Ignoring Unwanted Resources in APM

이 페이지는 아직 한국어로 제공되지 않으며 번역 작업 중입니다. 번역에 관한 질문이나 의견이 있으시면 언제든지 저희에게 연락해 주십시오.

A service can handle a variety of requests, some of which you might not want traced or included in trace metrics. An example of this is, possibly, health checks in a web application.

There are two ways to specify that such an endpoint should be untraced and excluded from trace metrics:

Note: Filtering traces using any of the following options removes these requests from trace metrics. For information on how to reduce ingestion without affecting the trace metrics, see ingestion controls.

If you need assistance, contact Datadog support.

Trace Agent configuration options

The Trace Agent component within the Datadog Agent has two methods to prevent certain traces from coming through: ignoring span tags or ignoring resources. If traces are dropped due to these settings, the trace metrics exclude these requests.

Configuring the Trace Agent to ignore certain spans or resources applies to all services that send traces to this particular Datadog Agent. If you have application-specific requirements, use the Tracer configuration method instead.

Ignoring based on span tags

Starting with Datadog Agent 6.27.0/7.27.0, the filter tags option drops traces with root spans that match specified span tags. This option applies to all services that send traces to this particular Datadog Agent. Traces that are dropped because of filter tags are not included in trace metrics.

If you can programmatically identify a set of traces that you know you don’t want to send to Datadog, and no other option in this guide solves your requirement, you can consider adding a custom span tag so you can drop the traces. Reach out to Support to discuss your use case further so Datadog can continue to expand this functionality.

The filter tags option requires an exact string match. If your use case requires ignoring by regex, see Ignoring based on resources.

You can specify span tags to require or reject by using a list of keys and values separated by spaces in environment variables:

DD_APM_FILTER_TAGS_REQUIRE
Collects only traces that have root spans with an exact match for the specified span tags and values. If it does not match this rule, the trace is dropped. For example, DD_APM_FILTER_TAGS_REQUIRE="key1:value1 key2:value2". In Datadog Agent 7.49+, regular expressions can be provided with DD_APM_FILTER_TAGS_REGEX_REQUIRE.
DD_APM_FILTER_TAGS_REJECT
Rejects traces that have root spans with an exact match for the specified span tags and values. If it matches this rule, the trace is dropped. For example, DD_APM_FILTER_TAGS_REJECT="key1:value1 key2:value2". In Datadog Agent 7.49+, regular expressions can be provided with DD_APM_FILTER_TAGS_REGEX_REJECT.

Alternatively, you can set them in the Agent configuration with a comma-separated list:

datadog.yaml

apm_config:
  filter_tags:
    require: ["db:sql", "db.instance:mysql"]
    reject: ["outcome:success", "key2:value2"]

For example, to ignore health checks where the http.url matches this endpoint:

datadog.yaml

apm_config:
  filter_tags:
    reject: ["http.url:http://localhost:5050/healthcheck"]

Datadog Operator

datadog-agent.yaml

apiVersion: datadoghq.com/v2alpha1
kind: DatadogAgent
metadata:
  name: datadog
spec:
  override:
    nodeAgent:
      containers:
        trace-agent:
          env:
            - name: DD_APM_FILTER_TAGS_REJECT
              value: tag_key1:tag_val2 tag_key2:tag_val2

After making your changes, apply the new configuration by using the following command:

kubectl apply -n $DD_NAMESPACE -f datadog-agent.yaml

Helm

datadog-values.yaml

agents:
  containers:
    traceAgent:
      env:
        - name: DD_APM_FILTER_TAGS_REJECT
          value: tag_key1:tag_val2 tag_key2:tag_val2

After making your changes, upgrade your Datadog Helm chart using the following command:

helm upgrade -f datadog-values.yaml <RELEASE NAME> datadog/datadog

Filtering traces this way removes these requests from trace metrics. For more information on how to reduce ingestion without affecting the trace metrics, see Ingestion Controls.

On the backend, Datadog creates and adds the following span tags to spans after ingestion. These tags cannot be used to drop traces at the Datadog Agent level.

