[v3] Slowdown because of large source maps for queries that don't error

[mohe2015]

I assume because of

postgres/lib/index.js

Line 187 in 51d033c

query.origin = options.debug ? new Error().stack : cachedError(xs)

stack traces are generated. It seems like the caching does not work as they seem to be re-generated for every request. nodejs/node#41541 also plays into this but still I don't understand why this is necessary. If you need a minimally reproducible example please tell me and I will try to provide one. The relevant thing is probably that cachedError should only calculate the stack trace the first time. What are some reasons this does not work? The queries are wrapped in a transaction. The following is the code that creates the pool.

export const sql = postgres(process.env["DATABASE_URL"], {
  host: process.env["DATABASE_HOST"],
});

[porsager]

Thanks a lot for this issue. I'll give it a closer look Sunday.

[porsager]

Yeah, that looks like an oversight on my part for .unsafe()... Since the caching relies on the first argument from the tagged template function it's not possible to make origin work for unsafe.

I think it should be changed to get the stack only if the error is thrown either way. Don't know why I hadn't thought of that.

[porsager]

Would you mind trying 0aacdb3 (current rewrite branch) out to see if there's an improvement?

I have a vague memory testing this at some point only to find it didn't have much of an effect, but let's see :) If it's still just as bad, I think adding origin to the stack should be removed from .unsafe().

[mohe2015]

It looks fixed. If it will happen again I will reopen. Thanks for fixing this!