Set up the Prometheus agent

Configure the Prometheus Agent

You need to place the examples below in the config section of the agent. Refer to the installation method you used to known which is in your case.

To configure the agent using Helm, you should set up your values.yaml in one of these two ways:

Dica

You need to add your New Relic in the values.yaml file.

global:
  licenseKey: YOUR_NEW_RELIC_LICENSE_KEY
  cluster: K8S_CLUSTER_NAME

newrelic-prometheus-agent:
  enabled: true
  config:
    # YOUR CONFIGURATION GOES HERE

Define which endpoints to scrape

By default, the New Relic Prometheus agent uses two annotations to discover targets: newrelic.io/scrape: "true" and prometheus.io/scrape: "true".

All the targets in the cluster that are annotated with newrelic.io/scrape: "true" are discovered and scraped by default. The targets annotated with prometheus.io/scrape: "true" will be scraped or not depending on the configuration.

Scrape metrics only from Prometheus integrations

The Prometheus agent can be configured to scrape metrics from the most popular integrations in Kubernetes. Have a look at the list of integrations containing a set of dashboards and to start monitoring out of the box.

That list is defined in the values.yaml of New Relic's Prometheus agent helm chart. You can modify this list, but some might not work out of the box with custom labels or values.

When upgrading, new integration filters might be available. Therefore, the amount of data scraped could increase, depending on the services in your cluster, after an upgrade involving integration filters. You can avoid this by saving a fixed list of app_values in your values.yaml file. For example:

app_values: ["redis", "traefik", "calico", "nginx", "coredns", "kube-dns", "etcd", "cockroachdb"]

Moreover, it might happen that a new version of the integrations filters may cause a target that was already scraped by one job to be scraped a second time. In order to receive a notification in the event of duplicated data (and prevent duplicate scraping altogether), you can create an alert based on the following query:

FROM Metric SELECT uniqueCount(job) FACET instance, cluster_name LIMIT 10 SINCE 2 minutes ago

If any value is different from 1 then you have two or more jobs scraping the same instance in the same cluster.

Scrape metrics from all the targets

If you need to scrape all the targets annotated with prometheus.io/scrape: "true", you need to perform one of the following actions, depending on the installation method you chose:

If you used the guided install, select the Scrape all Prometheus endpoints option.
If you used Helm, add the following values in your prometheus-agent configuration:
```
kubernetes:
  integrations_filter:
    enabled: false
```

Kubernetes target discovery

Kubernetes jobs discover targets and scrape targets according to the target_discovery configuration. If inside the target_discovery configuration, you set pod and endpoints toggles to true, Prometheus will create rules to discover any pod or endpoint in the cluster having an exposed port.

Use the target_discovery.filter configuration parameter to filter in the targets that Prometheus scrapes. Use the label and annotations labels to filter by current conditions, and the AND operator for all conditions.

The following example only scrapes Pods and Endpoints with the newrelic.io/scrape: "true" annotation, and the k8s.io/app label with postgres or mysql as values. For the endpoints, the annotation must be present in the service related to it. The regexes are anchored, that is, if you configure scrape: 'true', Prometheus evaluates true as ^true$. To avoid that, use .*true.* so it also matches a-true-example.

kubernetes:
  jobs:
    - job_name_prefix: example
      integrations_filter:
        enabled: false
      target_discovery:
        pod: true
        endpoints: true
        filter:
          annotations:
            # <string>: <regex>
            newrelic.io/scrape: "true"
          label:
            # <string>: <regex>
            k8s.io/app: "(postgres|mysql)"

Dica

If you don't add a value for the label or annotation, the filter will only check if it exists.

Setup static targets

The prometheus agent defines a static target job to scrape self-metrics by default, but you can set up additional static targets by including additional jobs.

This following example includes an additional job to scrape a server managed separately, and the self-metrics job to keep reporting the prometheus agent metrics, as defined by default.

static_targets:
  jobs:
    - job_name: managed-exporter
      targets:
        - "managed_exporter.your-company.tld:5432"
    - job_name: self-metrics
      skip_sharding: true  # sharding is skipped to obtain self-metrics from all Prometheus servers.
      targets:
        - "localhost:9090"
      extra_metric_relabel_config:
        - source_labels: [__name__]
          action: keep
          regex: "\
            prometheus_agent_active_series|\
            prometheus_target_interval_length_seconds|\
            prometheus_target_scrape_pool_targets|\
            prometheus_remote_storage_samples_pending|\
            prometheus_remote_storage_samples_in_total|\
            prometheus_remote_storage_samples_retried_total|\
            prometheus_agent_corruptions_total|\
            prometheus_remote_storage_shards|\
            prometheus_sd_kubernetes_events_total|\
            prometheus_agent_checkpoint_creations_failed_total|\
            prometheus_agent_checkpoint_deletions_failed_total|\
            prometheus_remote_storage_samples_dropped_total|\
            prometheus_remote_storage_samples_failed_total|\
            prometheus_sd_kubernetes_http_request_total|\
            prometheus_agent_truncate_duration_seconds_sum|\
            prometheus_build_info|\
            process_resident_memory_bytes|\
            process_virtual_memory_bytes|\
            process_cpu_seconds_total"

Cuidado

If you modify the static_targets section and don't include self-metrics job, the agent metrics aren't reported.

