 |
Index for
Section 8 |
|
 |
Alphabetical
listing for C |
|
 |
Bottom of
page |
|
cfsmgr(8)
NAME
cfsmgr - Manages a mounted physical file system in a cluster
SYNOPSIS
Query:
/sbin/cfsmgr [-h member] [-c] [-e] [-n] [-a attribute] [-q] [-v]
[-F raw | std ] [[-p] directory ] | [-d domain] | [-m filesystem]
Set:
/sbin/cfsmgr [-r] [-f] [-h member] -a attribute=value [-s] [-K directory]
[-u {filesystem | -d domain}]
{[[-p] directory ] | [-d domain] | [-m filesystem]}
Get version number:
/sbin/cfsmgr -V
OPTIONS
-h member Specifies the target member; that is, the member on which the
cfsmgr command is executed. If no member is specified, the local
member is assumed. A member name takes the same form as the
output from the command hostname -s.
-c Verifies the configuration without actually changing anything.
-e Shows servers for all mount points, including servers for mount
points served by the automount and autofsd daemons. By default,
cfsmgr does not list the servers for mount points served by
automount and autofsd.
-n Gets attributes without waiting for reconfiguration to complete.
If a reconfiguration is in progress at the time that a cfsmgr
query occurs, and the -n flag is specified, the attribute's
current value is returned. If the -n flag is not specified,
cfsmgr waits until reconfiguration is complete before getting the
attribute's value.
This flag is valid only when getting attributes.
-r Relocates underlying disks, if necessary, when a file system is
relocated. Keeping disks local to the serving member helps
performance.
-f Configures CFS as indicated, even if this means that some disks
will be remotely accessed. Without the -f flag, the relocation
fails unless all disks can be locally accessed on the target
member.
-a attribute
Returns the current value of the specified attribute.
-a attribute=value
Sets the attribute to the specified value.
The following are valid attributes. Attribute names are not case
sensitive, and any attribute can be preceded by CFS_. For
example, devices, DEVICES, and CFS_DEVICES are all synonyms.
DEVICES Use this attribute to get a list of the devices serving
the specified file systems or domains.
The DEVICES attribute can be queried, but it cannot be
set.
FSBSIZE Use this attribute to get and set the CFS block
transfer size for the specified file system. The
attribute can be set only for mounted file systems.
You cannot set FSBSIZE on an AdvFS domain (the -d
option).
The value of FSBSIZE controls the amount of data, in
bytes, that is processed per pass for client-side reads
and writes for the specified CFS file system.
Generally, a larger value yields better performance,
and a value of 64 kilobytes or larger is suitable for
FSBSIZE.
There are special cases where a smaller value for
FSBSIZE can be advantageous. If the reads and writes
for a file system are small and totally random (not
sequential) in nature, then the extra processing that
is done for a large value of FSBSIZE could be wasteful.
This is not the case when there are multiple threads
all doing small I/O to a filesystem but reading it
sequentially.
For example, if the I/O for a file system is 8
kilobytes or less and totally random, then a value of 8
kilobytes for FSBSIZE is appropriate for that file
system.
The default value for FSBSIZE is determined by the
value of the cfsiosize kernel attribute. To learn the
current value of cfsiosize, use the following command:
sysconfig -q cfs cfsiosize
Values for FSBSIZE must be between 8192 bytes (8k) and
131072 bytes (128k), inclusive. When you set FSBSIZE,
the value is automatically rounded to the nearest page.
For example, if you set FSBSIZE=100000, the new value
for FSBSIZE is 106496.
SERVER Use this attribute to manually relocate the file system
to another member or to get information about the
server of a file system.
To relocate a file system to a different member, assign
the name of the target member to this attribute. The
member name can be in either short or long form. For
example, either SERVER=mutt, or SERVER=mutt.tyron.com
is acceptable.
A member always serves its own member-specific root
domain. You cannot relocate the member root domain to
a different member. You can relocate the clusterwide
root.
File systems of type File-on-File Mount (FFM), NFS
client, and Memory File System (MFS) cannot be
relocated. An attempt to relocate file systems of
these types returns an error (EINVAL). Relocating a
file system of type AdvFS affects not only the selected
file system, but all file systems in the same domain.
The whole domain is relocated.
STATISTICS
Use this attribute to get statistics related to the
cluster file system or to reset the current statistics
to zero. If you do not use the -h option, the
statistics are for CFS activity on the member where the
cfsmgr command executes.
