[META] Backend performance testing

Chances are, this is related to Paragraphs Performance, so see also our connected meta issue about it.

Another thing is that sometimes media widgets can cause a slow UI, since it causes lots of filesystem operations and the filesystem can be slow in clusters with shared filesystems.

At Paragraphs Performance, we started some example setups to compare reference scenarios:
https://www.drupal.org/project/2855549/git-instructions

We should start setting up reference use cases for Thunder and then profile them when loading edit, triggering some ajax or saving.

Crediting myself and @chr.fritsch for code review.

I added some issues we might want to look into.

More interesting issues

Here's an idea that came up...

It would be very helpful if a standardised framework for contrib module performance was recommended that can be continuously applied on any contrib... Ultimately, it would be very interesting to see this rolled out to the testbot so per-method test runs and maybe sub-metrics could be passed into these results...

OK maybe testbot resources and their load varies that causes jitter and results are not directly comparable, but some rolling average should be.. Or we should make sure the resources are standardised.

And in case we want to have accurate results, you simply trigger multiple test runs.

Or do we need to take this out to a dedicated performance reference bot that has always the same resources and is free of other load?

@miro_dietiker We actually implemented something very similar for testing the performance regressions in Thunder :-)

And one of the results actually was, that testing on AWS is not very stable, even if you use a dedicated instance. We are still optimizing the testing infrastructure. Maybe this can result in a standardized framework for testing performance, but we are certainly not there yet.