Target scrape interval

By default, the Prometheus agent scrapes all targets for metrics every 30 seconds as defined in common.scrape_interval for all scraping jobs in the configuration. You can change this by using the scrape_interval key in that section.

This example shows two Kubernetes jobs with different scrape intervals:

common:
  scrape_interval: 30s
kubernetes:
  jobs:
    # this job will use the default scrape_interval defined in common.
    - job_name_prefix: default-targets-with-30s-interval
      target_discovery:
        pod: true
        filter:
          annotations:
            newrelic.io/scrape: "true"

    - job_name_prefix: slow-targets-with-60s-interval
      scrape_interval: 60s
      target_discovery:
        pod: true
        filter:
          annotations:
            newrelic.io/scrape_slow: "true"

Metric and label transformations

You can apply metric and label transformations at any configuration section, but setting it at remote write level makes the filtering or transformations wider.

If you set them at newrelic_remote_write level, the filters apply to all metrics that are being sent to New Relic. If you set them at any other section, they apply to the metrics scraped by that section. The metric filter process happens after the metrics have been scraped from the targets.

You can use the extra_metric_relabel_config parameter to apply the filters, which adds entries of the metric_relabel_config parameter. This parameter is present at static_targets.jobs, kubernetes.jobs, and the extra_write_relabel_configs parameter for newrelic_remote_write.

Here's an example of how to use it in different parts of the YAML configuration file:

static_targets:
  - name: self-metrics
    urls:
      - "http://static-service:8181"
    extra_metric_relabel_config:
      # Drop metrics with prefix 'go_' for this target.
      - source_labels: [__name__]
        regex: "go_.+"
        action: drop

newrelic_remote_write:
  extra_write_relabel_configs:
    # Drop all metrics with the specified name before sent to New Relic.
    - source_labels: [__name__]
      regex: "metric_name"
      action: drop

YAML file snippet samples

Add one of these examples in the YAML configuration file from the metric and label transformations section.

To filter metrics

To add or remove metric labels

Importante

Metric Labels names must comply with Prometheus DataModel.

Target authorization configuration

Some targets need access authorization to be scraped, like No-SQL Databases to fetch data that the user connecting has access, or exporters that expose sensible data in their metrics endpoint. All authorization methods supported by Prometheus can be configured in the static_targets and kubernetes sections.

As explained in the Installation guide we create a configuration for Prometheus based on our YAML. This part of the configuration passed to Prometheus as-is from our YAML, so you would refer to Prometheus documentation:

Here are some examples to deal with targets that need access authorization:

kubernetes:
  jobs:
    - job_name_prefix: skip-verify-on-https-targets
      target_discovery:
        pod: true
        filter:
          annotations:
            newrelic.io/scrape: "true"
    - job_name_prefix: bearer-token
      target_discovery:
        pod: true
        filter:
          label:
            k8s.io/app: my-app-with-token
      authorization:
        type: Bearer
        credentials_file: "/etc/my-app/token"

static_targets:
  jobs:
    - job_name: mtls-target
      scheme: https
      targets:
        - "my-mtls-target:8181"
      tls_config:
        ca_file: "/etc/my-app/client-ca.crt"
        cert_file: "/etc/my-app/client.crt"
        key_file: "/etc/my-app/client.key"

    - job_name: basic-auth-target
      targets:
        - "my-basic-auth-static:8181"
      basic_auth:
        password_file: "/etc/my-app/pass.htpasswd"

Configure the Prometheus Agent

Dica

Installed using our Helm bundle

Standalone Helm release

Define which endpoints to scrape

Scrape metrics only from Prometheus integrations

Scrape metrics from all the targets

Kubernetes target discovery

Dica

Setup static targets

Cuidado

Target scrape interval

Metric and label transformations

YAML file snippet samples

To filter metrics

Filters out metrics whose name start with 'prefix_'

Filters out metrics having a 'k8s.io/app=appLabelValue' Kubernetes label

Filters out metrics whose name start with 'prefix_' and that have a 'k8s.io/app=appLabelValue' Kubernetes label

Filters out all metrics except the ones whose name start with 'prefix_'

To add or remove metric labels

Importante

Add the 'new_label=newLabelValue' labels to metrics names starting with 'prefix_'

Filter out all metrics the label 'label_name'

Target authorization configuration

Set up the Prometheus agent

Configure the Prometheus Agent .css-21sua1{background:none;border:none;width:0;padding:0;}

Dica

Standalone Helm release

Define which endpoints to scrape

Scrape metrics only from Prometheus integrations

Scrape metrics from all the targets

Kubernetes target discovery

Dica

Setup static targets

Cuidado

Target scrape interval

Metric and label transformations

YAML file snippet samples

To filter metrics

Filters out metrics whose name start with 'prefix_'

Filters out metrics having a 'k8s.io/app=appLabelValue' Kubernetes label

Filters out metrics whose name start with 'prefix_' and that have a 'k8s.io/app=appLabelValue' Kubernetes label

Filters out all metrics except the ones whose name start with 'prefix_'

To add or remove metric labels

Importante

Add the 'new_label=newLabelValue' labels to metrics names starting with 'prefix_'

Filter out all metrics the label 'label_name'

Target authorization configuration

Configure the Prometheus Agent