Perry

Perry was used as a testing site to explore entity relationship concepts.

How it works

1. Write a JSON file specifying the entities and relationships, looking like data.json.
2. Create JSON schemas from the spec.
3. Create SQL schemas from the spec.
4. Create GraphQL schemas from the spec.
5. Use a generic UI to create/read/update objects (folder ui).

Perry includes:

• Models API to simply serve the spec mentioned above (models).
• a shared internal library (perrydl) in Typescript and Rust (WASM) that:
  • generates SQL schema from the spec
  • generates GraphQL schemas from spec
  • generates JSON schemas from the spec
• Dupe API (dupe) which is an extremely unoptimized backend that:
  • queries the model from Models API
  • uses perrydl to generate SQL schema, GraphQL schema and JSON schemas from the spec
  • bootstraps a sqliteDB using the generated SQL
  • serves a GraphQL API using the generated GraphQL schema (to serve and mutate the data in the sqliteDB)
  • uses JSON schemas to validate input data
  • allows bulk insert and update using CSVs
  • captures data changes and sends to an in-memory event-broker (gRPC-Web)
• A user interface in Typescript+React+Recoil ui
  • queries Models API to create entity relationship internal representation
  • allows exploring entities using a graph like structure
  • generates forms to create data objects
  • uses tables and filters to present data objects
  • listens to events served by an in-memory event-broker (gRPC-Web)

graph TD
    S(Data Spec in json) -->|served by| M[Models API]
    D[Dupe] -->|query models| M
    D -->|create DB and store data| Q[sqliteDB]
    D -->|publish mutation events| E[Event Broker]
    U[UI] -->|query model| M
    U -->|query data using GraphQL| D
    U -->|consume mutation events| E

Loading

Bugs and limitations

Perry is not production ready and should not be used in a production environment, it's just exploration:

• search by text is flawed and applies to first column only
• gRPC stream lacks keep alive
• gear icon on text search is useless
• items cannot be dragged & dropped even though there are drop zones
• sessions are volatile, if you reload tabs are lost
• mutation events published by dupe are meant as a "poor man's CDC" but are not consistent (if dupe fails to publish them, they won't be retried)
• the event broker is an in-memory broker with short-lived historic data
• changing fields which are unique yields cryptic errors

This project is not being actively maintained.

Demo

Go to https://perry-demo.pages.dev and use the credentials:

• user: [email protected]
• password: 1veryrandomPassword4perry

The demo is fairly limited:

• data resets every hour
• data sample does not explore all entities and relations in the data spec
• backend requests are rate limited to prevent abuse
• app is halted if high network traffic is detected
• translation to de_DE comes straight from https://translate.i18next.com/