開始使用 OpenTelemetry 收集器

透過集合功能整理內容 你可以依據偏好儲存及分類內容。

gcloud components list

kubectl create ns NAMESPACE_NAME

gcloud config set project PROJECT_ID \
&&
gcloud iam service-accounts create gmp-test-sa \
&&
gcloud iam service-accounts add-iam-policy-binding \
  --role roles/iam.workloadIdentityUser \
  --member "serviceAccount:PROJECT_ID.svc.id.goog[NAMESPACE_NAME/default]" \
  gmp-test-sa@PROJECT_ID.iam.gserviceaccount.com \
&&
kubectl annotate serviceaccount \
  --namespace NAMESPACE_NAME \
  default \
  iam.gke.io/gcp-service-account=gmp-test-sa@PROJECT_ID.iam.gserviceaccount.com

gcloud projects add-iam-policy-binding PROJECT_ID\
  --member=serviceAccount:gmp-test-sa@PROJECT_ID.iam.gserviceaccount.com \
  --role=roles/monitoring.metricWriter

receivers:
  prometheus:
    config:
      scrape_configs:
      - job_name: 'SCRAPE_JOB_NAME'
        kubernetes_sd_configs:
        - role: pod
        relabel_configs:
        - source_labels: [__meta_kubernetes_pod_label_app_kubernetes_io_name]
          action: keep
          regex: prom-example
        - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_path]
          action: replace
          target_label: __metrics_path__
          regex: (.+)
        - source_labels: [__address__, __meta_kubernetes_pod_annotation_prometheus_io_port]
          action: replace
          regex: (.+):(?:\d+);(\d+)
          replacement: $$1:$$2
          target_label: __address__
        - action: labelmap
          regex: __meta_kubernetes_pod_label_(.+)

processors:
  resourcedetection:
    detectors: [gcp]
    timeout: 10s

  transform:
    # "location", "cluster", "namespace", "job", "instance", and "project_id" are reserved, and
    # metrics containing these labels will be rejected.  Prefix them with exported_ to prevent this.
    metric_statements:
    - context: datapoint
      statements:
      - set(attributes["exported_location"], attributes["location"])
      - delete_key(attributes, "location")
      - set(attributes["exported_cluster"], attributes["cluster"])
      - delete_key(attributes, "cluster")
      - set(attributes["exported_namespace"], attributes["namespace"])
      - delete_key(attributes, "namespace")
      - set(attributes["exported_job"], attributes["job"])
      - delete_key(attributes, "job")
      - set(attributes["exported_instance"], attributes["instance"])
      - delete_key(attributes, "instance")
      - set(attributes["exported_project_id"], attributes["project_id"])
      - delete_key(attributes, "project_id")

  batch:
    # batch metrics before sending to reduce API usage
    send_batch_max_size: 200
    send_batch_size: 200
    timeout: 5s

  memory_limiter:
    # drop metrics if memory usage gets too high
    check_interval: 1s
    limit_percentage: 65
    spike_limit_percentage: 20

# Note that the googlemanagedprometheus exporter block is intentionally blank
exporters:
  googlemanagedprometheus:

service:
  pipelines:
    metrics:
      receivers: [prometheus]
      processors: [batch, memory_limiter, resourcedetection, transform]
      exporters: [googlemanagedprometheus]

receivers:
  prometheus:
    config:
      scrape_configs:
      - job_name: 'SCRAPE_JOB_NAME'
        kubernetes_sd_configs:
        - role: pod
        relabel_configs:
        - source_labels: [__meta_kubernetes_pod_label_app_kubernetes_io_name]
          action: keep
          regex: prom-example
        - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_path]
          action: replace
          target_label: __metrics_path__
          regex: (.+)
        - source_labels: [__address__, __meta_kubernetes_pod_annotation_prometheus_io_port]
          action: replace
          regex: (.+):(?:\d+);(\d+)
          replacement: $$1:$$2
          target_label: __address__
        - action: labelmap
          regex: __meta_kubernetes_pod_label_(.+)

