datadog_lambda 2.14.0

The Datadog AWS Lambda Layer

datadog-lambda-layer-python

Datadog Lambda Layer for Python (2.7, 3.6, 3.7 and 3.8) enables custom metric submission from AWS Lambda functions, and distributed tracing between serverful and serverless environments.

IMPORTANT NOTE

AWS Lambda is expected to recieve a breaking change on March 30th, 2020. If you are using Datadog Python Lambda layer version 7 or below, please upgrade to version 11.

Installation

Datadog Lambda Layer can be added to a Lambda function via AWS Lambda console, AWS CLI or Serverless Framework using the following ARN.

arn:aws:lambda:<AWS_REGION>:464622532012:layer:Datadog-<PYTHON_RUNTIME>:<VERSION>

Replace <AWS_REGION> with the AWS region where your Lambda function is published to. Replace <PYTHON_RUNTIME> with one of the following that matches your Lambda's Python runtime:

• Datadog-Python27
• Datadog-Python36
• Datadog-Python37
• Datadog-Python38

Replace <VERSION> with the latest layer version that can be found from CHANGELOG. For example,

arn:aws:lambda:us-east-1:464622532012:layer:Datadog-Python37:1

PyPI

When developing your Lambda function locally where AWS Layer doesn't work, the Datadog Lambda layer can be installed from PyPI by pip install datadog-lambda or adding datadog-lambda to your project's requirements.txt.

The minor version of the datadog-lambda package always match the layer version. E.g., datadog-lambda v0.5.0 matches the content in layer version 5.

The Serverless Framework

The Datadog Serverless Framework Plugin makes it easy to manage the Datadog instrumentation for all of your Lambda functions in one place.

Instead of the plugin, you can also use the sample serverless.yml below as a reference for manually including the Lambda Layer, enable AWS X-Ray tracing, and set up environment variables.

provider:
  name: aws
  runtime: python3.7
  tracing:
    lambda: true
    apiGateway: true

functions:
  hello:
    handler: handler.hello
    events:
      - http:
          path: hello
          method: get
    layers:
      - arn:aws:lambda:<AWS_REGION>:464622532012:layer:Datadog-<PYTHON_RUNTIME>:<VERSION>
    environment:
      DD_FLUSH_TO_LOG: true

Environment Variables

DD_FLUSH_TO_LOG

Set to true (recommended) to send custom metrics asynchronously (with no added latency to your Lambda function executions) through CloudWatch Logs with the help of Datadog Forwarder.

If set to false, you also need to set DD_API_KEY and DD_SITE.

DD_API_KEY

If DD_FLUSH_TO_LOG is set to false (not recommended), the Datadog API Key must be defined as one of the following environment variables:

• DD_API_KEY - the Datadog API Key in plain-text, NOT recommended
• DD_KMS_API_KEY - the KMS-encrypted API Key, requires the kms:Decrypt permission
• DD_API_KEY_SECRET_ARN - the Secret ARN to fetch API Key from the Secrets Manager, requires the secretsmanager:GetSecretValue permission (and kms:Decrypt if using a customer managed CMK)

You can also supply or override the API key at runtime (not recommended):

# Override DD API Key after importing datadog_lambda packages
from datadog import api
api._api_key = "MY_API_KEY"

DD_SITE

If DD_FLUSH_TO_LOG is set to false (not recommended), and your data need to be sent to the Datadog EU site, you must set DD_SITE to datadoghq.eu.

DD_LOGS_INJECTION

Inject Datadog trace id into logs for correlation. Defaults to true.

DD_LOG_LEVEL

Set to debug enable debug los from the Datadog Lambda Layer.

DD_ENHANCED_METRICS

Generate enhanced Datadog Lambda integration metrics, such as, aws.lambda.enhanced.invocations and aws.lambda.enhanced.errors. Defaults to true.

Basic Usage

import requests
from datadog_lambda.wrapper import datadog_lambda_wrapper
from datadog_lambda.metric import lambda_metric

@datadog_lambda_wrapper
def lambda_handler(event, context):
    lambda_metric("my_metric", 10, tags=['tag:value'])
    requests.get("https://www.datadoghq.com")

Custom Metrics

Custom metrics can be submitted using lambda_metric and the Lambda handler function needs to be decorated with @datadog_lambda_wrapper. The metrics are submitted as distribution metrics.

IMPORTANT NOTE: If you have already been submitting the same custom metric as non-distribution metric (e.g., gauge, count, or histogram) without using the Datadog Lambda Layer, you MUST pick a new metric name to use for lambda_metric. Otherwise that existing metric will be converted to a distribution metric and the historical data prior to the conversion will be no longer queryable.

from datadog_lambda.wrapper import datadog_lambda_wrapper
from datadog_lambda.metric import lambda_metric

@datadog_lambda_wrapper
def lambda_handler(event, context):
    lambda_metric(
        "coffee_house.order_value",  # metric
        12.45,  # value
        tags=['product:latte', 'order:online']  # tags
    )

VPC

If your Lambda function is associated with a VPC, you need to ensure it has access to the public internet.

