Amazon ECS Prometheus service discovery. For more information, see Detailed guide for autodiscovery on Amazon ECS clusters.

    The ecs_service_discovery section can contain the following fields:

    • sd_frequency is the frequency to discover the Prometheus exporters. Specify a number and a unit suffix. For example, 1m for once per minute or 30s for once per 30 seconds. Valid unit suffixes are ns, us, ms, s, m, and h.

      This field is optional. The default is 60 seconds (1 minute).

    • sd_target_cluster is the target Amazon ECS cluster name for auto-discovery. This field is optional. The default is the name of the Amazon ECS cluster where the CloudWatch agent is installed.

    • sd_cluster_region is the target Amazon ECS cluster's Region. This field is optional. The default is the Region of the Amazon ECS cluster where the CloudWatch agent is installed. .

    • sd_result_file is the path of the YAML file for the Prometheus target results. The Prometheus scrape configuration will refer to this file.

    • docker_label is an optional section that you can use to specify the configuration for docker label-based service discovery. If you omit this section, docker label-based discovery is not used. This section can contain the following fields:

      • sd_port_label is the container's docker label name that specifies the container port for Prometheus metrics. The default value is ECS_PROMETHEUS_EXPORTER_PORT. If the container does not have this docker label, the CloudWatch agent will skip it.

      • sd_metrics_path_label is the container's docker label name that specifies the Prometheus metrics path. The default value is ECS_PROMETHEUS_METRICS_PATH. If the container does not have this docker label, the agent assumes the default path /metrics.

      • sd_job_name_label is the container's docker label name that specifies the Prometheus scrape job name. The default value is job. If the container does not have this docker label, the CloudWatch agent uses the job name in the Prometheus scrape configuration.

    • task_definition_list is an optional section that you can use to specify the configuration of task definition-based service discovery. If you omit this section, task definition-based discovery is not used. This section can contain the following fields:

      • sd_task_definition_arn_pattern is the pattern to use to specify the Amazon ECS task definitions to discover. This is a regular expression.

      • sd_metrics_ports lists the containerPort for the Prometheus metrics. Separate the containerPorts with semicolons.

      • sd_container_name_pattern specifies the Amazon ECS task container names. This is a regular expression.

      • sd_metrics_path specifies the Prometheus metric path. If you omit this, the agent assumes the default path /metrics

      • sd_job_name specifies the Prometheus scrape job name. If you omit this field, the CloudWatch agent uses the job name in the Prometheus scrape configuration.

  • metric_declaration— are sections that specify the array of logs with embedded metric format to be generated. There are metric_declaration sections for each Prometheus source that the CloudWatch agent imports from by default. These sections each include the following fields:

    • label_matcher is a regular expression that checks the value of the labels listed in source_labels. The metrics that match are enabled for inclusion in the embedded metric format sent to CloudWatch.

      If you have multiple labels specified in source_labels, we recommend that you do not use ^ or $ characters in the regular expression for label_matcher.

    • source_labels specifies the value of the labels that are checked by the label_matcher line.

    • label_separator specifies the separator to be used in the label_matcher line if multiple source_labels are specified. The default is ;. You can see this default used in the label_matcher line in the following example.

    • metric_selectors is a regular expression that specifies the metrics to be collected and sent to CloudWatch.

    • dimensions is the list of labels to be used as CloudWatch dimensions for each selected metric.

See the following metric_declaration example.

"metric_declaration": [
  {
     "source_labels":[ "Service", "Namespace"],
     "label_matcher":"(.*node-exporter.*|.*kube-dns.*);kube-system",
     "dimensions":[
        ["Service", "Namespace"]
     ],
     "metric_selectors":[
        "^coredns_dns_request_type_count_total$"
     ]
  }
]

This example configures an embedded metric format section to be sent as a log event if the following conditions are met:

  • The value of Service contains either node-exporter or kube-dns.

  • The value of Namespace is kube-system.

  • The Prometheus metric coredns_dns_request_type_count_total contains both Service and Namespace labels.