processors:
  resource:
    attributes:
    - key: "cluster"
      value: "CLUSTER_NAME"
      action: upsert
    - key: "namespace"
      value: "NAMESPACE_NAME"
      action: upsert
    - key: "location"
      value: "REGION"
      action: upsert

  transform:
    # "location", "cluster", "namespace", "job", "instance", and "project_id" are reserved, and
    # metrics containing these labels will be rejected.  Prefix them with exported_ to prevent this.
    metric_statements:
    - context: datapoint
      statements:
      - set(attributes["exported_location"], attributes["location"])
      - delete_key(attributes, "location")
      - set(attributes["exported_cluster"], attributes["cluster"])
      - delete_key(attributes, "cluster")
      - set(attributes["exported_namespace"], attributes["namespace"])
      - delete_key(attributes, "namespace")
      - set(attributes["exported_job"], attributes["job"])
      - delete_key(attributes, "job")
      - set(attributes["exported_instance"], attributes["instance"])
      - delete_key(attributes, "instance")
      - set(attributes["exported_project_id"], attributes["project_id"])
      - delete_key(attributes, "project_id")

  batch:
    # batch metrics before sending to reduce API usage
    send_batch_max_size: 200
    send_batch_size: 200
    timeout: 5s

  memory_limiter:
    # drop metrics if memory usage gets too high
    check_interval: 1s
    limit_percentage: 65
    spike_limit_percentage: 20

exporters:
  googlemanagedprometheus:
    project: "PROJECT_ID"

service:
  pipelines:
    metrics:
      receivers: [prometheus]
      processors: [batch, memory_limiter, resource, transform]
      exporters: [googlemanagedprometheus]

kubectl -n NAMESPACE_NAME apply -f https://raw.githubusercontent.com/GoogleCloudPlatform/prometheus-engine/v0.15.3/examples/example-app.yaml

kubectl -n NAMESPACE_NAME create configmap otel-config --from-file config.yaml

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: NAMESPACE_NAME:prometheus-test
rules:
- apiGroups: [""]
  resources:
  - pods
  verbs: ["get", "list", "watch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: NAMESPACE_NAME:prometheus-test
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: NAMESPACE_NAME:prometheus-test
subjects:
- kind: ServiceAccount
  namespace: NAMESPACE_NAME
  name: default
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: otel-collector
spec:
  replicas: 1
  selector:
    matchLabels:
      app: otel-collector
  template:
    metadata:
      labels:
        app: otel-collector
    spec:
      containers:
      - name: otel-collector
        image: otel/opentelemetry-collector-contrib:0.106.0
        args:
        - --config
        - /etc/otel/config.yaml
        - --feature-gates=exporter.googlemanagedprometheus.intToDouble
        volumeMounts:
        - mountPath: /etc/otel/
          name: otel-config
      volumes:
      - name: otel-config
        configMap:
          name: otel-config

kubectl -n NAMESPACE_NAME create -f collector-deployment.yaml

receivers:
  prometheus:
    config:
      scrape_configs:
      - job_name: 'kubernetes-pods'
        kubernetes_sd_configs:
        - role: pod
        relabel_configs:
        - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_scrape]
          action: keep
          regex: true
        - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_path]
          action: replace
          target_label: __metrics_path__
          regex: (.+)
        - source_labels: [__address__, __meta_kubernetes_pod_annotation_prometheus_io_port]
          action: replace
          regex: (.+):(?:\d+);(\d+)
          replacement: $$1:$$2
          target_label: __address__
        - action: labelmap
          regex: __meta_kubernetes_pod_label_(.+)

service:
  pipelines:
    metrics:
      receivers: [prometheus]

- source_labels: [__address__, __meta_kubernetes_pod_annotation_prometheus_io_port]
  action: replace
  regex: (.+):(?:\d+);(\d+)
  replacement: $1:$2
  target_label: __address__

- source_labels: [__address__, __meta_kubernetes_pod_annotation_prometheus_io_port]
  action: replace
  regex: (.+):(?:\d+);(\d+)
  replacement: $$1:$$2
  target_label: __address__


For more information, see the OpenTelemetry documentation.


