Links

About the Spike

Classes

• We have the Abstract **LimitOffsetCursor (we need to check that class and verify if we can reuse)**

  • The class UserCursor implements the **LimitOffsetCursor.**

• We have PageQuery

• We have LimitOffsetPageQuery

• The UsersInput Type have some classes

  • LimitOffsetPageQuery

• In the view_model(service) was create the UserConnection.

  

• The UserConnection has a builder method.

Token Info

• The variables are something like this

  {
    "input": {
      "after": "eyJvZmZzZXQiOiAxMzAsICJzb3J0QnkiOiAiTGFzdCBOYW1lIiwgInNvcnRPcmRlciI6IG51bGx9"
    }
  }
  

• The Devin spike uses the next information in the token.

  { 
    "offset": 130,
    "sortBy": "Last Name",
    "sortOrder": null,
    "title": "icardenas"
  }
  

• And the page info in the next

  "pageInfo": {
    "startCursor": "eyJvZmZzZXQiOiAxMzEsICJzb3J0QnkiOiAiTGFzdCBOYW1lIiwgInNvcnRPcmRlciI6IG51bGx9",
    "endCursor": "eyJvZmZzZXQiOiAxNTAsICJzb3J0QnkiOiAiTGFzdCBOYW1lIiwgInNvcnRPcmRlciI6IG51bGx9",
    "hasNextPage": true,
    "hasPreviousPage": true
  }
  

Phases

• Phase 1: Use the pagination in the base of a cursor, but use the offset and not in the base of ID for the complex.

  • My personal opinion: if the problem with offset is because the large offset is the problem, maybe we can wait and see how many “page” the user request with the same filters.

    This is an article that talks about this problem, then I feel that we want to delay this because I saw it difficult to apply an offset of the 10000 results

    • Screenshot
• Phase 2: Monitoring and checking performances.
• Phase 3: Migrate In the base of the ID, using LD toggle.
  • I know that using the ID is the best way to return records, but creating the strategy and thinking of the edge case, need more time.

Convention - Workflow

1. The FE sends the request using the input parameters.
2. We need a limit by default, but the FE can set one, but we need a max value allowed.
3. All params are content in the token [filters, offset, limit] (a similar approach is used in facebook), then:
   1. If neither filter changes, we work with the after and before properties and any extra param in the input should be ignored.
   2. If any filter changes, the FE should not send the after or before parameter and should send all the filters.

Question

• In which case would we use the before parameter?

• Start with JobConnection
  • Create

Example to return info from dataclass

• The variables are something like this

  {
  "owner_ids": ["ASDF"]
  }
  

• info in token

  { 
    "offset": 130,
    "sortBy": "Last Name",
    "sortOrder": null,
    "owner_ids": [123]
  }