Closed chaoran-chen closed 2 weeks ago
Ah sorry, makes sense to revert if there's a chance for flakiness. Not that it has happened in practice I think.
Isn't it dangerous to work on one database? Do you assume that every test cleans up after itself?
Do you assume that every test cleans up after itself?
Yes, there is a clean up step after (or before) every test and thus every test assumes that it's running on a clean state.
Could we easily set up separate databases for each runner? It's not worth if it takes much work, but speeding up tests isn't completely without benefit!
speeding up tests isn't completely without benefit!
You have to weigh it against initial setup and long-term maintenance. My feeling is that it's not worth the effort yet.
Sure! I just don't know how hard it is to give each runner its own db.
Sure! I just don't know how hard it is to give each runner its own db.
Note I've reworded the PR to not auto close the flakiness because I think the flakiness predates parallelity.
I'm merging this now already because I introduced the change and shouldn't have. No need to discuss further at this point and waste our time ;)
Maybe helps with #2708 (hopefully, let's see)
This reverts the parallel test execution that was introduced in #1914. Some tests expect an empty database, so parallel execution can very likely be a source of flakiness.
I will run backend tests five times: https://github.com/loculus-project/loculus/actions/runs/10716872022/job/29715313279?pr=2710