Next, we strongly recommend that you use processors to format your metrics. In
many cases, processors must be used to properly format your metrics.

Add processors

OpenTelemetry
processors
modify telemetry data before it is exported. You can use the
following processors
to ensure that your metrics are written in a format compatible with
Managed Service for Prometheus.

Detect resource attributes

The Managed Service for Prometheus exporter for OpenTelemetry uses the
prometheus_target monitored
resource
to uniquely identify time series data points. The exporter parses the required
monitored-resource fields from resource attributes on the metric data points.
The fields and the attributes from which the values are scraped are:


project_id: auto-detected by Application Default
Credentials,
gcp.project.id, or project in exporter config (see configuring the
exporter)
location: location, cloud.availability_zone, cloud.region
cluster: cluster, k8s.cluster_name
namespace: namespace, k8s.namespace_name
job: service.name + service.namespace
instance: service.instance.id


Failure to set these labels to unique values can result in "duplicate
timeseries" errors when exporting to Managed Service for Prometheus.
In many cases, values can be automatically detected for these labels,
but in some cases, you might have to map them yourself. The rest of this
section describes these scenarios.

The Prometheus receiver automatically sets the service.name attribute
based on the job_name in the scrape config, and service.instance.id
attribute based on the scrape target's instance. The receiver also sets
k8s.namespace.name when using role: pod in the scrape config.

When possible, populate the other attributes automatically by using the
resource detection
processor.
However, depending on your environment, some attributes might not be
automatically detectable. In this case, you can use other processors to either
manually insert these values or parse them from metric labels. The following
sections illustrate configurations for detecting resources on various platforms.

GKE

When running OpenTelemetry on GKE, you need to enable the
resource-detection processor to fill out the resource labels. Be sure that your
metrics don't already contain any of the reserved resource labels. If this is
unavoidable, see Avoid resource attribute collisions by renaming
attributes.

processors:
  resourcedetection:
    detectors: [gcp]
    timeout: 10s


This section can be copied directly into your config file, replacing the
processors section if it already exists.

Amazon EKS

The EKS resource detector does not automatically fill in the cluster or
namespace attributes. You can provide these values manually by using
the resource
processor,
as shown in the following example:

processors:
  resourcedetection:
    detectors: [eks]
    timeout: 10s

  resource:
    attributes:
    - key: "cluster"
      value: "my-eks-cluster"
      action: upsert
    - key: "namespace"
      value: "my-app"
      action: upsert


You can also convert these values from metric labels using the groupbyattrs
processor (see move metric labels to resource labels below).

Azure AKS

The AKS resource detector does not automatically fill in the cluster or
namespace attributes. You can provide these values manually by using the
resource
processor,
as shown in the following example:

processors:
  resourcedetection:
    detectors: [aks]
    timeout: 10s

  resource:
    attributes:
    - key: "cluster"
      value: "my-eks-cluster"
      action: upsert
    - key: "namespace"
      value: "my-app"
      action: upsert


You can also convert these values from metric labels by using the groupbyattrs
processor; see Move metric labels to resource labels.

On-premises and non-cloud environments

With on-premises or non-cloud environments, you probably can't
detect any of the necessary resource attributes automatically. In this case, you
can emit these labels in your metrics and move them to resource attributes (see
Move metric labels to resource labels), or manually set all
of the resource attributes as shown in the following example:

processors:
  resource:
    attributes:
    - key: "cluster"
      value: "my-on-prem-cluster"
      action: upsert
    - key: "namespace"
      value: "my-app"
      action: upsert
    - key: "location"
      value: "us-east-1"
      action: upsert


Create your collector config as a ConfigMap describes how
to use the config. That section assumes you have put your config in a file
called config.yaml.

