Download as PDF, PPTX
![type Mutation {
addUser(name: String!, email: AWSEmail!, role: Role!): User
@aws_auth(cognito_groups: ["Admin", "SuperUser"])
…
}](https://image.slidesharecdn.com/lessonsfromrunningappsyncinprod-211118155245/85/Lessons-from-running-AppSync-in-prod-95-320.jpg)
![type Mutation {
addUser(name: String!, email: AWSEmail!, role: Role!): User
@aws_auth(cognito_groups: ["Admin", "SuperUser"])
…
}](https://image.slidesharecdn.com/lessonsfromrunningappsyncinprod-211118155245/85/Lessons-from-running-AppSync-in-prod-96-320.jpg)
![type Mutation {
addUser(name: String!, email: AWSEmail!, role: Role!): User
@aws_auth(cognito_groups: ["Admin", "SuperUser"])
…
}](https://image.slidesharecdn.com/lessonsfromrunningappsyncinprod-211118155245/75/Lessons-from-running-AppSync-in-prod-95-2048.jpg)
![type Mutation {
addUser(name: String!, email: AWSEmail!, role: Role!): User
@aws_auth(cognito_groups: ["Admin", "SuperUser"])
…
}](https://image.slidesharecdn.com/lessonsfromrunningappsyncinprod-211118155245/75/Lessons-from-running-AppSync-in-prod-96-2048.jpg)
Uploaded byYan Cui
Lessons from running AppSync in prod
This document provides lessons learned from running three AppSync projects in production over 18 months. It discusses preferences for using VTL over Lambda resolvers for CRUD operations due to VTL being faster, cheaper and simpler. It also recommends per-resolver caching over full request caching. Other tips include not leaving logging on full in production, handling user errors gracefully, planning for nested CloudFormation stacks for large projects, and modeling multi-tenancy using Cognito groups and attributes.
More Related Content
Similar to Lessons from running AppSync in prod
Lessons from running AppSync in prod
- 1. Lessons from running AppSyncin production Yan Cui
- 2.
- 3.
- 4.
- 5. three AppSync projectsin the last 18 months
- 6. social network healthcaredocument assembly B2B & B2C B2B & B2C B2B multi-tenant multi-tenant multi-tenant 168 resolvers 111 resolvers 195 resolvers
- 7. social network healthcaredocument assembly B2B & B2C B2B & B2C B2B multi-tenant multi-tenant multi-tenant 168 resolvers 111 resolvers 195 resolvers AppSync Lambda DynamoDB AppSync Lambda DynamoDB DynamoDB Lambda AppSync
- 8. social network healthcaredocument assembly B2B & B2C B2B & B2C B2B multi-tenant multi-tenant multi-tenant 168 resolvers 111 resolvers 195 resolvers AppSync Lambda DynamoDB AppSync Lambda DynamoDB DynamoDB Lambda AppSync Cognito Algolia Cognito ElasticSearch Algolia Auth0
- 9. social network healthcaredocument assembly B2B & B2C B2B & B2C B2B multi-tenant multi-tenant multi-tenant 168 resolvers 111 resolvers 195 resolvers AppSync Lambda DynamoDB AppSync Lambda DynamoDB DynamoDB Lambda AppSync Cognito Algolia Cognito ElasticSearch Algolia Auth0 Lumigo Lumigo Lumigo
- 10. Prefer VTL overLambda Faster, cheaper, simpler and more scalable 01.
- 11.
- 12.
- 13.
- 14. AppSync DynamoDB Lambda complex businesslogic call 3rd party services
- 15. AppSync DynamoDB Lambda complex businesslogic call 3rd party services retries w/ exponential backoff
- 16. AppSync DynamoDB Lambda complex businesslogic call 3rd party services retries w/ exponential backoff fallbacks/circuit-breakers
- 17. AppSync DynamoDB Lambda complex businesslogic call 3rd party services retries w/ exponential backoff fallbacks/circuit-breakers do more than one thing
- 18. AppSync DynamoDB Lambda complex businesslogic call 3rd party services retries w/ exponential backoff fallbacks/circuit-breakers do more than one thing can use pipeline resolvers instead
- 19. AppSync DynamoDB Lambda another sourceof latency
- 20. AppSync DynamoDB Lambda another sourceof latency cold starts
- 21. AppSync DynamoDB Lambda another sourceof latency cold starts another cost
- 22. AppSync DynamoDB Lambda another sourceof latency cold starts another cost concurrent executions limit
- 23. For CRUD operations,prefer VTL to Lambda resolver
- 24. AppSync DynamoDB Cheaper Faster Morescalable Simpler
- 25. “But I endup writing a lot of VTL and I don’t like VTL!”
- 26. “But I endup writing a lot of VTL and I don’t like VTL!” - someone who uses single-table design
- 27. You don’t needSTD* * STD = single-table design 02.
- 28. single-table design A collectionof practices & modelling techniques for DynamoDB that solves the problem of data joining by pre-joining data into collections in a single table.
