Serverless for the Enterprise

Using serverless technologies brings a number challenges for large, enterprise scale systems. This talk covers the problems faced and offers insights and recommendations to help others wanting to use serverless for non-trivial use cases.

Transcript

1. 1
   Serverless for the Enterprise
   Rafal Gancarz
   Lead Consultant
   rafal.gancarz@opencredo.com
   @RafalGancarz
   

   View Slide

2. 2
   So you want to do Serverless?
   fun, cool, cutting edge, cost effective,
   agile, competitive, fast to deliver
   @RafalGancarz
   

   View Slide

3. 3
   … but you are an Enterprise
   security, operability, support, compliance
   @RafalGancarz
   

   View Slide

4. 4
   … or you just want to
   build something more substantial
   multiple development teams, CI/CD, DevOps,
   automated testing, loose coupling,
   microservices, event-driven
   @RafalGancarz
   

   View Slide

5. Simple use case - chat bot
   5
   @RafalGancarz
   

   View Slide

6. Simple use case - image resizer
   6
   @RafalGancarz
   

   View Slide

7. • API-first platform for the cloud
   management vendor
   • 10s of APIs
   • CRUD, scheduled/triggered
   background jobs, event-driven
   processing
   • Strict security policies
   • SAML SSO integration
   • Rich RBAC model
   • Legacy system integration
   • CI/CD
   • Operations
   • Tooling
   7
   Advanced use case
   ?
   @RafalGancarz
   

   View Slide

8. • Tight coupling
   • Sprawling web of
   dependencies
   • All or nothing deployments
   • Team dependencies/
   coordination challenges
   8
   Serverless monolith
   ABC
   API
   @RafalGancarz
   

   View Slide

9. • Scoped around domain bounded
   contexts or platform capabilities
   • Isolated infrastructure stack
   (Terraform state)
   • Consistent naming of resources
   • Dedicated build/deployment
   pipeline
   • 0 .. N Lambda functions
   • Explicitly defined security policies
   and dependencies
   • Optionally exposes an API
   • Optionally exposes domain events
   • Optionally subscribes to event
   sources
   9
   Serverless Components
   A B C
   API API
   @RafalGancarz
   

   View Slide

10. • Synchronous call to another
    component
    • via HTTPS API
    • via direct Lambda
    invokation
    • no buffering
    • you pay for the wait
    time :)
    10
    Integration patterns
    @RafalGancarz
    

    View Slide

11. • Asynchronous call to
    another component
    • send a message to SQS
    queue or SNS topic
    • fire and forget
    • decoupled
    • flexible message formats
    helpful
    11
    Integration patterns
    @RafalGancarz
    

    View Slide

12. • Asynchronous notifications
    • subscribe to SNS topic
    • decoupled
    • flexible message formats
    helpful
    12
    Integration patterns
    @RafalGancarz
    

    View Slide

13. • Swagger for API specifications
    • API definitions can be used for model
    validation and contract testing
    • API versioning (for non backward-
    compatible changes)
    • Developer portal for API discovery
    and documentation
    13
    APIs
    

    View Slide

14. 14
    Continuous delivery/deployment
    @RafalGancarz
    

    View Slide

15. 15
    Continuous delivery/deployment
    @RafalGancarz
    

    View Slide

16. 16
    Infrastructure provisioning
    @RafalGancarz
    

    View Slide

17. 17
    Code deployments
    ES 2015
    +
    @RafalGancarz
    (babeljs.io)
    (rollupjs.org)
    (apex.run)
    

    View Slide

18. 18
    Testing
    Unit testing
    (local/CI)
    Acceptance testing
    (test environment)
    Smoke testing
    (post deployment)
    testing pyramid
    @RafalGancarz
    

    View Slide

19. 19
    Testing
    + +
    Unit testing (local + CI)
    +
    Acceptance/functional testing (AWS)
    +
    Smoke testing (AWS)
    +
    @RafalGancarz
    (mochajs.org) (chaijs.com) (sinonjs.org) (proxyquire)
    

    View Slide

20. • Least privilege policy based on IAM
    roles
    • IAM credentials and STS used for
    operations/tooling access
    • Credential/key rotation
    • Encryption in transfer
    • Encryption in storage (for sensitive
    data)
    20
    Security principles
    @RafalGancarz
    

    View Slide

21. 21
    Security in depth
    @RafalGancarz
    

    View Slide

22. • Authentication
    • Bespoke SSO solution integrated with the legacy system
    • Serverless SAML SSO + JWT token
    • Authorisation
    • Custom authorisation library
    • API Gateway custom authorisers not flexible enough
    • Sensitive configuration values
    • Encrypted objects in the component’s S3 bucket
    • Secret management/storage component
    • Bespoke serverless solution based on DynamoDB and
    KMS
    22
    Security
    @RafalGancarz
    

    View Slide

23. • Bunyan logging library
    • JSON formatted logs
    • business specific metadata
    • transaction tracing
    • CloudWatch Logs for log collection
    • Log aggregation pipeline (Lambda + Kinesis + Lambda)
    • Bespoke log filtering solution
    • Elastic Search for log storage
    • Kibana (on a server!) for browsing
    23
    Logging
    @RafalGancarz
    

    View Slide

24. • CloudWatch metrics and alarms
    • API Gateway
    • Lambda
    • DynamoDB
    • Billing
    • Bespoke Operations Health Dashboard application
    • CloudTrail for auditing
    • API Gateway
    • AWS API calls
    24
    Monitoring & Auditing
    @RafalGancarz
    

    View Slide

25. • Serverless NPM registry (S3 + Lambda + API GW)
    • Jenkins Pipeline DSL scripts (Groovy)
    • Deployment framework (Python)
    • Operations Dashboard (Node)
    • more to come …
    25
    Tooling
    @RafalGancarz
    

    View Slide

26. • Serverless is still in its infancy - lots of problems to solve
    (sometimes not the problems you’d like to be solving)
    • Serverless evolves rapidly - new tools/solutions are
    emerging, new features becoming available, new
    ideas&patterns are being shared
    • Strong dependency on the cloud provider (SDKs, tools,
    support, limits)
    • Plan/adapt your capacity (DynamoDB, Kinesis)
    • Serverless frameworks are great for some use cases but
    quite opinionated and limiting for others (particularly
    around stack provisioning/management)
    • Tooling is sparse, a lot of ‘build your own’
    Lessons learned
    26
    @RafalGancarz
    

    View Slide

27. • Automated rollback (when deployment fails)
    • Data migration (when DynamoDB table name or hashing
    key changes)
    • Involving the wider engineering team
    27
    Challenges to be addressed
    @RafalGancarz
    

    View Slide

28. 28
    Thank you!
    Rafal Gancarz
    Lead Consultant
    rafal.gancarz@opencredo.com
    @RafalGancarz
    Questions?
    

    View Slide