Configure your local and multihost disks for Solstice DiskSuite software by using the procedures in this appendix, along with the information in Chapter 1, Planning the Sun Cluster Configuration. Refer to your Solstice DiskSuite documentation for additional details.
The following procedures are in this appendix.
The following table lists the tasks to perform to configure Solstice DiskSuite software for Sun Cluster configurations.
Table A-1 Task Map: Configuring Solstice DiskSuite for Sun Cluster Configurations
Task |
For Instructions, Go To ... |
---|---|
Plan the layout of your Solstice DiskSuite configuration. | |
Calculate the number of metadevice names and disksets needed for your configuration, and modify the /kernel/drv/md.conf file. | |
Create metadevice state database replicas on the local disks. | |
Optional. Mirror file systems on the root disk. | |
Create disksets using the metaset command. | |
Add disk drives to the disksets. | |
Repartition drives in a diskset to allocate space to slices 1 through 6. | |
List device ID pseudo-driver mappings and define metadevices in the /etc/lvm/md.tab files. | |
Initialize the md.tab files. | |
For dual-string configurations, configure mediator hosts, check the status of mediator data, and, if necessary, fix bad mediator data. |
"How to Add Mediator Hosts" "How to Check the Status of Mediator Data" "How to Fix Bad Mediator Data" |
Create and mount cluster file systems. |
This procedure describes how to calculate the quantity of metadevice names needed for your configuration, and how to modify the /kernel/drv/md.conf file.
The default number of metadevice names is 128. Many configurations need more than the default. Increasing this number before implementing a configuration saves administration time later.
Calculate the quantity of metadevice names needed by determining the largest of the metadevice names to be used in each diskset.
This requirement is based on the metadevice name value rather than on the actual quantity. For example, if your metadevice names range from d950 to d1000, Solstice DiskSuite software requires 1000 names, not 50.
The cluster can have a maximum of 8192 metadevice names per diskset.
Calculate the total expected number of disksets in the cluster, then add one for private disk management.
The cluster can have a maximum of 32 disksets.
Edit the /kernel/drv/md.conf file.
All cluster nodes (or cluster pairs in the cluster-pair topology) must have identical /kernel/drv/md.conf files, regardless of the number of disksets served by each node. Failure to follow this guideline can result in serious Solstice DiskSuite errors and possible loss of data.
If the calculated quantity of metadevice names exceeds 128, set the nmd field to the largest metadevice name value used in a diskset.
Set the md_nsets field to the quantity calculated in Step 2.
Changes to the /kernel/drv/md.conf file become operative after a reconfiguration reboot is performed.
To create local replicas, go to "How to Create Metadevice State Database Replicas".
Perform this procedure on each node in the cluster.
Become superuser on the cluster node.
Create replicas on one or more local disks for each cluster node by using the metadb command.
Refer to the metadb(1M) man page and your Solstice DiskSuite documentation for details.
To provide protection of metadevice state data, which is necessary to run Solstice DiskSuite software, create at least three replicas for each node. Also, placing replicas on more than one disk provides protection if one of the disks fails.
Verify the replicas.
# metadb |
The following example shows three metadevice state database replicas, each created on a different disk.
# metadb -af c0t0d0s7 c0t1d0s7 c1t0d0s7 # metadb flags first blk block count a u 16 1034 /dev/dsk/c0t0d0s7 a u 1050 1034 /dev/dsk/c0t1d0s7 a u 2084 1034 /dev/dsk/c1t0d0s7 |
If you want to mirror file systems on the root disk, go to "Mirroring the Root Disk". Otherwise, to create Solstice DiskSuite disksets, go to "How to Create a Diskset".
Mirroring the root disk prevents the cluster node itself from shutting down because of a system disk failure. Four types of file systems can reside on the root disk. Each file system type is mirrored by using a different method.
Use the following procedures to mirror each type of file system.
Some of the steps in these mirroring procedures can cause an error message similar to the following, which is harmless and can be ignored.
metainit: dg-schost-1: d1s0: not a metadevice |
For local disk mirroring, do not use /dev/global as the path when specifying the disk name. Specifying this path for anything other than cluster file systems prevents the system from booting.
