test(qe-wasm): WASM query engine benchmarks isolated from DB

[miguelff]

The present PR adds continuous benchmarking of the current query-engine WASM, comparing it with:

• NAPI on the same branch
• WASM in main
• WASM baseline (at the beginning of our performance inquiries)

It provides a comment like this with the benchmark results for different queries in text mode.

Benchmarks are considered to have regressed if there is more than one benchmark 5% slower in the current WASM query engine compared to the latest version (coming from main)

Close https://github.com/prisma/team-orm/issues/725
Close https://github.com/prisma/team-orm/issues/732
Close https://github.com/prisma/team-orm/issues/733
Close https://github.com/prisma/team-orm/issues/734

Implementation:

The code uses a decorator over the driver adapter that fetches query results from a local cache after warmup, in a way such as the database IO latency is a no-op.

The queries to benchmark are defined in the queries.json file. Once this PR is reviewed for a correct approach and implementation, we can add more queries the query-benchmarks suite if necessary.

[github-actions]

WASM Size

Engine	This PR	Base branch	Diff
WASM	3.133MiB	3.133MiB	0.000B
WASM (gzip)	1.184MiB	1.184MiB	0.000B

[codspeed-hq]

CodSpeed Performance Report

Merging #4567 will not alter performance

_{Comparing da-benchmarks (ee4fc7a) with main (e78ce01)}

Summary

✅ 11 untouched benchmarks

[miguelff]

Builds a custom image that loads the seed file, see Dockerfile at https://github.com/prisma/prisma-engines/pull/4567/files#diff-96c0901b002d607e8553d7b5b38b2dd0dccdf904ade81ff4c8f4dffa6ab54140

[miguelff]

Yup, quite large. The alternatives are:

1. Store this file in a different place in the internet and pull it (ex. a dedicated repository)
2. Seed the database each time benchmark runs (10s -> 5 minutes)

[miguelff]

@janpio Did you try to mean "Format the json file?" 😂

[miguelff]

Thank you for the feedback @SevInf and @jkomyno I have addressed it and also decrease the tolerance of the benchmarks to 1% to consider it a regression.