NameDescription
http.path_groupThe full URL path from the http.url tag.
http.url_details.hostThe host name portion of the http.url tag.
http.url_details.pathThe full request target as passed in an HTTP request line or equivalent.
http.url_details.schemeThe request scheme from the http.url tag.
http.url_details.queryStringThe query string portion from the http.url tag.
http.url_details.portThe HTTP port from the http.url tag.
http.useragent_details.os.familyThe OS family reported by the User-Agent.
http.useragent_details.browser.familyThe browser family reported by the User-Agent.
http.useragent_details.device.familyThe device family reported by the User-Agent.
Note: Starting from October 1st 2022, Datadog backend applies a remapping in order to apply Span Tags Semantics across tracers on all ingested spans. If you want to drop spans based on tags at the Datadog Agent level, use tags in the Remap from column.

Network communications

NameRemap from
network.host.iptcp.local.address - Node.js
network.destination.ipout.host - All languages
network.destination.portgrpc.port - Python
tcp.remote.port - Node.js
out.port - All languages

HTTP requests

NameRemap from
http.routeaspnet_core.route - .NET
aspnet.route - .NET
laravel.route - PHP
symfony.route - PHP
http.useragentuser_agent - Java
http.url_details.queryStringhttp.query.string - Python

Database

NameRemap from
db.systemdb.type - Java, Python, Node.js, Go
active_record.db.vendor - Ruby
sequel.db.vendor - Ruby
db.instancemongodb.db - Python
sql.db - Python
db.name - All languages
db.statementcassandra.query - Go
consul.command - Python
memcached.query - Python
mongodb.query - Python, .NET, Go
redis.command - Python
redis.raw_command - Python
sql.query - Python, PHP, Node.js, Java
db.row_countcassandra.row_count - Python
db.rowcount - Python, PHP
mongodb.rows - Python
sql.rows - Python
db.cassandra.clustercassandra.cluster - Python, Go
db.cassandra.consistency_levelcassandra.consistency_level - Python, Go
db.cassandra.tablecassandra.keyspace - Python, Go
db.redis.database_indexdb.redis.dbIndex - Java
out.redis_db - Python, Ruby
db.mongodb.collectionmongodb.collection - Python, .NET, Ruby, PHP
db.cosmosdb.containercosmosdb.container - .NET

Message Queue

NameRemap from
messaging.destinationamqp.destination - Node.js
amqp.queue - .NET
msmq.queue.path - .NET
aws.queue.name - .NET
messaging.urlaws.queue.url - .NET, Java
messaging.message_idserver_id - Go
messaging.message_payload_sizemessage.size - .NET, Java
messaging.operationamqp.command - .NET
msmq.command - .NET
messaging.rabbitmq.routing_keyamqp.routing_key - Java
amqp.routingKey - Nodes.js
messaging.rabbitmq.delivery_modemessaging.rabbitmq.exchange - .NET
messaging.msmq.message.transactionalmsmq.message.transactional - .NET
messaging.msmq.queue.transactionalmsmq.queue.transactional - .NET
messaging.kafka.consumer_groupkafka.group - Java
messaging.kafka.tombstonekafka.tombstone - .NET
tombstone - Java
messaging.kafka.partitionkafka.partition - .NET
partition - Node.js, Go, Java
messaging.kafka.offsetkafka.offset - .NET
messaging.msmq.message.transactionalmsmq.message.transactional - .NET

Remote procedure calls

NameRemap from
rpc.servicegrpc.method.service - Python, .NET
rpc.methodgrpc.method.name - Python, .NET, Go
rpc.grpc.packagegrpc.method.package - Python, .NET, Go
rpc.grpc.status_codegrpc.code - Go
status.code - Python, .NET, Node.js
grpc.status.code - Python, .NET, Node.js
rpc.grpc.kindgrpc.method.kind - Python, Node.js, Go, .NET
rpc.grpc.pathrpc.grpc.path - Python, Node.js, Go, .NET
rpc.grpc.request.metadata.*grpc.request.metadata.* - Python, Node.js
rpc.grpc.request.metadata - Go
rpc.grpc.response.metadata.*grpc.response.metadata.* - Python, Node.js

Errors

NameRemap from
error.messageerror.msg - All languages

Ignoring based on resources

The ignore resources option allows resources to be excluded if the global root span of the trace matches certain criteria. See Exclude resources from being collected. This option applies to all services that send traces to this particular Datadog Agent. Traces that are dropped because of ignore resources are not included in trace metrics.