Use this procedure to mirror the root (/) file system.
Become superuser on a node of the cluster.
Use the metainit(1M) command to put the root slice in a single-slice (one-way) concatenation.
# metainit -f submirror1 1 1 root-disk-slice |
Create a second concatenation.
# metainit -f submirror2 1 1 submirror-disk-slice |
Create a one-way mirror with one submirror.
The metadevice name for the mirror must be unique throughout the cluster.
# metainit mirror -m submirror1 |
Run the metaroot(1M) command.
This command edits the /etc/vfstab and /etc/system files so the system can be booted with the root (/) file system on a metadevice.
# metaroot mirror |
Run the lockfs(1M) command.
This command flushes all transactions out of the log and writes the transactions to the master file system on all mounted UFS file systems.
# lockfs -fa |
Evacuate any resource groups or device groups from the node.
# scswitch -S -h node |
Evacuates all resource groups and device groups
Specifies the name of the node from which to evacuate resource or device groups
Reboot the node.
# shutdown -g 0 -y -i 6 |
Use the metattach(1M) command to attach the second submirror to the mirror.
# metattach mirror submirror2 |
If the disk used to mirror the root disk is physically connected to more than one node (multiported), enable the localonly property of the raw disk device group for the disk used to mirror the root disk.
You must enable the localonly property to prevent unintentional fencing of a node from its boot device if the boot device is connected to multiple nodes.
If necessary, use the scdidadm -L command to display the full device ID (DID) pseudo-driver name of the raw disk device group.
In the following example, the raw disk device group name dsk/d2 is part of the third column of output, which is the full DID pseudo-driver name.
# scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 # scconf -c -D name=dsk/d2,localonly=true |
For more information about the localonly property, refer to the scconf_dg_rawdisk(1M) man page.
Use the scconf(1M) command to enable the localonly property.
# scconf -c -D name=rawdisk_groupname,localonly=true |
Specifies the name of the raw disk device group
Record the alternate boot path for possible future use.
# ls -l /dev/rdsk/root-disk-slice |
Repeat Step 1 through Step 11 on each remaining node of the cluster.
Ensure that each metadevice name for a mirror is unique throughout the cluster.
The following example shows creation of mirror d0 on the node phys-schost-1, which consists of submirror d10 on partition c0t0d0s0 and submirror d20 on partition c2t2d0s0. Disk c2t2d0 is a multiported disk, so the localonly property is enabled.
(Create the mirror:) # metainit -f d10 1 1 c0t0d0s0 d11: Concat/Stripe is setup # metainit -f d20 1 1 c2t2d0s0 d12: Concat/Stripe is setup # metainit d0 -m d10 d10: Mirror is setup # metaroot d0 # lockfs -fa (Reboot the node:) # scswitch -S -h phys-schost-1 # shutdown -g 0 -y -i 6 (Attach the second submirror:) # metattach d0 d20 d0: Submirror d20 is attached (Enable the localonly property of the mirrored disk's raw disk device group:) # scconf -c -D name=dsk/d2,localonly=true (Record the alternate boot path:) # ls -l /dev/rdsk/c2t2d0s0 lrwxrwxrwx 1 root root 57 Apr 25 20:11 /dev/rdsk/c2t2d0s0 -> ../../devices/node@1/pci@1f,0/pci@1/scsi@3,1/disk@2,0:a,raw |
To mirror the global namespace, /global/.devices/node@nodeid, go to "How to Mirror the Global Namespace".
Use this procedure to mirror the global namespace, /global/.devices/node@nodeid.
Become superuser on a node of the cluster.
Put the global namespace slice in a single slice (one-way) concatenation.
# metainit -f submirror1 1 1 disk-slice |
Create a second concatenation.
# metainit -f submirror2 1 1 submirror-disk-slice |
Create a one-way mirror with one submirror.
The metadevice name for the mirror must be unique throughout the cluster.
# metainit mirror -m submirror1 |
Attach the second submirror to the mirror.
This attachment starts a sync of the submirrors.
