Monitoring Serverless Architectures on AWS Lambda - Nordics Lego themed

@srhtcn

@srhtcn Monitoring Serverless Architectures on AWS Lambda Technical Evangelist at
Atlassian AWS Community Hero @srhtcn Serhat Can LEGO brand and images belong to LEGO® lego.com

@srhtcn Hygge

@srhtcn Fires and dragons

@srhtcn

@srhtcn Lego!

@srhtcn Opsgenie’s Serverless journey

@srhtcn Opsgenie’s Serverless journey 2015 Small scale custom integrations

@srhtcn Opsgenie’s Serverless journey 2015 2016 Small scale custom integrations
First production usage with async like dynamodb auto scale

@srhtcn Opsgenie’s Serverless journey 2015 2016 2017 Small scale custom
integrations First production usage with async like dynamodb auto scale Customer facing feature “Service and Incident Management”

@srhtcn Opsgenie’s Serverless journey 2015 2016 2018 2017 Small scale
custom integrations First production usage with async like dynamodb auto scale Customer facing feature “Service and Incident Management” Thundra: Observability for AWS Lambda

@srhtcn Some background

@srhtcn Atlassian acquires Opsgenie

@srhtcn Thundra was spun off before the acquisition

@srhtcn Two seperate companies but we share some serverless history

@srhtcn Agenda • Introduction to Serverless computing and AWS Lambda
• What is available and what is not for monitoring • Monitoring challenges and our solutions

@srhtcn

@srhtcn What is Serverless? From IaaS → CaaS → PaaS
→ FaaS

@srhtcn What is Serverless? From IaaS → CaaS → PaaS
→ FaaS Serverless is an event driven, utility based, stateless, code execution environment. Simon Wardley @swardley

@srhtcn Why should I go Serverless? Two main reasons

@srhtcn ServerLess is more Less code to maintain, less ops,
less toil • Scaling • Provisioning • OS or Language updates • Resource utilization • Network monitoring • Fault tolerance • Shipping logs https://landing.google.com/sre/book/chapters/eliminating-toil.html

@srhtcn Economics No payment for idle time or hosting Easy
to get started Faster time to market

@srhtcn How it works?

@srhtcn AWS Lambda Can be triggered by 20+ different services
Native support for many languages including Java, Node, .Net Core, Golang, Python Bring any language with Runtime API Layers for sharing code Also take a look at: SAM, AWS Serverless Application Repository, Step Functions, X-Ray

@srhtcn Give me examples!

Elasticsearch data indexing @srhtcn

Service and incident management @srhtcn

@srhtcn But, there are always challenges...

@srhtcn Be aware of some challenges Cold start Local development
Concurrent execution limit Well-known good practices Debugging and monitoring

@srhtcn What is available for monitoring?

@srhtcn Out of the box basic monitoring • Cloudwatch metrics:
Invocations, errors (permission, timeout, out of memory etc), concurrent executions

Invocations, errors (permission, timeout, out of memory etc), concurrent executions • Cloudwatch logs: Fast, easy, simple. Now, easier to search with its new interactive log analytics!

Invocations, errors (permission, timeout, out of memory etc), concurrent executions • Cloudwatch logs: Fast, easy, simple. Now, easier to search with its new interactive log analytics! • Distributed tracing with AWS X-Ray Easy to use end to end requests in distributed applications

@srhtcn • Logs: application logs • Metrics: CPU, memory etc.
• Traces ◦ Local: Debugging inside the functions ◦ Distributed: Trace external service calls • Aggregate logs, metrics and traces What data do we need in Serverless monitoring?

@srhtcn Why are current solutions not enough? Existing solutions do
not play well with serverless environments 1. Only sync data senders 2. Only manual instrumentation 3. Distributed tracing for distributed systems

@srhtcn 1. Only sync data senders • Longer request duration
• Stateless environment • Data publish failures • Access within VPC https://medium.com/thundra/4-reasons-why-you-should-publish-monitoring-data-async-in-aws-lambda-fd1e56473941 https://twitter.com/legobatmanmovie

@srhtcn Our approach Sync for development environment • Faster to
send the data • Easier to set up and debug • Cheaper

@srhtcn Our approach Async first • Doesn’t block Lambda functions
• Actual function is not affected • Best option for production Write data as log Data in Cloudwatch Consumer Lambda Thundra

@srhtcn 2. Only manual instrumentation • No way to attach
an agent • Polluting the code • Error-prone • Maintenance burden

@srhtcn Our approach - Automated instrumentation Use the automated way
whenever possible to keep things simple

@srhtcn Our approach - Local tracing for more context •
High resolution local tracing • Debugging at some sense • Trace external dependencies and 3rd party libraries

@srhtcn

@srhtcn 3. Distributed tracing for distributed systems Functions talk with
other services and functions AWS X-Ray does a good job at tracing external calls Topology detection matters for cost monitoring and optimization

@srhtcn Our approach • Local traces with automatic or manual
instrumentation • Integrates with AWS X-Ray to enrich distributed tracing data • Service map (topology) and distributed tracing is coming this April!

@srhtcn Integrates with AWS X-Ray to enrich distributed tracing

@srhtcn We also suggest: • Plugin for reducing cold-starts •
Open tracing compatible API and data model • Metrics, traces, and logs all in one view • Looking at outliers and go deep in what actually happened • Highlight outliers in a heat map, especially for external resources

@srhtcn Serverless computing will become the default computing paradigm of
the Cloud Era. @srhtcn https://www2.eecs.berkeley.edu/Pubs/TechRpts/2019/EECS-2019-3.pdf

@srhtcn Thundra in action! demo.thundra.io

@srhtcn

Monitoring Serverless Architectures on AWS Lamb...

Monitoring Serverless Architectures on AWS Lambda - Nordics Lego themed

More Decks by Serhat Can

Other Decks in Technology

Featured

Transcript