You can specify resources to ignore either in the Agent configuration file, datadog.yaml, or with the DD_APM_IGNORE_RESOURCES environment variable. See examples below.

datadog.yaml

apm_config:
## @param ignore_resources - list of strings - optional
## A list of regular expressions can be provided to exclude certain traces based on their resource name.
## All entries must be surrounded by double quotes and separated by commas.

  ignore_resources: ["(GET|POST) /healthcheck","API::NotesController#index"]

Notes:

  • The regex syntax that the Trace Agent accepts is evaluated by Go’s regexp.
  • Depending on your deployment strategy, you may have to adjust the regex by escaping special characters.
  • If you use dedicated containers with Kubernetes, make sure that the environment variable for the ignore resource option is being applied to the trace-agent container.

Example

Consider a trace that contains calls to /api/healthcheck that you don’t want traces from:

Flame graph of a resource you want the tracer to ignore

Take note of the resource name of the global root span.

  • Operation name: rack.request
  • Resource name: Api::HealthchecksController#index
  • Http.url: /api/healthcheck

To use the ignore resource option correctly, the regex rule written must match with the resource name, Api::HealthchecksController#index. A few regex options are possible, but to filter out traces from this resource exactly as is, a potential regex to use is Api::HealthchecksController#index$.

Depending on how you deploy, the syntax looks a little different:

datadog.yaml

apm_config:
  ignore_resources: Api::HealthchecksController#index$

For multiple values:

apm_config:
  ignore_resources: ["value1","Api::HealthchecksController#index$"]

In the Datadog Agent container’s list of environment variables, add DD_APM_IGNORE_RESOURCES with a pattern like the example below. Docker Compose has its own variable substitution to consider when you use special characters like $.

    environment:
      // other Datadog Agent environment variables
      - DD_APM_IGNORE_RESOURCES=Api::HealthchecksController#index$$

For multiple values:

    environment:
      // other Datadog Agent environment variables
      - DD_APM_IGNORE_RESOURCES="value1","Api::HealthchecksController#index$$"

In your docker run command to spin up the Datadog Agent, add DD_APM_IGNORE_RESOURCES:

docker run -d --name datadog-agent \
              --cgroupns host \
              --pid host \
              -v /var/run/docker.sock:/var/run/docker.sock:ro \
              -v /proc/:/host/proc/:ro \
              -v /sys/fs/cgroup/:/host/sys/fs/cgroup:ro \
              -e DD_API_KEY=<> \
              -e DD_APM_IGNORE_RESOURCES="Api::HealthchecksController#index$" \
              -e DD_APM_ENABLED=true \
              -e DD_APM_NON_LOCAL_TRAFFIC=true \
              gcr.io/datadoghq/agent:latest

For multiple values:

              -e DD_APM_IGNORE_RESOURCES=["value1","Api::HealthchecksController#index$"] \

In the dedicated trace-agent container, add the environment variable DD_APM_IGNORE_RESOURCES:

    - name: trace-agent
        image: "gcr.io/datadoghq/agent:latest"
        imagePullPolicy: IfNotPresent
        command: ["trace-agent", "-config=/etc/datadog-agent/datadog.yaml"]
        resources: {}
        ports:
        - containerPort: 8126
          hostPort: 8126
          name: traceport
          protocol: TCP
        env:
        - name: DD_API_KEY
          valueFrom:
            secretKeyRef:
              name: "datadog-secret"
              key: api-key
        - name: DD_KUBERNETES_KUBELET_HOST
          valueFrom:
            fieldRef:
              fieldPath: status.hostIP
        - name: KUBERNETES
          value: "yes"
        - name: DOCKER_HOST
          value: unix:///host/var/run/docker.sock
        - name: DD_LOG_LEVEL
          value: "INFO"
        - name: DD_APM_ENABLED
          value: "true"
        - name: DD_APM_NON_LOCAL_TRAFFIC
          value: "true"
        - name: DD_APM_RECEIVER_PORT
          value: "8126"
        - name: DD_KUBELET_TLS_VERIFY
          value: "false"
        - name: DD_APM_IGNORE_RESOURCES
          value: "Api::HealthchecksController#index$"

For multiple values:

        - name: DD_APM_IGNORE_RESOURCES
          value: '"value1","Api::HealthchecksController#index$"'

