6.2.2 HP IBRIX 9000 Storage Release Notes (AW549-96061, January 2013)

• Fusion Manager (FM) restarted a CFR job even if it is marked “stopped by the user.” Fusion
Manager has been modified so that it does not automatically restart a CFR job when the user
requests the CFR job to be stopped.
• A soft lockup of the CPU was occurring during connection/delegation recovery.
• The creation of directory tree quota names with a colon (:) in the name produced undesirable
results and it further impacted the creation of new "normal" directory tree quota names until the
entry with the colon character was removed. Users are now told in the HP IBRIX 9000 Storage
File System User Guide that they cannot create a directory tree quota containing a comma (,) or
colon (:) character.
• When performing a "graceful" shutdown (ibrix_umount) of the IBRIX system for maintenance
reasons, the segment server hosting the root segment took 30 minutes per NFS export to timeout
and finally unmount.
• When accessing IBRIX file systems through NFS, the client NFS access would hang intermittently.
• When segments reached 70–80% of capacity, clients were unable to write to an SMB share.
• The quota current size would be reduced after running the rebalance task, even though files were
not modified or deleted. During movement of files across segments, there are intermediate replica
files that are created. Once data from an original file is synced with its replica, the original file
is deleted and the replica is promoted as the original file. These replicas have to be marked with
the directory quota ID, which is the same as that of the original master file. That was not occurring.
As the original file was deleted post replication, the directory quota accounted for it was getting
decremented. Since the newly created file was not marked with the directory quota ID, it was
never was accounted for. As a result, a reduction in quota usage was shown post replication.
The fix was to mark the replica created with the quota ID same as that of the original file, so any
data written into this file is accounted for against the set quota ID.
• Data tiering could result in zero length files.
• When a snapshot was taken, files could be deleted from the source directory.
• Stack overflows were occurring, as a result of some of the IBRIX commands, such as ibrix_fs
requiring a large buffer. The fix lessened the risk of a stack overflow.
• Dangling dentries were created when mv and cp commands were killed during the file creation
operation.
• The upgrade60.sh upgrade script was sometimes showing an invalid file system message in
environments where clusters were on a flat network.
• A delay would occur between serial sequential file writes, about 2 to 20+ seconds between the
first file transfer and the next. This delay would be seen during the drag and drop of multiple 4
GB or larger files from Windows Explorer onto the IBRIX SMB share.
• Checksum errors were being reported while accessing multiple database files from SUSE Linux
NFS Clients, which were reading multiple database pages (8 KB) into the NFS Client memory for
a SAP application. Although some of the data inside the 8 KB page was missing, when the same
page was re-read, the page data was completely intact and the checksum was correct. This
situation only occurred when the read operations involved a cluster network hop to a remote FSN
on a remote couplet, and the issue has been resolved.
• During a mkdir operation, the xdr_encode_netobj dereferencing NULL
xdr_netobj->data pointer message was displayed. The underlying issue has been resolved
so the message no longer appears.
• During Directory Change Notification (DCN) processing, a serialization issue could cause the
thread to loop forever while traversing the list of consumers of the notification. The Linux kernel
would subsequently detect a soft lockup and log messages in regards to the situation would appear
8 Fixes