The log event that is sent includes the following highlighted section:

{
   "CloudWatchMetrics":[
      {
         "Metrics":[
            {
               "Name":"coredns_dns_request_type_count_total"
            }
         ],
         "Dimensions":[
            [
               "Namespace",
               "Service"
            ]
         ],
         "Namespace":"ContainerInsights/Prometheus"
      }
   ],
   "Namespace":"kube-system",
   "Service":"kube-dns",
   "coredns_dns_request_type_count_total":2562,
   "eks_amazonaws_com_component":"kube-dns",
   "instance":"192.168.61.254:9153",
   "job":"kubernetes-service-endpoints",
   ...
}

Tutorial for adding a new Prometheus scrape target: Prometheus API Server metrics

The Kubernetes API Server exposes Prometheus metrics on endpoints by default. The official example for the Kubernetes API Server scraping configuration is available on Github.

The following tutorial shows how to do the following steps to begin importing Kubernetes API Server metrics into CloudWatch:

  • Adding the Prometheus scraping configuration for Kubernetes API Server to the CloudWatch agent YAML file.

  • Configuring the embedded metric format metrics definitions in the CloudWatch agent YAML file.

  • (Optional) Creating a CloudWatch dashboard for the Kubernetes API Server metrics.

Note

The Kubernetes API Server exposes gauge, counter, histogram, and summary metrics. In this release of Prometheus metrics support, CloudWatch imports only the metrics with gauge, counter, and summary types.

To start collecting Kubernetes API Server Prometheus metrics in CloudWatch

  1. Download the latest version of the prometheus-eks.yaml, prometheus-eks-fargate.yaml, or prometheus-k8s.yaml file by entering one of the following commands.

    For an Amazon EKS cluster with the EC2 launch type, enter the following command:

    curl -O https://raw.githubusercontent.com/aws-samples/amazon-cloudwatch-container-insights/latest/k8s-deployment-manifest-templates/deployment-mode/service/cwagent-prometheus/prometheus-eks.yaml

    For an Amazon EKS cluster with the Fargate launch type, enter the following command:

    curl -O https://raw.githubusercontent.com/aws-samples/amazon-cloudwatch-container-insights/latest/k8s-deployment-manifest-templates/deployment-mode/service/cwagent-prometheus/prometheus-eks-fargate.yaml

    For a Kubernetes cluster running on an Amazon EC2 instance, enter the following command:

    curl -O https://raw.githubusercontent.com/aws-samples/amazon-cloudwatch-container-insights/latest/k8s-deployment-manifest-templates/deployment-mode/service/cwagent-prometheus/prometheus-k8s.yaml

  2. Open the file with a text editor, find the prometheus-config section, and add the following section inside of that section. Then save the changes:

        # Scrape config for API servers
    - job_name: 'kubernetes-apiservers'
      kubernetes_sd_configs:
        - role: endpoints
          namespaces:
            names:
              - default
      scheme: https
      tls_config:
        ca_file: /var/run/secrets/kubernetes.io/serviceaccount/ca.crt
        insecure_skip_verify: true
      bearer_token_file: /var/run/secrets/kubernetes.io/serviceaccount/token
      relabel_configs:
      - source_labels: [__meta_kubernetes_service_name, __meta_kubernetes_endpoint_port_name]
        action: keep
        regex: kubernetes;https
      - action: replace
        source_labels:
        - __meta_kubernetes_namespace
        target_label: Namespace
      - action: replace
        source_labels:
        - __meta_kubernetes_service_name
        target_label: Service

  3. While you still have the YAML file open in the text editor, find the cwagentconfig.json section. Add the following subsection and save the changes. This section puts the API server metrics onto the CloudWatch agent allow list. Three types of API Server metrics are added to the allow list:

