during prerendering hypermodel has gotten very slow #3
Description
This is a fairly recent change I think.
More experiments are needed but if you check out this repo: https://github.com/catmando/HyperAQ
and go the prerendering-on branch you will see the problem.
The code is fetching about 70 records, and for each record is displaying 4 pieces of data in a table.
This takes 35 seconds on my box, this is compared to 1 second to fetch and process the data if done on the client.
Some experiments show that for fetching 1 piece of data (instead of 4) it only takes 3 seconds to render the page. Then if you get two fields from the same records it now takes 6 seconds, then if you get 3 pieces its up to 12, and finally the fourth piece (which is an attribute from a belongs_to relationship) takes 35 seconds. It seems like each additional piece of data is basically doubling the time.
I suspect the caching mechanism that keeps track of all the data fetched during pre-rerendering (which is needed so we can dump it the client and instantly rerender without re-fetching) is sub-optimal.
I also wonder if that caching mechanism got moved to the client (in the V8 engine) instead of the server.