Michael Härtl

Ok I would say that we should *never* release any lock before rotation. This will always cause concurreny issues. A fix should probably: * Keep `flock()` active until writing +...

After more analysis I would go so far and say that there's no easy fix. What we basically want is a mutex here. We try to re-implement it with file...

> solution from PHP manual: That was my second suggestion above. > or by using flock() parameter $would_block This has nothing to do with it (also see [here](https://stackoverflow.com/a/24823415/767752)). We already...

See the linked SO answer that I linked in my last comment above. `$would_block` is basically just a flag that indicates that something is seriously going wrong when accessing the...

Both [suggested fixes](https://github.com/yiisoft/yii2/issues/19259#issuecomment-1049710928) come with a cost (BC break). But I would vote for option 2: * Fix too early release of lock file * Drop `rotateByCopy` and make it...

@cebe I think you should also add a test for batch queries using ActiveRecord. I found, that they currently don't work at all (unless you have the schema already cached...

I currently don't have time to take a closer look here. But I'm quite surprised that there are *still* problems with URL rules. If you check the current set of...

All you need should be phpunit. I'm using version 6 here via the global .phar file: https://phpunit.de/getting-started/phpunit-6.html

Do we need more live testing before merging? Locking is tricky and it's too easy to overlook something.

I can not really help here. Maybe @stu2162583 can try it?