slides.md

theme	background	class	highlighter	lineNumbers	info	drawings	transition	css
seriph	https://source.unsplash.com/collection/94734566/1920x1080	text-center	shiki	false	## Best Practices for Using Python to Power Serverless Applications ### PyCon 2023 Presentation Materials	persist true	slide-left	unocss

Best Practices for Using Python to Power Serverless Applications

Capital One

PyCon 2023

Press Space for next page

---

transition: fade-out layout: two-cols

::right::

Dan Furman

• @djfurman4tech
• :/djfurman

Distinguished Engineer

Card Tech, Data

::default::

Brian McNamara

• @mcnamarabrian
• :/mcnamarabrian

Distinguished Engineer

Retail Bank, Serverless

---

transition: fade-out

Why Python for Serverless?

• Simple is better than complex
• Fast development cycle
• Fantastic scaling
• Lower cost

---

transition: slide-down

Kinds of Serverless Applications

• API provider
• Streaming data processor
• Machine Learning Inference

---

transition: slide-down

API Application

---

transition: slide-down

Streaming Data Application

---

transition: slide-down

ML Inference Application

---

The Serverless Mindset

- Code is ephemeral

- Code is isolated

- Code is self-contained

- Code scales without intervention

---

Key Questions

- How do you observe your application (e.g., logs, metrics, performance)?

- How do you test, build, and deploy your application?

- How do you architect your application?

- Should you use Lambda for your application?

---

transition: slide-up

How do you know what your Lambda application is doing?

• Rules of observability change
• Logs, metrics, and traces all must persist outside of the function

---

transition: slide-up layout: image-right image: https://source.unsplash.com/collection/94734566/1920x1080

Observability Pro Tips

- Structure your logs - JSON is nice

- Set a retention period on your logs

- Log what you need but allow for debugging if necessary

- Emit custom metrics asynchronously using Embedded Metrics Format (EMF)

- Understand the different types of Lambda metrics and the statistics that should be used with each

- Metric math functions are available to you

- Don't just look at the average - look at the edges (p90, p99)

...just use AWS Lambda Powertools

---

transition: slide-up

Logging

Sample of using AWS Lambda Powertools package for logging¹ and metrics collection²

from aws_lambda_powertools import Logger, Metrics
from aws_lambda_powertools.metrics import MetricUnit
from aws_lambda_powertools.utilities.typing import LambdaContext

logger = Logger()
metrics = Metrics()


@logger.inject_lambda_context
@metrics.log_metrics
def handler(event: dict, context: LambdaContext) -> str:
    logger.info("Collecting payment")

    logger.info({"operation": "charge", "charge_id": event["charge_id"]})
    metrics.add_metric(name="SuccessfulCharge", unit=MetricUnit.Count, value=event["charge_amount"])
    return {
      "operation": "charge",
      "charge_id": event["charge_id"],
      "charge_amount": event["charge_amount"]
    }

<style> .footnotes-sep { @apply mt-20 opacity-10; } .footnotes { @apply text-sm opacity-75; } .footnote-backref { display: none; } </style>

---

Tracing Pro Tips

• Use Powertools to seamlessly
  • Annotate your functions
  • Instrument your packages
  • Inject keys for your telemetry

---

Trace your code

from uuid import uuid4
from aws_lambda_powertools import Tracer

# Automatically instrument calls to external dependencies
tracer = Tracer(patch_modules=["requests"])


@tracer.capture_method
def generate_uuid4() -> str:
    # You can apply this at any level to understand what is needed
    return str(uuid4())


@tracer.capture_method
def post_payment(user_id: str) -> dict:
    tracer.put_annotation(key="user_id", value=user_id)
    # Process the payment here

---

Live Code

These examples are nice; but we're at PyCon! Let's do some real code.

Join us at github.com/mcnamarabrian/pycon203

---

Thank you for joining us!

Footnotes

1. Logging ↩

2. Metrics ↩