We believe the problem is fixed. But we are still monitoring. There was a compute node in a bad state which may have created a deadlock in the filesystem. We removed this node and after that the deadlock went away.
We are still in contact with our vendor to confirm the root cause of the problem. But tests indicate the problem is solved as we can access both scratch and the PL new spaces after the action above was taken.
If you still find any problems, please contact us via email@example.com