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.
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-Python27Datadog-Python36Datadog-Python37Datadog-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
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 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_API_KEY: <DD_API_KEY>The Datadog API Key must be defined as one of the following environment variables via AWS CLI or Serverless Framework:
- DD_API_KEY - the Datadog API Key in plain-text, NOT recommended
- DD_KMS_API_KEY - the KMS-encrypted API Key, requires the
kms:Decryptpermission - DD_API_KEY_SECRET_ARN - the Secret ARN to fetch API Key from the Secrets Manager, requires the
secretsmanager:GetSecretValuepermission (andkms:Decryptif using a customer managed CMK)
You can also supply or override the API key at runtime:
# Override DD API Key after importing datadog_lambda packages
from datadog import api
api._api_key = "MY_API_KEY"Set the following Datadog environment variable to datadoghq.eu to send your data to the Datadog EU site.
- DD_SITE
If your Lambda function powers a performance-critical task (e.g., a consumer-facing API). You can avoid the added latencies of metric-submitting API calls, by setting the following Datadog environment variable to True. Custom metrics will be submitted asynchronously through CloudWatch Logs and the Datadog log forwarder.
- DD_FLUSH_TO_LOG
To connect logs and traces, set the environment variable below to True. The default format of the AWS provided LambdaLoggerHandler will be overridden to inject dd.trace_id and dd.span_id. The default Datadog lambda log integration pipeline will automatically parse them and map the dd.trace_id into the reserved trace_id attribute.
- DD_LOGS_INJECTION
To debug the Datadog Lambda Layer, set the environment variable below to DEBUG.
- DD_LOG_LEVEL
To increment aws.lambda.enhanced.invocations and aws.lambda.enhanced.errors Datadog Lambda integration metrics set this environment variable to true:
- DD_ENHANCED_METRICS
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 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
)If your Lambda function is associated with a VPC, you need to ensure it has access to the public internet.
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.
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)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.jsonThe 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
B1D6
:
{
"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"
}
}
}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.
To connect logs and traces, set the environment variable DD_LOGS_INJECTION to True. The log format of the AWS provided LambdaLoggerHandler will be overridden to inject dd.trace_id and dd.span_id. The default Datadog lambda log integration pipeline will automatically parse them and map the dd.trace_id into the reserved attribute trace_id.
If you use a custom logger handler to log in json, you can manually inject the ids using the helper function get_correlation_ids.
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
}
})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.
If you find an issue with this package and have a fix, please feel free to open a pull request following the procedures.
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.