add: override PGRST_CMD for postgrest-loadtest from env

[develop7]

Enables overriding postgrest executable with PGRST_CMD environment variable in postgrest-loadtest script. This enables, for example, running loadtest against profiled executable with simple PGRST_CMD=postgrest-profiled-run postgrest-loadtest, which wasn't possible before, because PGRST_CMD was defined internally.

[steve-chavez]

PGRST_CMD=postgrest-profiled-run postgrest-loadtest

The above is real nice, we can use it to improve perf.

Looking at this #4220 (comment), is this PR dependent on #4220? Or how do you confirm this PR is working as expected?

[develop7]

Looking at this #4220 (comment), is this PR dependent on #4220? Or how do you confirm this PR is working as expected?

It doesn't depend on #4220, I've merely used PGRST_CMD=postgrest-profiled-run postgrest-loadtest to verify it works. One could use /usr/bin/true just as well.

[steve-chavez]

How about using argbash ARG_USE_ENV instead?

There's an example on

postgrest/nix/tools/withTools.nix

Lines 28 to 32 in cc4f920

	"ARG_USE_ENV([PGUSER], [postgrest_test_authenticator], [Authenticator PG role])"
	"ARG_USE_ENV([PGDATABASE], [postgres], [PG database name])"
	"ARG_USE_ENV([PGRST_DB_SCHEMAS], [test], [Schema to expose])"
	"ARG_USE_ENV([PGTZ], [utc], [Timezone to use])"
	"ARG_USE_ENV([PGOPTIONS], [-c search_path=public,test], [PG options to use])"

[steve-chavez]

This way we also document the command has the PGRST_CMD option when doing --help

[develop7]

And have it documented too! Thanks for the suggestion, will do!

[wolfgangwalther]

With the default of postgrest-run, this will change the logic heavily now. We will now have to explicitly set this to the empty string to get the default behavior from before.

Also, we can't make use of PGRST_BUILD_CABAL anymore.

I think we should not control this via env variable at all, but instead add a --profiled flag to postgrest-loadtest and then use the profiled package in both variants if this is set.

[develop7]

PGRST_CMD env argument doesn't have a default value anymore and should be good to go now

re: separate flag> I agree this is a cleaner behavior, but it'd require to propagate the desired executable name (path) from pgrst-loadtest to postgrest-with-pgrst script, which I'm yet to wrap my head around of.