Migrate off vegeta for dynamic target loadtests

[steve-chavez]

Problem

We use vegeta + a python script to generate dynamic targets for our JWT loadtests.

Once something goes wrong on the python script, it's really hard to debug and modify. This happened recently #4544 (comment) and the fix was quite laborious (#4563).

Proposal

• Use the k6 tool also used on https://github.com/PostgREST/postgrest-benchmark. It supports embedded scripting in javascript.

• Use vegeta as a library?

[steve-chavez]

A slight problem after #4563, the jwt-rsa-cache-worst loadtest generates a lot of 0s in the CPU column for the markdown report (see here). This is because the process monitoring starts when withPgrst is invoked, the target generation runs afterwards (which takes around ~230 seconds while PostgREST does nothing, see here) and only after that the vegeta loadtest runs.

postgrest/nix/tools/loadtest.nix

Lines 114 to 117 in 2bcd336

	${withTools.withPgrst} -m "$_arg_monitor" \
	${withGenTargets} --method "$_arg_method" --rsa="$_arg_testdir"/gen_jwk.json --private-key="$_arg_testdir"/gen_private.json "$_arg_testdir"/gen_targets.http \
	sh -c "cd \"$_arg_testdir\" && \
	${runner} -lazy -targets gen_targets.http -output \"$abs_output\" \"''${_arg_leftovers[@]}\""

This problem would also be nonexistent with k6 or maybe vegeta as library.