The project_id resource attribute can still be automatically set when running
the Collector with Application Default
Credentials.
If your Collector does not have access to Application Default Credentials, see
Setting project_id.

Alternatively, you can manually set the resource attributes you need in an
environment variable, OTEL_RESOURCE_ATTRIBUTES, with a comma-separated list of
key-value pairs, for example:

export OTEL_RESOURCE_ATTRIBUTES="cluster=my-cluster,namespace=my-app,location=us-east-1"


Then use the env resource detector
processor
to set the resource attributes:

processors:
  resourcedetection:
    detectors: [env]


Avoid resource attribute collisions by renaming attributes

If your metrics already contain labels that collide with the required
resource attributes (such as location, cluster, or namespace), rename them
to avoid the collision. The Prometheus convention is to add the prefix exported_
to the label name. To add this prefix, use the transform
processor.

The following processors config renames any potential collisions and
resolves any conflicting keys from the metric:

processors:
  transform:
    # "location", "cluster", "namespace", "job", "instance", and "project_id" are reserved, and
    # metrics containing these labels will be rejected.  Prefix them with exported_ to prevent this.
    metric_statements:
    - context: datapoint
      statements:
      - set(attributes["exported_location"], attributes["location"])
      - delete_key(attributes, "location")
      - set(attributes["exported_cluster"], attributes["cluster"])
      - delete_key(attributes, "cluster")
      - set(attributes["exported_namespace"], attributes["namespace"])
      - delete_key(attributes, "namespace")
      - set(attributes["exported_job"], attributes["job"])
      - delete_key(attributes, "job")
      - set(attributes["exported_instance"], attributes["instance"])
      - delete_key(attributes, "instance")
      - set(attributes["exported_project_id"], attributes["project_id"])
      - delete_key(attributes, "project_id")


Move metric labels to resource labels

In some cases, your metrics might be intentionally reporting labels such as
namespace because your exporter is monitoring multiple namespaces. For
example, when running the
kube-state-metrics
exporter.

In this scenario, these labels can be moved to resource attributes using the
groupbyattrs
processor:

processors:
  groupbyattrs:
    keys:
    - namespace
    - cluster
    - location


In the previous example, given a metric with the labels namespace, cluster,
or location, those labels will be converted to the matching resource
attributes.

Limit API requests and memory usage

Two other processors, the batch
processor
and memory limiter
processor
allow you to limit the resource consumption of your collector.

Batch processing

Batching requests lets you define how many data points to send in a single
request. Note that Cloud Monitoring
limit of 200 time series per
request. Enable the batch processor by using the following settings:

processors:
  batch:
    # batch metrics before sending to reduce API usage
    send_batch_max_size: 200
    send_batch_size: 200
    timeout: 5s


Memory limiting

We recommend enabling the memory-limiter processor to prevent your collector
from crashing at times of high throughput. Enable the processing by using
the following settings:

processors:
  memory_limiter:
    # drop metrics if memory usage gets too high
    check_interval: 1s
    limit_percentage: 65
    spike_limit_percentage: 20


Configure the googlemanagedprometheus exporter

By default, using the googlemanagedprometheus exporter on GKE
requires no additional configuration. For many use cases you only need to enable
it with an empty block in the exporters section:

exporters:
  googlemanagedprometheus:


However, the exporter does provide some optional configuration settings. The
following sections describe the other configuration settings.

Setting project_id

To associate your time series with a Google Cloud project, the
prometheus_target monitored resource must have project_id set.

When running OpenTelemetry on Google Cloud, the
Managed Service for Prometheus exporter defaults to setting this value
based on the Application Default
Credentials
it finds. If no credentials are available, or you want to override the default
project, you have two options:


Set project in the exporter config
Add a gcp.project.id resource attribute to your metrics.


We strongly recommend using the default (unset) value for project_id rather
than explicitly setting it, when possible.

