Does postgrest cache? Help with broken benchmark

[cptwunderlich]

Hi there!
Via some reddit discussion, I was made aware of the Techempower Framework Benchmarks. Looking through it, I found that most of the Haskell benchmarks were derelict.
Now, let's not discuss whether these benchmarks are good or bad. I just wanted to play around a bit with this.

Here is my PR: TechEmpower/FrameworkBenchmarks#10064

The "Fortunes" benchmark fails with this message:

Only 506 rows read in the database out of roughly 6144 expected.

It uses pg_stats to get this metric, AFAIK. So I'm wondering, how that is possible? Does PostgREST cache results internally?
Is there a way to circumvent this?

You can see for yourselves by checking out the PR.
Running the benchmark: tfb --mode verify --test postgrest
There is also tfb --mode debug --test postgrest to start a server and leave it running for debugging (i.e., you can make requests to the server).

[steve-chavez]

Hey @cptwunderlich, thanks for updating the benchmark 🙏

Only 506 rows read in the database out of roughly 6144 expected.
Does PostgREST cache results internally?

Nope, we don't. Maybe pg_stats is outdated and needs an ANALYZE?

[cptwunderlich]

I have a hunch that their query to check selected rows might be wrong, but I just spent half an hour trying to find where they start the postgres docker container, so that I can expose the port. Without success...