mbkroese

I'm also affected by this problem when I use `pyspark` as a dependency.

No, this problem occurs when either the data structure has non-deterministic order or the code generating the parametrised test cases is for some reason not deterministic.

> downside is ofc that if one parametrised test is very time consuming vs the rest of the suite, the splits would not be great. Yes, and I wonder if...

>Incidentally, I think you can get slightly better performance out of least_duration by sorting the tests by duration first so they are processed from longest to shortest. @alexforencich you're right,...

Just imagine you work for a company that has a large suite of tests and the corresponding technical debt. The tests already have order dependency (by accident) but it's too...

I also recognised the problem that you are flagging here. In my case the groups run on different nodes, and they get their own CI_NODE_INDEX assigned automatically by Gitlab. After...

@sondrelg I think that could work - right now we update the data that we read before and write the result, but it could be changed to just create a...

Suppose currently we have these durations: ``` {'a': 1, 'b': 2, 'c': 3} ``` And we run one group that only executes test 'a' with new duration 5, then (in...

I see what you mean now. I assumed you always just wanted to write the durations of tests we ran. But instead you want to _update_ the durations at the...

@alexforencich thanks for reporting. Is it possible to run your test suite with this version: `0.2.3`? This is the version before the last one, and would help us pin down...