Slonik is 50x slower when used with ncc

[thomasthiebaud]

Slonik is 10x slower when I compiled my code using ncc. I'm not sure where the issue lies, because pg does not have the same issue.

Here is a minimal repo to reproduce the issue https://github.com/thomasthiebaud/slonik-perf-issue

[guybedford]

@thomasthiebaud source maps can considerably slow down execution time. Did you try compiling with ncc without the source map?

[thomasthiebaud]

@guybedford it does not seem that this is the issue. Without source and without minifying, I still have a huge perf difference for Slonik

pg - query: 6.555ms
ncc-pg - query: 12.474ms

slonik - query: 9.418ms
ncc-slonik - query: 606.6ms

You can easily reproduce the issue in that repo https://github.com/thomasthiebaud/slonik-perf-issue

[guybedford]

@thomasthiebaud do any of the dependencies have native implementations that are only used when a native platform version is available? ncc not managing to inline such native code resulting in a JS fallback might explain the issue?

[thomasthiebaud]

yes slonik has an optional dependency on pg-native https://github.com/gajus/slonik/blob/6bafb002e2c7fff468b4baeeb7ae57d8c2dd86bd/src/factories/createPool.ts#L48-L58

[guybedford]

Right - it would be worth checking if pg-native is working via ncc itself directly.

[thomasthiebaud]

It does not seem that pg-native has an issue with ncc

pg - query: 23.174ms
ncc-pg - query: 26.919ms

slonik - query: 8.378ms
ncc-slonik - query: 566.233ms

pg-native - query: 2.083ms
ncc-pg-native - query: 2.627ms

[guybedford]

Is there any other dependency you think it could be? Because this is performance and not an actual breakage, missing native deps / slow optional deps seems the most likely cause.

[thomasthiebaud]

It does not seem like there is another native dependency