    • etcd object counts

    • API Server registration controller metrics

    • API Server request metrics

    {"source_labels": ["job", "resource"],
  "label_matcher": "^kubernetes-apiservers;(services|daemonsets.apps|deployments.apps|configmaps|endpoints|secrets|serviceaccounts|replicasets.apps)",
  "dimensions": [["ClusterName","Service","resource"]],
  "metric_selectors": [
  "^etcd_object_counts$"
  ]
},
{"source_labels": ["job", "name"],
   "label_matcher": "^kubernetes-apiservers;APIServiceRegistrationController$",
   "dimensions": [["ClusterName","Service","name"]],
   "metric_selectors": [
   "^workqueue_depth$",
   "^workqueue_adds_total$",
   "^workqueue_retries_total$"
  ]
},
{"source_labels": ["job","code"],
  "label_matcher": "^kubernetes-apiservers;2[0-9]{2}$",
  "dimensions": [["ClusterName","Service","code"]],
  "metric_selectors": [
   "^apiserver_request_total$"
  ]
},
{"source_labels": ["job"],
  "label_matcher": "^kubernetes-apiservers",
  "dimensions": [["ClusterName","Service"]],
  "metric_selectors": [
  "^apiserver_request_total$"
  ]
},

  4. If you already have the CloudWatch agent with Prometheus support deployed in the cluster, you must delete it by entering the following command:

    kubectl delete deployment cwagent-prometheus -n amazon-cloudwatch

  5. Deploy the CloudWatch agent with your updated configuration by entering one of the following commands. For an Amazon EKS cluster with the EC2 launch type, enter:

    kubectl apply -f prometheus-eks.yaml

    For an Amazon EKS cluster with the Fargate launch type, enter the following command. Replace MyCluster and region with values to match your deployment.

    cat prometheus-eks-fargate.yaml \
| sed "s/{{cluster_name}}/MyCluster/;s/{{region_name}}/region/" \
| kubectl apply -f -

    For a Kubernetes cluster, enter the following command. Replace MyCluster and region with values to match your deployment.

    cat prometheus-k8s.yaml \
| sed "s/{{cluster_name}}/MyCluster/;s/{{region_name}}/region/" \
| kubectl apply -f -

Once you have done this, you should see a new log stream named kubernetes-apiservers in the /aws/containerinsights/cluster_name/prometheus log group. This log stream should include log events with an embedded metric format definition like the following:

{
   "CloudWatchMetrics":[
      {
         "Metrics":[
            {
               "Name":"apiserver_request_total"
            }
         ],
         "Dimensions":[
            [
               "ClusterName",
               "Service"
            ]
         ],
         "Namespace":"ContainerInsights/Prometheus"
      }
   ],
   "ClusterName":"my-cluster-name",
   "Namespace":"default",
   "Service":"kubernetes",
   "Timestamp":"1592267020339",
   "Version":"0",
   "apiserver_request_count":0,
   "apiserver_request_total":0,
   "code":"0",
   "component":"apiserver",
   "contentType":"application/json",
   "instance":"192.0.2.0:443",
   "job":"kubernetes-apiservers",
   "prom_metric_type":"counter",
   "resource":"pods",
   "scope":"namespace",
   "verb":"WATCH",
   "version":"v1"
}

You can view your metrics in the CloudWatch console in the ContainerInsights/Prometheus namespace. You can also optionally create a CloudWatch dashboard for your Prometheus Kubernetes API Server metrics.

(Optional) Creating a dashboard for Kubernetes API Server metrics

To see Kubernetes API Server metrics in your dashboard, you must have first completed the steps in the previous sections to start collecting these metrics in CloudWatch.

To create a dashboard for Kubernetes API Server metrics

  1. Open the CloudWatch console at https://console.aws.amazon.com/cloudwatch/.

  2. Make sure you have the correct AWS Region selected.

  3. In the navigation pane, choose Dashboards.

  4. Choose Create Dashboard. Enter a name for the new dashboard, and choose Create dashboard.

  5. In Add to this dashboard, choose Cancel.

  6. Choose Actions, View/edit source.

  7. Download the following JSON file: Kubernetes API Dashboard source.

  8. Open the JSON file that you downloaded with a text editor, and make the following changes:

    • Replace all the {{YOUR_CLUSTER_NAME}} strings with the exact name of your cluster. Make sure not to add whitespaces before or after the text.

    • Replace all the {{YOUR_AWS_REGION}} strings with the name of the Region where the metrics are collected. For example us-west-2. Be sure not to add whitespaces before or after the text.

  9. Copy the entire JSON blob and paste it into the text box in the CloudWatch console, replacing what is already in the box.

  10. Choose Update, Save dashboard.