README.md

Lambda Powertools

A suite of utilities for AWS Lambda Functions that makes tracing with AWS X-Ray, structured logging, and creating custom metrics asynchronously easier - Compatible with Python >=3.6.

Status: Release candidate
How long until GA?: Current progress

Features

Tracing

Capture cold start as annotation, and response and exceptions as metadata
Run functions locally with SAM CLI without code change to disable tracing
Explicitly disable tracing via env var POWERTOOLS_TRACE_DISABLED="true"
Support tracing async methods

Logging

Capture key fields from Lambda context, cold start and structures logging output as JSON
Log Lambda event when instructed (disabled by default)
- Enable via POWERTOOLS_LOGGER_LOG_EVENT="true" or explicitly via decorator param
Log sampling enables DEBUG log level for a percentage of requests (disabled by default)
- Enable via POWERTOOLS_LOGGER_SAMPLE_RATE=0.1, ranges from 0 to 1, where 0.1 is 10% and 1 is 100%
Append additional keys to structured log at any point in time

Metrics

Aggregate up to 100 metrics using a single CloudWatch Embedded Metric Format object (large JSON blob)
Context manager to create an one off metric with a different dimension than metrics already aggregated
Validate against common metric definitions mistakes (metric unit, values, max dimensions, max metrics, etc)

Bring your own middleware

Utility to easily create your own middleware
Run logic before, after, and handle exceptions
Receive lambda handler, event, context
Optionally create sub-segment for each custom middleware

Bring your own middleware

Utility to easily create your own middleware
Run logic before, after, and handle exceptions
Receive lambda handler, event, context
Optionally create sub-segment for each custom middleware

Environment variables used across suite of utilities

Environment variable	Description	Default	Utility
POWERTOOLS_SERVICE_NAME	Sets service name used for tracing namespace, metrics dimensions and structured logging	"service_undefined"	all
POWERTOOLS_TRACE_DISABLED	Disables tracing	"false"	Tracing
POWERTOOLS_TRACE_MIDDLEWARES	Creates sub-segment for each middleware created by lambda_handler_decorator	"false"	middleware_factory
POWERTOOLS_LOGGER_LOG_EVENT	Logs incoming event	"false"	Logging
POWERTOOLS_LOGGER_SAMPLE_RATE	Debug log sampling	0	Logging
POWERTOOLS_METRICS_NAMESPACE	Metrics namespace	None	Metrics
LOG_LEVEL	Sets logging level	"INFO"	Logging

Usage

See example of all features, testing, and a SAM template with all Powertools env vars. All features also provide full docs, and code completion for VSCode and PyCharm.

Installation

With pip installed, run: pip install aws-lambda-powertools

Tracing

Tracing Lambda handler and a function

from aws_lambda_powertools.tracing import Tracer
tracer = Tracer()
# tracer = Tracer(service="payment") # can also be explicitly defined

@tracer.capture_method
def collect_payment(charge_id):
  ret = requests.post(PAYMENT_ENDPOINT) # logic
  tracer.put_annotation("PAYMENT_STATUS", "SUCCESS") # custom annotation
  return ret

@tracer.capture_lambda_handler
def handler(event, context)
  charge_id = event.get('charge_id')
  payment = collect_payment(charge_id)
  ...

Tracing asynchronous functions

import asyncio

from aws_lambda_powertools.tracing import Tracer
tracer = Tracer()
# tracer = Tracer(service="payment") # can also be explicitly defined

@tracer.capture_method
async def collect_payment(charge_id):
    ...

@tracer.capture_lambda_handler
def handler(event, context)
  charge_id = event.get('charge_id')
  payment = asyncio.run(collect_payment(charge_id)) # python 3.7+  
  ...

Tracing concurrent asynchronous with gather

⚠️ This will no longer be necessary after this X-Ray recorder issue is resolved as it's an edge case. ⚠️

To safely workaround this issue, use @tracer.capture_method on functions not being run with async.gather, and instead use in_subsegment_async context manager escape hatch to have the same tracing effect.

import asyncio

from aws_lambda_powertools.tracing import Tracer
tracer = Tracer()
# tracer = Tracer(service="payment") # can also be explicitly defined

async def another_async_task():
    async with tracer.provider.in_subsegment_async("## another_async_task"):
        ...

async def another_async_task_2():
    async with tracer.provider.in_subsegment_async("## another_async_task_2"):
        ...

@tracer.capture_method
async def collect_payment(charge_id):
    asyncio.gather(another_async_task(), another_async_task_2())
    ...

