Going serverless with aws

Going serverless with aws

• 1.
  Going Serverless with AWS AlexLanda
• 2.
  Services in theCloud • Infrastructure (IaaS) • virtualized basic computing, networking and storage resources. • EC2, VPC, EBS … • Platform (PaaS) • complete development and deployment environment in the cloud • RDS, ElastiCache, Elastic Beanstalk … • Software (SaaS) – complete application offered through the web.
• 3.
  Serverless - don’tthink about servers • A trending execution model in which the server management and the resource management are completely hidden. • AWS offers a set of serverless services • Storage – S3 • DB – DynomoDB • Queues – SQS, Kinesis • Notifications – SNS • Monitoring – CloudWatch • Compute - Lambda
• 4.
  Used for • Automation– ”maintenance scripts” for the cloud • ETL – extraction and transformation layer in front of the backend services • External services • Glue code – connect between several services • Full blown applications – not the main purpose (yet)
• 5.
  Lambda • AWS Lambdaallows you to define a function that will be executed when some event happens. • You do not need to care about the physical execution regarding resources as AWS Lambda will automatically scale and provide CPU/Network/IO and Memory for it. • A major part in a serverless architecture.
• 6.
  Lambda • The Lambdacode you write must be in one of the following languages: • Java. • C#. • Python. • Node.js (JavaScript). • Go
• 7.
  How Does ItWork? • The main component of the Lamda is the handler. • The point-of-entry. • When some event happens (more about it in the next slides), Lamda will create a container ands runs your function there (by invoking the handler). • The container may be retain for some time to allow consecutive executions without paying the bootstrapping price. • Note that you can’t count on it and a new container may be created.
• 8.
  Event Sources • AWSLambda can connect and be activated by a large variety of events. • For example: • S3 – respond to object_created event. • SNS – respond to object published on a configured topic. • API Gateway – wire the Lambda to a REST API. • CloudWatch Events – respond to object changing state (e.g., EC2 instance state changes). • You can also invoke the Lambda directly from the CLI or your application (custom events).
• 9.
  Example HTTP Endpoint
• 10.
  The Container • Lamdacreates a container to run your function. • When your function exits, the container is frozen for some time and may be thawed for consecutive execution. • You are provided with 500MB of /tmp directory which contents are maintained between invocations. • Note that you can’t count on it as a new container may be created for you. • A good use-case for /tmp is caching.
• 11.
  The Container • Globalobjects can be reused and are maintained cross freeze/thaw. • For example, static objects in Java. • In is considered a good practice to try and reuse global resources (e.g., connection pools).
• 12.
  What about scale? •Scale in Lambda == “Concurrent Executions” • Handled by execution pools – 1000 per account, per region • Can be changed • If not set function uses “unreserved concurrency pool” • All specified limits should reach together 900 • 100 reserved for the ”unreserved concurrency”
• 13.
  Throttling • Behaves differentlyfor synchronous and asynchronous events • HTTP – synchronous • S3 trigger – asynchronous • In case of synchronous - invoking application receives a 429 error • In case of asynchronous - will automatically retry the invocation twice, with delays between retries • If DLQ is configured the failed event will be sent to SNS topic or SQS Queue
• 14.
  Advanced example Machine LearningHTTP Endpoint
• 15.
  Logging – CloudWatch •Lambda integrates with Cloud Watch for logging • Pushes the logs to /aws/lambda/<function name> • Common errors: • permissions denied – modify IAM permission • timeout exceeded – increase timeout • memory exceeded - increase memory • Exception stack trace - go and handle :)
• 16.
  Monitoring – CloudWatch • Provides you a full view on the execution metrics • Number of executions • Duration • Failures • Concurrent executions • .. • You should set alarms on those metrics to be notified in case of unexpected behavior
• 17.
  Serverless != Effortless •How do we manage all of this? • Manually – Good for demo and job security, bad for everything else.. • Automatically – using Terraform / CloudFormation + scripts • Good but requires a lot of knowledge • Doesn’t solve the debugging • We want a framework to manage it for us!
• 18.
  Serverless Framework • Opensource CLI for building serverless applications • Allows you to create, deploy, test , invoke and debug your serverless services • Serverless service is composed of : • Functions – mapped to AWS Lambda • Resources • Permissions
• 19.
  Serverless Framework • Usesserverless.yml to describe the functions of your service as well as their permissions and dependent resources • By running serverless deploy it will deploy your changes using cloud formation
• 20.
  Serverless Framework Demo
• 21.
  Questions ? ? ?? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?