improve cache.json generation implementation (reduce duplicated grapqhl calls in serialization workflow)

[thescientist13]

Type of Change

• New Feature Request
• Documentation / Website
• Improvement / Suggestion
• Bug
• Other (please clarify below)

Summary

Technical debt tracking for some of the issue rough and dirty technical design decisions made as part of implementing #115 .

Details

The way cache.json data gets written is a little convoluted since each request needs to be done twice before it can get written as cache.json as part of the build / serialization lifecycle.

1. Puppeteer renders a page
2. Request gets made to GraphQL server
3. GraphQL intercepts the request and calls createCache
4. createCache then instantiates an instance of Apollo client and makes a call back to the GraphQL server
5. The response of that is then written to disk

This seems cumbersome and will likely not scale well? Or just seems slow. It would be great if on the response handler of the GraphQL server could handle writing to disk, or at least avoiding the duplicate query call, which has required us to do this or else we will get stuck in an infinite loop.

if (req.query.q !== 'internal') {
  await createCache(req, context);
}

Ideally, we could use some form of memoization so duplicated queries for a specific route subset are not repeated, which would certainly help with performance. This kind of caching improvement (memoization) could / should also applies to client.js as well to avoid unnecessary fetch calls.

[thescientist13]

Note sure if using ssrMode would be of any help?

Also, maybe we could backfill the remaining properties of a response from calling client.query?