Set project in the exporter config



The following config excerpt sends metrics to
Managed Service for Prometheus in the Google Cloud project MY_PROJECT:

receivers:
  prometheus:
    config:
    ...

processors:
  resourcedetection:
    detectors: [gcp]
    timeout: 10s

exporters:
  googlemanagedprometheus:
    project: MY_PROJECT

service:
  pipelines:
    metrics:
      receivers: [prometheus]
      processors: [resourcedetection]
      exporters: [googlemanagedprometheus]


The only change from previous examples is the new line project: MY_PROJECT.
This setting is useful if you know that every metric coming through this
Collector should be sent to MY_PROJECT.

Set gcp.project.id resource attribute

You can set project association on a per-metric basis by adding a
gcp.project.id resource attribute to your metrics. Set the value of the
attribute to the name of the project the metric should be associated with.

For example, if your metric already has a label project, this label can be
moved to a resource attribute and renamed to gcp.project.id by using
processors in the Collector config, as shown in the following example:

receivers:
  prometheus:
    config:
    ...

processors:
  resourcedetection:
    detectors: [gcp]
    timeout: 10s

  groupbyattrs:
    keys:
    - project

  resource:
    attributes:
    - key: "gcp.project.id"
      from_attribute: "project"
      action: upsert

exporters:
  googlemanagedprometheus:

service:
  pipelines:
    metrics:
      receivers: [prometheus]
      processors: [resourcedetection, groupbyattrs, resource]
      exporters: [googlemanagedprometheus]


Setting client options

The googlemanagedprometheus exporter uses gRPC clients for
Managed Service for Prometheus. Therefore, optional settings
are available for configuring the gRPC client:


compression: Enables gzip compression for gRPC requests, which is useful for
minimizing data transfer fees when sending data from other clouds to
Managed Service for Prometheus (valid values: gzip).
user_agent: Overrides the user-agent string sent on requests to
Cloud Monitoring; only applies to metrics.
Defaults to the build and version number of your OpenTelemetry Collector,
for example, opentelemetry-collector-contrib 0.106.0.
endpoint: Sets the endpoint to which metric data is going to be sent.
use_insecure: If true, uses gRPC as the communication transport. Has an
effect only when the endpoint value is not "".
grpc_pool_size: Sets the size of the connection pool in the gRPC client.
prefix: Configures the prefix of metrics sent to
Managed Service for Prometheus. Defaults to
prometheus.googleapis.com.
Don't change this prefix; doing so causes metrics to not be
queryable with PromQL in the Cloud Monitoring UI.


In most cases, you don't need to change these values from their
defaults. However, you can change them to accommodate special
circumstances.

All of these settings are set under a metric block in the
googlemanagedprometheus exporter section, as shown in the following example:

receivers:
  prometheus:
    config:
    ...

processors:
  resourcedetection:
    detectors: [gcp]
    timeout: 10s

exporters:
  googlemanagedprometheus:
    metric:
      compression: gzip
      user_agent: opentelemetry-collector-contrib 0.106.0
      endpoint: ""
      use_insecure: false
      grpc_pool_size: 1
      prefix: prometheus.googleapis.com

service:
  pipelines:
    metrics:
      receivers: [prometheus]
      processors: [resourcedetection]
      exporters: [googlemanagedprometheus]


What's next




Use PromQL in Cloud Monitoring to query Prometheus metrics.
Use Grafana to query Prometheus metrics.
Set up the OpenTelemetry Collector as a sidecar agent in Cloud Run.















































































































































The Cloud Monitoring Metrics Management page provides information
that can help you control the amount you spend on billable metrics
without affecting observability. The Metrics Management page reports the
following information:

  
Ingestion volumes for both byte- and sample-based billing, across metric
    domains and for individual metrics.
  
Data about labels and cardinality of metrics.
  
Number of reads for each metric.
  
Use of metrics in alerting policies and custom dashboards.
  
Rate of metric-write errors.

You can also use the Metrics Management page to
exclude unneeded metrics,
eliminating the cost of ingesting them.






For more information about the Metrics Management page, see
View and manage metric usage.