# metattach mirror submirror2 |
Edit the /etc/vfstab file entry for the /global/.devices/node@nodeid file system.
Replace the names in the device to mount and device to fsck columns with the mirror name.
# vi /etc/vfstab #device device mount FS fsck mount mount #to mount to fsck point type pass at boot options # /dev/md/dsk/mirror /dev/md/rdsk/mirror /global/.devices/node@nodeid ufs 2 no global |
Repeat Step 1 through Step 6 on each remaining node of the cluster.
Ensure that each metadevice name for a mirror is unique throughout the cluster.
Wait for the sync of the mirrors, started in Step 5, to complete.
Use the metastat(1M) command to view mirror status.
# metastat mirror |
If the disk used to mirror the global namespace is physically connected to more than one node (multiported), enable the localonly property of the raw disk device group for the disk used to mirror the global namespace.
You must enable the localonly property to prevent unintentional fencing of a node from its boot device if the boot device is connected to multiple nodes.
If necessary, use the scdidadm -L command to display the full device ID (DID) pseudo-driver name of the raw disk device group.
In the following example, the raw disk device group name dsk/d2 is part of the third column of output, which is the full DID pseudo-driver name.
# scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 # scconf -c -D name=dsk/d2,localonly=true |
For more information about the localonly property, refer to the scconf_dg_rawdisk(1M) man page.
Use the scconf(1M) command to enable the localonly property.
# scconf -c -D name=rawdisk_groupname,localonly=true |
Specifies the name of the raw disk device group
The following example shows creation of mirror d101, which consists of submirror d111 on partition c0t0d0s3 and submirror d121 on partition c2t2d0s3. The /etc/vfstab file entry for /global/.devices/node@1 is updated to use the mirror name d101. Disk c2t2d0 is a multiported disk, so the localonly property is enabled.
(Create the mirror:) # metainit -f d111 1 1 c0t0d0s3 d111: Concat/Stripe is setup # metainit -f d121 1 1 c2t2d0s3 d121: Concat/Stripe is setup # metainit d101 -m d111 d101: Mirror is setup # metattach d101 d121 d101: Submirror d121 is attached (Edit the /etc/vfstab file:) # vi /etc/vfstab #device device mount FS fsck mount mount #to mount to fsck point type pass at boot options # /dev/md/dsk/d101 /dev/md/rdsk/d101 /global/.devices/node@1 ufs 2 no global (View the sync status:) # metastat d101 d101: Mirror Submirror 0: d111 State: Okay Submirror 1: d121 State: Resyncing Resync in progress: 15 % done ... (Identify the DID name of the mirrored disk's raw disk device group:) # scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c2t2d0 /dev/did/rdsk/d2 (Enable the localonly property of the mirrored disk's raw disk device group:) # scconf -c -D name=dsk/d2,localonly=true |
To mirror file systems that cannot be unmounted during normal system usage, such as /usr, /opt, or swap, go to "How to Mirror File Systems That Cannot Be Unmounted". To mirror user-defined file systems, go to "How to Mirror User-Defined File Systems".
Otherwise, to create a diskset, go to "How to Create a Diskset".
Use this procedure to mirror file systems that cannot be unmounted during normal system usage, such as /usr, /opt, or swap.
Become superuser on a node of the cluster.
Put the slice on which an unmountable file system resides in a single slice (one-way) concatenation.
# metainit -f submirror1 1 1 disk-slice |
Create a second concatenation.
# metainit -f submirror2 1 1 submirror-disk-slice |
Create a one-way mirror with one submirror.
The metadevice name for the mirror does not need to be unique throughout the cluster.
# metainit mirror -m submirror1 |
Repeat Step 1 through Step 4 for each unmountable file system to be mirrored.
On each node, edit the /etc/vfstab file entry for each unmountable file system you mirrored.
Replace the names in the device to mount and device to fsck columns with the mirror name.
# vi /etc/vfstab #device device mount FS fsck mount mount #to mount to fsck point type pass at boot options # /dev/md/dsk/mirror /dev/md/rdsk/mirror /file-system ufs 2 no global |
Evacuate any resource groups or device groups from the node.
