Description of problem: On a distribute cluster with NUFA, upon the disk getting full the files are distributed across the cluster deviating from NUFA behavior. Shouldn't the client be reporting ENOSPACE? Is this the expected behavior? Version-Release number of selected component (if applicable): lusterfs 3.4.0.19rhs built on Aug 14 2013 00:11:42 How reproducible: Always Steps to Reproduce: 1. Create a distribute volume and switch on nufa 2. Mount from one of the servers and write till disk is full 3. Observe that data is written on other servers as well.
NUFA doesn't override all the behavior of DHT, its just 'tries' to create the file locally as long as possible, if the disks are getting full, then it won't continue to write the file locally.