Distributed Tracing

Distributed tracing allows you to propagate a trace context from a service running on a host to a service running on AWS Lambda, and vice versa, so you can see performance end-to-end. Linking is implemented by injecting Datadog trace context into the HTTP request headers.

Distributed tracing headers are language agnostic, e.g., a trace can be propagated between a Java service running on a host to a Lambda function written in Python.

Because the trace context is propagated through HTTP request headers, the Lambda function needs to be triggered by AWS API Gateway or AWS Application Load Balancer.

To enable this feature, you simple need to decorate your Lambda handler function with @datadog_lambda_wrapper.

import requests
from datadog_lambda.wrapper import datadog_lambda_wrapper

@datadog_lambda_wrapper
def lambda_handler(event, context):
    requests.get("https://www.datadoghq.com")

Note, the Datadog Lambda Layer is only needed to enable distributed tracing between Lambda and non-Lambda services. For standalone Lambda functions, traces can be found in Datadog APM after configuring the X-Ray integration.

Patching

By default, widely used HTTP client libraries, such as requests, urllib2 and urllib.request are patched automatically to inject Datadog trace context into outgoing requests.

You can also manually retrieve the Datadog trace context (i.e., http headers in a Python dict) and inject it to request headers when needed.

import requests
from datadog_lambda.wrapper import datadog_lambda_wrapper
from datadog_lambda.tracing import get_dd_trace_context

@datadog_lambda_wrapper
def lambda_handler(event, context):
    headers = get_dd_trace_context()
    requests.get("https://www.datadoghq.com", headers=headers)

Sampling

The traces for your Lambda function are converted by Datadog from AWS X-Ray traces. X-Ray needs to sample the traces that the Datadog tracing agent decides to sample, in order to collect as many complete traces as possible. You can create X-Ray sampling rules to ensure requests with header x-datadog-sampling-priority:1 or x-datadog-sampling-priority:2 via API Gateway always get sampled by X-Ray.

These rules can be created using the following AWS CLI command.

aws xray create-sampling-rule --cli-input-json file://datadog-sampling-priority-1.json
aws xray create-sampling-rule --cli-input-json file://datadog-sampling-priority-2.json

The file content for datadog-sampling-priority-1.json:

{
  "SamplingRule": {
    "RuleName": "Datadog-Sampling-Priority-1",
    "ResourceARN": "*",
    "Priority": 9998,
    "FixedRate": 1,
    "ReservoirSize": 100,
    "ServiceName": "*",
    "ServiceType": "AWS::APIGateway::Stage",
    "Host": "*",
    "HTTPMethod": "*",
    "URLPath": "*",
    "Version": 1,
    "Attributes": {
      "x-datadog-sampling-priority": "1"
    }
  }
}

The file content for datadog-sampling-priority-2.json:

{
  "SamplingRule": {
    "RuleName": "Datadog-Sampling-Priority-2",
    "ResourceARN": "*",
    "Priority": 9999,
    "FixedRate": 1,
    "ReservoirSize": 100,
    "ServiceName": "*",
    "ServiceType": "AWS::APIGateway::Stage",
    "Host": "*",
    "HTTPMethod": "*",
    "URLPath": "*",
    "Version": 1,
    "Attributes": {
      "x-datadog-sampling-priority": "2"
    }
  }
}

Non-proxy integration

If your Lambda function is triggered by API Gateway via the non-proxy integration, then you have to set up a mapping template, which passes the Datadog trace context from the incoming HTTP request headers to the Lambda function via the event object.

If your Lambda function is deployed by the Serverless Framework, such a mapping template gets created by default.

Log and Trace Correlations

By default, the Datadog trace id gets automatically injected into the logs for correlation, if using the standard python logging library.

If you use a custom logger handler to log in json, you can inject the ids using the helper function get_correlation_ids manually.

Set the environment variable DD_LOGS_INJECTION to false to disable this feature.

from datadog_lambda.wrapper import datadog_lambda_wrapper
from ddtrace.helpers import get_correlation_ids

@datadog_lambda_wrapper
def lambda_handler(event, context):
  trace_id, span_id = get_correlation_ids()
  logger.info({
    "message": "hello world",
    "dd": {
      "trace_id": trace_id,
      "span_id": span_id
    }
  })

Opening Issues

If you encounter a bug with this package, we want to hear about it. Before opening a new issue, search the existing issues to avoid duplicates.

When opening an issue, include the Datadog Lambda Layer version, Python version, and stack trace if available. In addition, include the steps to reproduce when appropriate.

You can also open an issue for a feature request.

Contributing

If you find an issue with this package and have a fix, please feel free to open a pull request following the procedures.

License

Unless explicitly stated otherwise all files in this repository are licensed under the Apache License Version 2.0.

This product includes software developed at Datadog (https://www.datadoghq.com/). Copyright 2019 Datadog, Inc.