In this post to his site Joe Ferguson gives some advice on solidifying tests in your system that are a bit more fragile. Every test suite of any larger size has these kinds of tests - ones that usually pass but sometimes fail (and then pass just fine on the next run).

On my first week at the new job I was tasked to fix some tests that were logging data. While the fix was simple enough, by using `PsrLogNullLogger as Logger` instead of `MonologLogger` in the test, during the process I ran into another test that appeared quite fragile.

He gives an example of a fragile test, one based on a method that returns a "food" value, that would potentially fail if the data returned is not in the right order. He found the issue was with the use of the assertArraySubset check and how, thankfully, the fix was as easy as changing the assertion (and using an array_diff to help with the check).