@tracer.capture_lambda_handler
def handler(event, context)
  charge_id = event.get('charge_id')
  payment = asyncio.run(collect_payment(charge_id)) # python 3.7+  
  ...

Using escape hatch mechanisms

You can use tracer.provider attribute to access all methods provided by xray_recorder. This is useful when you need a feature available in X-Ray that is not available in the Tracer middleware, for example thread-safe, or context managers.

Example using aiohttp with an async context manager

NOTE: It expects you have aiohttp as a dependency. aiohttp_trace_config uses lazy import to create a trace_config object following aiohttp protocol.

import asyncio
import aiohttp

from aws_lambda_powertools.tracing import Tracer, aiohttp_trace_config
tracer = Tracer()

# aiohttp_trace_config is x-ray extension for aiohttp trace config known as aws_xray_trace_config

async def aiohttp_task():
    # Async context manager as opposed to `@tracer.capture_method`
    async with tracer.provider.in_subsegment_async("## aiohttp escape hatch"):
        async with aiohttp.ClientSession(trace_configs=[aiohttp_trace_config()]) as session:
            async with session.get("https://httpbin.org/json") as resp:
                resp = await resp.json()
                return resp

@tracer.capture_method
async def async_tasks():
    ret = await aiohttp_task()
    ...

    return {
        "task": "done",
        **ret
    }

@tracer.capture_lambda_handler
def handler(event, context)
  ret = asyncio.run(async_tasks()) # python 3.7+  
  ...

Using a pre-configured tracer anywhere

# handler.py
from aws_lambda_powertools.tracing import Tracer
tracer = Tracer(service="payment")

@tracer.capture_lambda_handler
def handler(event, context)
  charge_id = event.get('charge_id')
  payment = collect_payment(charge_id)
  ...

# another_file.py
from aws_lambda_powertools.tracing import Tracer
tracer = Tracer(auto_patch=False) # new instance using existing configuration with auto patching overriden

Logging

Structuring logs with Lambda context info

from aws_lambda_powertools.logging import Logger

logger = Logger()
# Logger(service="payment", level="INFO") # also accepts explicit service name, log level

@logger.inject_lambda_context
def handler(event, context)
  logger.info("Collecting payment")
  ...
  # You can log entire objects too
  logger.info({
    "operation": "collect_payment",
    "charge_id": event['charge_id']
  })
  ...

Exerpt output in CloudWatch Logs

{  
   "timestamp":"2019-08-22 18:17:33,774",
   "level":"INFO",
   "location":"collect.handler:1",
   "service":"payment",
   "lambda_function_name":"test",
   "lambda_function_memory_size":"128",
   "lambda_function_arn":"arn:aws:lambda:eu-west-1:12345678910:function:test",
   "lambda_request_id":"52fdfc07-2182-154f-163f-5f0f9a621d72",
   "cold_start": "true",
   "sampling_rate": 0.1,
   "message": "Collecting payment"
}

{  
   "timestamp":"2019-08-22 18:17:33,774",
   "level":"INFO",
   "location":"collect.handler:15",
   "service":"payment",
   "lambda_function_name":"test",
   "lambda_function_memory_size":"128",
   "lambda_function_arn":"arn:aws:lambda:eu-west-1:12345678910:function:test",
   "lambda_request_id":"52fdfc07-2182-154f-163f-5f0f9a621d72",
   "cold_start": "true",
   "sampling_rate": 0.1,
   "message":{  
      "operation":"collect_payment",
      "charge_id": "ch_AZFlk2345C0"
   }
}

Appending additional keys to current logger

from aws_lambda_powertools.logging import Logger

logger = Logger()

@logger.inject_lambda_context
def handler(event, context)
  if "order_id" in event:
      logger.structure_logs(append=True, order_id=event["order_id"])
  logger.info("Collecting payment")
  ...

Exerpt output in CloudWatch Logs

{  
   "timestamp":"2019-08-22 18:17:33,774",
   "level":"INFO",
   "location":"collect.handler:1",
   "service":"payment",
   "lambda_function_name":"test",
   "lambda_function_memory_size":"128",
   "lambda_function_arn":"arn:aws:lambda:eu-west-1:12345678910:function:test",
   "lambda_request_id":"52fdfc07-2182-154f-163f-5f0f9a621d72",
   "cold_start": "true",
   "sampling_rate": 0.1,
   "order_id": "order_id_value",
   "message": "Collecting payment"
}

Metrics

This feature makes use of CloudWatch Embedded Metric Format (EMF), and metrics are created asynchronously by CloudWatch service.

