%BOOK_ENTITIES; ]> The following are the known issues: Issues related to Distributed Replicated Volumes: When process has done cd into a directory, stat of deleted file recreates it (directory self- heal not triggered). In GlusterFS replicated setup, if you are inside a directory (for example, Test directory) of replicated volume. From another node, you will delete a file inside Test directory. Then if you perform stat operation on the same file name, the file will be automatically created. (that is, a proper directory self-heal is not triggered when process has done cd into a path). Issues related to Distributed Volumes: Rebalance does not happen if bricks are down. Currently while running rebalance, make sure all the bricks are in operating or connected state. glusterfsd - Error return code is not proper after daemonizing the process. Due to this, scripts that mount glusterfs or start glusterfs process must not depend on its return value. After # gluster volume replace-brick VOLNAME Brick New-Brick commit command is issued, the file system operations on that particular volume, which are in transit will fail. Command # gluster volume replace-brick ... will fail in a RDMA set up. If files and directories have different GFIDs on different backends, GlusterFS client may hang or display errors. Work Around: The workaround for this issue is explained at . Issues related to Directory Quota: Some writes can appear to pass even though the quota limit is exceeded (write returns success). This is because they could be cached in write-behind. However disk-space would not exceed the quota limit, since when writes to backend happen, quota does not allow them. Hence it is advised that applications should check for return value of close call. If a user has done cd into a directory on which the administrator is setting the limit, even though the command succeeds and the new limit value will be applicable to all the users except for those users’ who has done cd in to that particular directory. The old limit value will be applicable until the user has cd out of that directory. Rename operation (that is, removing oldpath and creating newpath) requires additional disk space equal to file size. This is because, during rename, it subtracts the size on oldpath after rename operation is performed, but it checks whether quota limit is exceeded on parents of newfile before rename operation. With striped volumes, Quota feature is not available. Issues related to POSIX ACLs: Even though POSIX ACLs are set on the file or directory, the + (plus) sign in the file permissions will not be displayed. This is for performance optimization and will be fixed in a future release. When glusterfs is mounted with -o acl, directory read performance can be bad. Commands like recursive directory listing can be slower than normal. When POSIX ACLs are set and multiple NFS clients are used, there could be inconsistency in the way ACLs are applied due to attribute caching in NFS. For a consistent view of POSIX ACLs in a multiple client setup, use -o noac option on NFS mount to switch off attribute caching. This could have a performance impact on operations involving attributes. If you have enabled Gluster NLM, you cannot mount kernel NFS client on your storage nodes. Error with lost and found directory while using multiple disks. Work Around: You must ensure that the brick directories are one of the extra directories in the back-end mount point. For Example, if /dev/sda1 is mounted on /export1, use /export1/volume as the glusterfs's export directory. Due to enhancements in Graphs, you may experience excessive memory usage with this release. After you restart the NFS server, the unlock within the grace-period may fail and previously held locks may not be reclaimed. After a rebalancing a volume, if you run rm -rf command at the mount point to remove all contents of the current working directory recursively without prompting, you may get "Directory not Empty" error message. The following is a known missing (minor) feature: locks - mandatory locking is not supported.