test: use postgrest.read_stdout in io tests

[steve-chavez]

It's easier to maintain this way in case there are new log lines.

Related to #3229 (comment)

[wolfgangwalther]

This takes away from the previous test, because the log format itself is not tested anymore, right?

How about just doing snapshot tests for each log level, and thus saving the full log output for each? This way we'd have a much better picture of the log level differences.

[steve-chavez]

This takes away from the previous test, because the log format itself is not tested anymore, right?

Just added back the previous regexes. I thought the user agent and roles on the log lines were irrelevant for this test but I'm cool with adding them back.

[steve-chavez]

How about just doing snapshot tests for each log level, and thus saving the full log output for each? This way we'd have a much better picture of the log level differences.

I'd rather not invest more time on this for now. This was just a step for #3229, which I need for clearing an unknown pool issue #3214.

[steve-chavez]

This is still an improvement over the current tests, I think. Will merge.

[steve-chavez]

How about just doing snapshot tests for each log level, and thus saving the full log output for each? This way we'd have a much better picture of the log level differences.

Wanted to open an issue for this, but giving it more thought.. maybe snapshot testing is not adequate for logs because the lines order is not deterministic. The snapshots could change frequently and we'd have to update them everytime (more maintenance).

Currently we do sorted() on the log lines to make the output deterministic, which has been working fine.