mbudm / fotopia-serverless Goto Github PK

Need a node.js api to replicate what amplify does in the client for functional testing (and bulk uploader). Or, apply solutions mentioned in an aws amplify issue claiming that the library works in node.js with a polyfill for DOM globals

The put indexes lambda is invoked by the streams lambda that is triggered by dynamodb creates, update and removes.

Occasionally two events occur simultaneously (within a few 100ms of each other) and they get the same existing state and write a new state. Not a big deal at small scale but will become a bigger issue (and before I'm ready to implement proper search indexes.

So the additon of an SQS resource should do the trick - force each put index to complete before the next one is tackled. And/or use a dynamodb table for faster get/put.

Separate the stack currently defined in one serverless.yml file into db, s3, api and cognito.

This allows for more control over what is deployed when but it also introduces complexity with coordinating release of components.

Related to #1

Currently when bulk uploading around 300 images the person categorisation ends up with multiples of the same person - more if it's a baby/child.

I think there may be a clever way to use the face matches of all faces in person A to see if there is a certain level of similarity to the face matches of all faces in person B.

This could be a cron job lambda. It could prompt the user to merge people or if above a certain confidence level, do it automagically in the background.

Delete a person, and:

• remove person from associated images
• update person indexes
• delete faces from rekognition collection?

Sometimes you need to redeploy - eg if you tweak any of the dynamodb indexes. This means all the data in the db and S3 buckets are gone.

Need a way to handle migration when the stack is recreated.

Breaking up serverless.yml will help, as there's more control over what gets removed and re deployed - but this might still be a need

The ability to share an image publicly

Need to automate deletion of the AWS Rekognition collection as part of a redeploy.

Modify the query API to default to return paginated records - with some limit of say 50 or 100 results.

Or leave as is and expect the client to default to time limited (month?) search?

Probably both.

Choose a different thumbnail image for the person. Currently it just uses the face id of the first image found.