- 29. id name 1 YanCui 2 theburningmonk UsersTable OrdersTable userId orderId 1 007 1 008
- 30. id name 1 YanCui 2 theburningmonk UsersTable OrdersTable userId orderId 1 007 1 008 GetItem
- 31. id name 1 YanCui 2 theburningmonk UsersTable OrdersTable userId orderId 1 007 1 008 GetItem Query
- 32. pk sk user#1 YanCui user#2 theburningmonk MySingleTable name orderId user user user#1 order#007 007 user#1 order#008 008
- 33. pk sk user#1 YanCui user#2 theburningmonk MySingleTable name orderId user user user#1 order#007 007 user#1 order#008 008 Query
- 34. pk sk user#1 YanCui user#2 theburningmonk MySingleTable name orderId user user user#1 order#007 007 user#1 order#008 008 Query
- 35. pk sk user#1 YanCui user#2 theburningmonk MySingleTable name orderId user user user#1 order#007 007 user#1 order#008 008 Query
- 36.
- 37. single-table design Steep learningcurve. Difficult to add new access patterns.
- 38. single-table design Steep learningcurve. Difficult to add new access patterns. Can’t monitor usage cost by entity type.
- 39. single-table design Steep learningcurve. Difficult to add new access patterns. Can’t monitor usage cost by entity type. Difficult to use DynamoDB streams.
- 40.
- 41.
- 42. You should learnsingle-table design.
- 43. You should learnsingle-table design. But don’t make it your religion.
- 44. “But I endup writing a lot of VTL and I don’t like VTL!” - someone who uses single-table design
- 45. “But I endup writing a lot of VTL and I don’t like VTL!” - someone who uses single-table design to use single-table design with AppSync, you have to write a lot of custom logic in VTL…
- 46. “But what aboutall the cost savings from Single-Table Design?!”
- 47. “But what aboutall the cost savings from Single-Table Design?!” Only matters when running at scale.
- 48. “But what aboutall the cost savings from Single-Table Design?!” Caching is a far more effective cost-saving strategy.
- 49.
- 51.
- 52. You need touse per-resolver caching Because full-request caching just won’t do 03.
- 53.
- 55.
- 56. full request caching $context.arguments &$context.identity
- 57. full-request caching won’tget you very far
- 58. this profile appearsin many requests full request caching
- 59. this tweet appearsin many requests full request caching
- 60. per-resolver caching each resolvercan opt-in for caching
- 61. per-resolver caching each resolvercan opt-in for caching can specify different TTL & cache key for each
- 62. this profile canbe cached across different requests, by different users per-resolver caching
- 63. Caching seldom changeddata for everyone is an important part of a good caching strategy
- 64. You can flushthe cache, but not individual items
- 65. You may seeoccassional overhead when AppSync establishes connections to the cache node
- 67. Don’t leave loggingon full in production Great for debugging, even better for going bankrupt 04.
- 69.
- 73. don’t leave resolverlog level at ALL in production
- 74.
- 75. Handle user errorsgracefully Because no one likes to wake up at 3am for a false alarm 05.
- 76. HTTP 5xx = ourbad 4xx = your bad
- 77. HTTP 5xx = ourbad 4xx = your bad Alert on these!
- 78. HTTP 5xx = ourbad 4xx = your bad Alert on these! Keep an eye on these, but don’t wake me up at 3am
- 79. API Gateway module.exports.handler =async (event) => { …. // validation logic return { statusCode: 404, } }
- 80. AppSync (Lambda resolver) don’thave a way to return 4xx errors
- 81. AppSync (Lambda resolver) module.exports.handler= async (event) => { …. // validation logic throw new Error(“not found”) }
- 82. AppSync (Lambda resolver) module.exports.handler= async (event) => { …. // validation logic throw new Error(“not found”) } Trips Lambda error alerts
- 83. AppSync (Lambda resolver) module.exports.handler= async (event) => { …. // validation logic return { error: { message: “blah blah”, type: “SomeErrorType” } } }
- 84. AppSync (Lambda resolver) if(!$util.isNull($ctx.result.error)) $util.error($ctx.result.error.message, $ctx.result.error.type) #end $utils.toJson($ctx.result)
- 85. Plan for goingbeyond 500 CloudFormation resources Because you will get there, sooner than you think 06.
- 87.
- 89. No need torestructure your project. !Ref and !GetAtt works as before.
- 90. Avoid circular referenesbetween the root and nested stack(s)
- 92. Model multi-tenant withCognito groups and attributes Because SaaS is the new black 07.
- 93. Users belong totenants. Users have roles within the tenant. Some actions are limited to particular roles.
- 95. type Mutation { addUser(name:String!, email: AWSEmail!, role: Role!): User @aws_auth(cognito_groups: ["Admin", "SuperUser"]) … }
- 96. type Mutation { addUser(name:String!, email: AWSEmail!, role: Role!): User @aws_auth(cognito_groups: ["Admin", "SuperUser"]) … }
- 97. NEVER accept tenantID as argument.
- 98. NEVER accept tenantID as argument. Use the identity in the request (resolved by Cognito).
- 100. https://theburningmonk.com/hire-me Advise Training Delivery “Fundamentally, Yanhas improved our team by increasing our ability to derive value from AWS and Lambda in particular.” Nick Blair Tech Lead
- 101. Learn GraphQL andAppSync by building a Twitter clone with these technologies appsyncmasterclass.com “Doing the AppSync Masterclass course has given me the confidence and skills to start a project I've been wanting to do for 5 years.” “If you are thinking about starting a new side project, take this course first!” “This is by far the best course available of AppSync!” “If you can only afford 1 course, I would suggest you go for this one.”
- 102.