These release notes support ptx®/Cluster File System (ptx/CFS) V1.1.2, which enables ptx/Enhanced File System (ptx/EFS) filesystems to be mounted on any cluster node and shared among all cluster nodes. Together, ptx/CFS and ptx/EFS let you create highly available filesystems with fast reboot and recovery times, on-line maintenance operations, and full UNIX File System (UFS) compatibility for use in a cluster.
Use this version of ptx/CFS with the following products:
DYNIX/ptx® V4.5.2 or later
ptx/EFS V1.4.2 or later
ptx/CLUSTERS V2.2.1 or later
ptx/SVM V2.2.1 or later (optional)
This release of ptx/CFS, like ptx/CLUSTERS, supports clusters containing up to four NUMA-Q nodes (or two Symmetry nodes). A CFS filesystem can be mounted on any node in a cluster. For example, it can be mounted on all cluster nodes, on one cluster node, on two cluster nodes, etc.
See the DYNIX/ptx and Layered Products Software Installation Release Notes for information on how to install ptx/CFS.
This procedure tells how to create a new ptx/CFS filesystem. The procedure in the next section, "Converting a ptx/EFS Filesystem to ptx/CFS," tells how to convert a ptx/EFS filesystem to ptx/CFS.
ATTENTION This procedure assumes you are using ptx/SVM with ptx/CFS.
Use ptx/SVM to create a log area for each node. Each node in the cluster needs its own log for each filesystem. The default log size is 4 MB if the log device is big enough; otherwise, the default is 1 MB.
# vxassist -g shared_disk_group make log1 4M (on node 1)
# vxassist -g shared_disk_group make log2 4M (on node 2)
Use ptx/SVM to create a volume for the filesystem:
# vxassist -g shared_disk_group make volume 100m
Use newfs to create a filesystem:
# newfs -F cfs /dev/vx/rdsk/shared_disk_group/volume
Edit the vfstab file to create an entry for the new filesystem. Include the logdev= option to specify the log areas for the nodes. For example:
mount -F cfs -o logdev=/dev/vx/dsk/shared_disk_group_name/log1 \
/dev/vx/dsk/shared_disk_group/volume /mountpoint (on node 1)
mount -F cfs -o logdev=/dev/vx/dsk/shared_disk_group_name/log2 \
/dev/vx/dsk/shared_disk_group/volume /mountpoint (on node 2)
Mount the filesystem.
The following procedure tells how to convert an existing ptx/EFS filesystem into a ptx/CFS filesystem. To create a new ptx/CFS filesystem from scratch, see the preceding section, "Creating a New ptx/CFS Filesystem."
ptx/EFS filesystems must be V1.3.0 or later before they can be converted to ptx/CFS filesystems. To upgrade pre-V1.3.0 ptx/EFS filesystems, follow these steps:
Unmount the filesystem.
Use ptx/SVM to create a log area for each node. Each node in the cluster needs its own log for each filesystem. The default log size is 4 MB if the log device is big enough; otherwise, the default is 1 MB.
# vxassist -g shared_disk_group make log1 4M (on node 1)
# vxassist -g shared_disk_group make log2 4M (on node 2)
Edit the vfstab file to add entries that specify a log device for each node. Change the efs entry to cfs. Add the logdev= option and specify the location of the log areas. Set the vfs_automnt field to yes.
Mount the filesystem as a ptx/CFS filesystem with the mount -F cfs command.
ptx/CFS filesystems are mounted automatically with the /etc/rc2.d/S07MOUNTCFS script, which runs when the system goes to multiuser mode. When ptx/CFS filesystems are mounted for the first time, the fsck command is performed automatically.
To back out of ptx/CFS format and return to ptx/EFS filesystems, perform this procedure:
Unmount cluster filesystems on all nodes in the cluster with the umount command.
Edit the vfstab file, changing the relevant entry from cfs to efs and removing the logdev= option.
Run the mount command.
The following documentation is available on the online documentation CD and at http://webdocs.sequent.com/:
ptx/EFS and ptx/CFS Administration Guide
This section lists the following problem report summaries:
The number that appears in parentheses in the title of each problem report is the problem-tracking system number assigned to the report.
This release of ptx/CFS includes fixes for the following software defects:
(251805) A race condition occurred invx_cllock_notifyd/vx_cldestroy.
(251869) The CFS resize function panicked in xted_iinactive.
(252437) ptx/CFS hung on unmount operations that were followed by mount operations.
(252945) A race condition between vx_cllock_notifyd and vx_add_notify led to a lost BAST.
(253027) An mmu fault occurred during a mount/unmount operation.
In DYNIX/ptx, filesystems are required to ensure that all but the last part of the file is in page-size contiguous disk blocks. For ptx/EFS, this requirement requires special handling; for ptx/CFS, this special handling can lead to a deadlock. The deadlock does not exist in 4K (default) and 8K filesystems.
Workaround. Use 4K or 8K filesystems.
The fsadm resize command does not work on ptx/CFS filesystems.
Unmount the filesystem as CFS on all the nodes.
Mount the filesystem as EFS.
Run fsadm on the filesystem.
Remount the filesystem as CFS.
On manufacturing (debug) kernels, it is possible in a rare case to encounter a DEBUG ASSERT "ted2_call_demon: vx_freeze_inode:2a" error. This error can occur under the following conditions: If a filesystem is unmounted shortly after a copy of a very large file that fills the buffer cache with a large amount of dirty buffers, or if the bdflush process starts writing that file just before the filesystem is unmounted.
Workaround. Run the sync command and wait for I/O to finish before running the umount command.
Audit files and locale files must be on local filesystems only.
CFS filesystems run ptx/CFS recovery operations automatically when necessary. These recovery operations can deadlock while waiting to write the audit record or read the locale file. Access to the CFS filesystem blocks until the ptx/CFS recovery operations finish.
Workaround. Do not put audit files or locale files on CFS filesystems.