To reset statistics, assign the value 0 (zero) to the
attribute: STATISTICS=0. An attempt to assign a value
other than zero results in an error message.
If you specify a directory in the command line for
cfsmgr, the statistics for activity in that directory
are displayed or reset. If you specify a domain name
in the command line, the statistics for all filesets in
the domain are either totaled and displayed or reset.
Statistics are reported in the following categories:
read, write, lookup, getattr, readlink, access, and
other (such as mkdir, rename, create, and remove). The
statistics for reads, lookups, getattrs, and accesses
are for those activities that are not cached.
-K directory
Forcibly unmounts all AutoFS intercept points and auto-mounted
file systems served on the cluster member serving the specified
directory, even if the intercept points and auto-mounted file
systems are busy. Specify the directory on which an AutoFS
intercept point or auto-mounted file system is mounted.
-q Shows the cluster attributes for the current configuration.
-s Specifies silent mode. A valid exit value is returned. This can
be useful in scripts that execute cfsmgr.
-u {filesystem | -d domain}
Forcibly unmounts an AdvFS file system or domain that is not
being served by any cluster member. Specify an AvdFS file system
mount directory, or the -d flag with the domain to unmount.
To determine if a domain or file system is not served, use the
following command and check the Server Status line:
cfsmgr {filesystem | domain}
To forcibly unmount a file system, use the following command:
cfsmgr -u filesystem
To forcibly unmount a domain, use the following command:
cfsmgr -u -d domain
The unmount is not performed if the file system or domain is
being served. If there are nested mounts on the file system being
unmounted, the forced unmount is not performed. Similarly, if
there are nested mounts on the file system on the domain being
unmounted, and the nested mount is not in the same domain, the
forced unmount is not performed.
-v Enables verbose mode.
-V Returns the version number of cfsmgr.
-p directory
The name of the directory on which the file system is mounted.
This is valid for all types of file systems, and must be used for
NFS clients, MFS, and FFM.
If you do not specify any of -p, -m, or -d, then -p is assumed.
-d domain The name of an AdvFS domain. Not valid when setting the FSBSIZE
attribute.
An AdvFS file system is mounted on a fileset in a domain. All
file systems in the domain must be served by the same member.
Because of this, certain AdvFS operations, such as manual
relocation, can be performed only on the whole domain.
-m filesystem
The mounted-from filesystem. For AdvFS, enter the domain and
fileset; for example, cluster_usr#usr. For UFS, enter the name
of the block device special file; for example, /dev/disk/dsk3c.
For NFS, enter the full path name and host in the same format
used to mount the file system, either path@host or host:path.
For an ISO 9660 CD-ROM file system (cdfs), enter the name of the
block device special file; for example, /dev/disk/cdrom0c.
-F [std | raw]
Specifies output format. The default, std, is intended to be read
by people. The output is clearly labeled and formatted for that
purpose.
The raw output is intended to be used as input to scripts. Data
is unlabeled. Data fields are separated by spaces. Data is
grouped into lines as follows:
Line 1, data about member on which the command was executed:
Member name.
Cluster member ID (an integer).
The date (year month day hour minute second).
For example:
mutt 2 1998 12 08 09 12 50
Line 2, data about the CFS entity:
Name of the the CFS entity: a file system or domain.
Type of CFS entity: AdvFS, ADV, UFS, NFS, CDFS, or
Unknown.
Class of CFS entity: FS or Domain.
Mount point (in response to the -d flag, which queries
a domain but not a fileset). Otherwise, -.
CFS Server name. This cluster member is the CFS server
of the entity.
Number of devices that create and sustain the CFS
entity.
Number of member-specific statistics (0 or 1).
For example:
cluster_var AdvFS Domain - jeff 1 1
Line 3, device data:
Device name.
Hardware ID.
Device type: 1=direct-access I/O device, 2=served
device, 0=unknown. Being a direct-access I/O device
does not ensure that the cluster member where the
cfsmgr command executed has direct I/O capability to
that device.
Device kind: 1=disk, 2=tape, 3=changer device,
0=unknown.
Whether it is a served device (T, or F).
Server name (or none).
Status of the server: 1=OK, 2=no service available for
the device, 3=cluster reconfiguration in progress,
4=device not handled by drdmgr, 0=status unknown.
Number of servers.
For example:
dsk4h 27 1 1 T mutt 1 2
Line 4, statistics on the CFS entity:
C (A parsing aid.)
Cluster member ID (an integer).
