The scenario can happen when even two reframe processes on different nodes try to append to the same perflog file multiple hundrends of lines at the same time. This can be triggered easier when perflog_compat=True with multiple performance variables being logged in separate lines. The result is that multiple NULL characters are inserted in the log file and some lines are also corrupted.

The problem is that NFS does not guaranteed atomic appends, so to fix this we need to support locking.