In the traceAgent section of the values.yaml file, add DD_APM_IGNORE_RESOURCES in the env section, then spin up helm as usual.

values.yaml

    traceAgent:
      # agents.containers.traceAgent.env -- Additional environment variables for the trace-agent container
      env:
        - name: DD_APM_IGNORE_RESOURCES
          value: Api::HealthchecksController#index$

For multiple values:

        - name: DD_APM_IGNORE_RESOURCES
          value: value1, Api::HealthchecksController#index$

Alternatively, you can set agents.containers.traceAgent.env in the helm install command:

helm install dd-agent -f values.yaml \
  --set datadog.apiKeyExistingSecret="datadog-secret" \
  --set agents.containers.traceAgent.env[0].name=DD_APM_IGNORE_RESOURCES, \
    agents.containers.traceAgent.env[0].value="Api::HealthchecksController#index$" \
  datadog/datadog

If you use Amazon ECS (such as on EC2), in your Datadog Agent container definition, add the environment variable DD_APM_IGNORE_RESOURCES with the values such that the JSON evaluates to something like this:

    "environment": [
	// other environment variables for the Datadog Agent
        {
          "name": "DD_APM_IGNORE_RESOURCES",
          "value": "Api::HealthchecksController#index$"
        }
     ]
Note: Filtering traces this way removes these requests from trace metrics. For information on how to reduce ingestion without affecting the trace metrics, see ingestion controls.

Tracer configuration options

Some of the language-specific tracers have an option to modify spans before they are sent to the Datadog Agent. Use this option if you have application-specific requirements and are using a language listed below.

Important: If the request is associated with a distributed trace, the resulting trace can have sampling inaccuracy if you drop portions of it due to these filtering rules.

The Ruby tracer has a post-processing pipeline that removes traces that meet certain criteria. More information and examples can be found in Post-processing traces.

For example, if the resource name is Api::HealthchecksController#index, use the Datadog::Tracing::Pipeline::SpanFilter class to remove traces that contain the resource name. This filter can also be used to match on other metadata available for the span object.

Datadog::Tracing.before_flush(
   Datadog::Tracing::Pipeline::SpanFilter.new { |span| span.resource =~ /Api::HealthchecksController#index/ }
)

The Python tracer has a FilterRequestsOnUrl filter you can configure to remove traces from certain endpoints. Alternatively, you can write a custom filter. See Trace Filtering for more information.

Suppose the root span’s http.url span tag has a value of http://<domain>/healthcheck. Use the following regex to match against any endpoint ending in healthcheck:

from ddtrace import tracer
from ddtrace.filters import FilterRequestsOnUrl
tracer.configure(settings={
    'FILTERS': [
        FilterRequestsOnUrl(r'http://.*/healthcheck$'),
    ],
})

Configure a blocklist on the Http plugin. Take note of what the blocklist matches on from the API docs. For example, incoming Http requests matches on URL paths, so if the trace’s http.url span tag is http://<domain>/healthcheck, write a rule that matches the healthcheck URL:

const tracer = require('dd-trace').init();
tracer.use('http', {
  // incoming http requests match on the path
  server: {
    blocklist: ['/healthcheck']
  },
  // outgoing http requests match on a full URL
  client: {
    blocklist: ['https://telemetry.example.org/api/v1/record']
  }
})

//import http
Note: The tracer configuration for the integration must come before that instrumented module is imported.

The Java tracer has an option for a custom TraceInterceptor to filter out certain spans. See Extending Tracers.

For example, if your resource name is GET /healthcheck, write a trace interceptor that drops traces containing this resource name. Adjust the logic to meet your use case.

public class GreetingController {
   static {
       // In a class static block to avoid initializing multiple times.
       GlobalTracer.get().addTraceInterceptor(new TraceInterceptor() {
           @Override
           public Collection<? extends MutableSpan> onTraceComplete(Collection<? extends MutableSpan> trace) {
               for (MutableSpan span : trace) {
                   if ("GET /healthcheck".contentEquals(span.getResourceName())) {
                       return Collections.emptyList();
                   }
               }
               return trace;
           }
           @Override
           public int priority() {
               return 200;  // Some unique number
           }
       });
   }
}
Note: Filtering traces this way removes these requests from trace metrics. For information on how to reduce ingestion without affecting the trace metrics, see ingestion controls.