Metrics middleware validates against the minimum necessary for a metric to be published:

At least of one Metric and Dimension
Maximum of 9 dimensions
Only one Namespace
Any Metric unit supported by CloudWatch

Creating multiple metrics

If using multiple middlewares, use log_metrics as the last decorator, or else it will fail with SchemaValidationError if no metrics are recorded.

from aws_lambda_powertools.metrics import Metrics, MetricUnit

metrics = Metrics()
metrics.add_namespace(name="ServerlessAirline")
metrics.add_metric(name="ColdStart", unit="Count", value=1)
metrics.add_dimension(name="service", value="booking")

@metrics.log_metrics
@tracer.capture_lambda_handler
def lambda_handler(evt, ctx):
    metrics.add_metric(name="BookingConfirmation", unit="Count", value=1)
    some_code()
    return True

def some_code():
    metrics.add_metric(name="some_other_metric", unit=MetricUnit.Seconds, value=1)
    ...

CloudWatch EMF uses the same dimensions across all metrics. If you have metrics that should have different dimensions, use single_metric to create a single metric with any dimension you want. Generally, this would be an edge case since you pay for unique metric

unique metric = (metric_name + dimension_name + dimension_value)

from aws_lambda_powertools.metrics import MetricUnit, single_metric

with single_metric(name="ColdStart", unit=MetricUnit.Count, value=1) as metric:
    metric.add_dimension(name="function_context", value="$LATEST")

NOTE: When using Metrics() in multiple places in your code, make sure to use POWERTOOLS_METRICS_NAMESPACE env var, or setting namespace param.

Bring your own middleware

This feature allows you to create your own middleware as a decorator with ease by following a simple signature.

Accept 3 mandatory args - handler, event, context
Always return the handler with event/context or response if executed
- Supports nested middleware/decorators use case

Middleware with no params

from aws_lambda_powertools.middleware_factory import lambda_handler_decorator

@lambda_handler_decorator
def middleware_name(handler, event, context):
    return handler(event, context)

@lambda_handler_decorator
def middleware_before_after(handler, event, context):
    logic_before_handler_execution()
    response = handler(event, context)
    logic_after_handler_execution()
    return response


# middleware_name will wrap Lambda handler 
# and simply return the handler as we're not pre/post-processing anything
# then middleware_before_after will wrap middleware_name
# run some code before/after calling the handler returned by middleware_name
# This way, lambda_handler is only actually called once (top-down)
@middleware_before_after # This will run last
@middleware_name # This will run first
def lambda_handler(event, context):
    return True

Middleware with params

@lambda_handler_decorator
def obfuscate_sensitive_data(handler, event, context, fields=None):
    # Obfuscate email before calling Lambda handler
    if fields:
        for field in fields:
            field = event.get(field, "")
            event[field] = obfuscate_pii(field)

    return handler(event, context)

@obfuscate_sensitive_data(fields=["email"])
def lambda_handler(event, context):
    return True

Tracing middleware execution

This makes use of an existing Tracer instance that you may have initialized anywhere in your code. If no Tracer instance is found, it'll initialize one using default options.

from aws_lambda_powertools.middleware_factory import lambda_handler_decorator

@lambda_handler_decorator(trace_execution=True)
def middleware_name(handler, event, context):
    return handler(event, context)

@middleware_name
def lambda_handler(event, context):
    return True

Optionally, you can enrich the final trace with additional annotations and metadata by retrieving a copy of the Tracer used.

from aws_lambda_powertools.middleware_factory import lambda_handler_decorator
from aws_lambda_powertools.tracing import Tracer

@lambda_handler_decorator(trace_execution=True)
def middleware_name(handler, event, context):
    tracer = Tracer() # Takes a copy of an existing tracer instance
    tracer.add_anotation...
    tracer.metadata...
    return handler(event, context)

@middleware_name
def lambda_handler(event, context):
    return True

Debug mode

By default, all log statements from AWS Lambda Powertools package are suppressed. If you'd like to enable them, use set_package_logger utility:

import aws_lambda_powertools
aws_lambda_powertools.logging.logger.set_package_logger()
...

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

Lambda Powertools

Features

Usage

Installation

Tracing

Tracing Lambda handler and a function

Tracing asynchronous functions

Tracing concurrent asynchronous with gather

Using escape hatch mechanisms

Using a pre-configured tracer anywhere

Logging

Structuring logs with Lambda context info

Appending additional keys to current logger

Metrics

Creating multiple metrics

Bring your own middleware

Middleware with no params

Middleware with params

Tracing middleware execution

Debug mode