Client name.
Number of read operations.
Number of write operations.
Number of lookup operations.
Number of getattr operations.
Number of readlink operations.
Number of access operations.
Number of other operations.
For example:
C 2 jeff 2383 2863 96551 32448 121975 10169 253236
See the description of the STATISTICS attribute for
more information about CFS statistics.
DESCRIPTION
Use the cfsmgr command to query attributes of the cluster file system and
to set the SERVER and STATISTICS attributes. By setting the SERVER
attribute for various file systems, you can manually balance the loads on
the CFS servers.
Entering the cfsmgr command with no options returns the names of domains
and file systems, where each is mounted, the name of the server of each,
and the server status. When setting or querying a CFS attribute, you must
specify a file system.
If you request a file system relocation, cfsmgr verifies that all disks
used by the file system can be accessed locally from the target member. If
the disk check succeeds, the file system is moved to the target member. If
the check fails, the file system is not relocated, and a message is
displayed recommending against relocating the file system to the target
member. Use the -f flag if you want the relocation to take place even when
the disks used by the file system cannot be accessed locally from the
target member. Use the -r flag to move all underlying disks to the target
member. After the disks are being served by the target member, the file
system is moved to the new server.
No checks are made as to whether disks serving the subject file system are
in use by another file system. Relocating a served disk can mean that
another file system suddenly does remote disk I/O. It is up to the system
administrator to manage the disk pool appropriately.
If the mount path of a file system is configured to traverse the mount of
an inaccessible file system, the first file system becomes inaccessible as
well. In that case, cfsmgr displays an error message stating the reason for
the inaccessibility. It is up to the system administrator to arrange the
mount points in a way that maintains availability.
The values of attributes set with cfsmgr do not survive a reboot of the
host on which the attribute was set.
To set an attribute and have it survive a reboot, create a startup script
that calls cfsmgr to make the desired configuration.
The cfsmgr command includes a -u flag that forcibly unmounts AdvFS file
systems or domains that are not served by any cluster member. A domain or
file system could become not served if one or more storage elements
containing the domain or file system become unavailable. The forced unmount
operation fails if the file system or domain you specify is currently being
served. You cannot use this flag to unmount the root (/), usr, or var file
systems.
The cfsmgr command also includes a -K flag that provides a way to forcibly
unmount all AutoFS intercept points and auto-mounted file systems. This
flag is needed if AutoFS is stopped on a cluster member where there are
busy file systems, and then started on another member. In this case, the
busy AutoFS intercept points continue to recognize the cluster member as
the server, even after the autofsd daemon has been started on the other
cluster member. These intercept points do not allow new auto-mounts.
To correct this problem without halting the cluster member, use the -K flag
to forcibly unmount all AutoFS intercept points and auto-mounted file
systems served on the cluster member, even if the intercept points and
auto-mounted file systems are busy. This flag results in applications
receiving I/O errors for open files in affected file systems. An
application with its current working directory in an affected file system
will no longer be able to navigate the file system namespace using relative
names. See the Cluster Administration guide for additional information
before using this flag.
EXAMPLES
1. In the following example, cfsmgr gets the names of the servers of
/var. In the output, the null value for host name indicates that the
command ran on the local member.
# cfsmgr -v -a server /var
cfsmgr:
Domain or fs name = /var
cfsmgr:
cfs_attribute = 0, host name = (null)
cfsmgr:
op_flag = 0, clu_cfs_flag = 0x8
Server Name = sys1
Server Status : OK
2. In the following example, cfsmgr relocates /usr from the host mutt to
jeff.
# cfsmgr -h mutt -r -a SERVER=jeff /usr
3. The following example gets the CFS statistics for the root file
system:
# cfsmgr -a STATISTICS /
Counters for the filesystem /:
read_ops = 46435
write_ops = 105459
lookup_ops = 2367579
getattr_ops = 381126
readlink_ops = 4154614
access_ops = 75064263
other_ops = 5072410
Server Status : OK
EXIT VALUES
Upon successful completion the command returns 0 (zero).
In an attempt to relocate a file system, if all disks serving the file
system are not locally connected to the target member, the exit status is
1. The relocation does not occur unless the user specifies the -f flag.
If an error occurs, the exit status is greater than 1.
FILES
/sbin/cfsmgr
The command path.
SEE ALSO
Commands: drdmgr(8)
|
Index for
Section 8 |
|
|
Alphabetical
listing for C |
|
 |
Top of
page |
|