# scswitch -S -h node |
Evacuates all resource groups and device groups
Specifies the name of the node from which to evacuate resource or device groups
Reboot the node.
# shutdown -g 0 -y -i 6 |
Attach the second submirror to each mirror.
This attachment starts a sync of the submirrors.
# metattach mirror submirror2 |
Wait for the sync of the mirrors, started in Step 9, to complete.
Use the metastat(1M) command to view mirror status.
# metastat mirror |
If the disk used to mirror the unmountable file system is physically connected to more than one node (multiported), enable the localonly property of the raw disk device group for the disk used to mirror the unmountable file system.
You must enable the localonly property to prevent unintentional fencing of a node from its boot device if the boot device is connected to multiple nodes.
If necessary, use the scdidadm -L command to display the full device ID (DID) pseudo-driver name of the raw disk device group.
In the following example, the raw disk device group name dsk/d2 is part of the third column of output, which is the full DID pseudo-driver name.
# scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 # scconf -c -D name=dsk/d2,localonly=true |
For more information about the localonly property, refer to the scconf_dg_rawdisk(1M) man page.
Use the scconf(1M) command to enable the localonly property.
# scconf -c -D name=rawdisk_groupname,localonly=true |
Specifies the name of the raw disk device group
The following example shows the creation of mirror d1 on the node phys-schost-1 to mirror /usr, which resides on c0t0d0s1. Mirror d1 consists of submirror d11 on partition c0t0d0s1 and submirror d21 on partition c2t2d0s1. The /etc/vfstab file entry for /usr is updated to use the mirror name d1. Disk c2t2d0 is a multiported disk, so the localonly property is enabled.
(Create the mirror:) # metainit -f d11 1 1 c0t0d0s1 d11: Concat/Stripe is setup # metainit -f d21 1 1 c2t2d0s1 d21: Concat/Stripe is setup # metainit d1 -m d11 d1: Mirror is setup (Edit the /etc/vfstab file:) # vi /etc/vfstab #device device mount FS fsck mount mount #to mount to fsck point type pass at boot options # /dev/md/dsk/d1 /dev/md/rdsk/d1 /usr ufs 2 no global (Reboot the node:) # scswitch -S -h phys-schost-1 # shutdown -g 0 -y -i 6 (Attach the second submirror:) # metattach d1 d21 d1: Submirror d21 is attached (View the sync status:) # metastat d1 d1: Mirror Submirror 0: d11 State: Okay Submirror 1: d21 State: Resyncing Resync in progress: 15 % done ... (Identify the DID name of the mirrored disk's raw disk device group:) # scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c2t2d0 /dev/did/rdsk/d2 (Enable the localonly property of the mirrored disk's raw disk device group:) # scconf -c -D name=dsk/d2,localonly=true |
To mirror user-defined file systems, go to "How to Mirror User-Defined File Systems". Otherwise, to create a diskset, go to "How to Create a Diskset".
Use this procedure to mirror user-defined file systems. In this procedure, nodes do not need to be rebooted.
Become superuser on a node of the cluster.
Put the slice on which a user-defined file system resides in a single slice (one-way) concatenation.
# metainit -f submirror1 1 1 disk-slice |
Create a second concatenation.
# metainit -f submirror2 1 1 submirror-disk-slice |
Create a one-way mirror with one submirror.
The metadevice name for the mirror does not need to be unique throughout the cluster.
# metainit mirror -m submirror1 |
Repeat Step 1 through Step 4 for each user-defined file system to be mirrored.
On each node, edit the /etc/vfstab file entry for each user-defined file system you mirrored.
Replace the names in the device to mount and device to fsck columns with the mirror name.
# vi /etc/vfstab #device device mount FS fsck mount mount #to mount to fsck point type pass at boot options # /dev/md/dsk/mirror /dev/md/rdsk/mirror /file-system ufs 2 no global |
Attach the second submirror to the mirror.
This attachment starts a sync of the submirrors.
# metattach mirror submirror2 |
Wait for the sync of the mirrors, started in Step 7, to complete.
Use the metastat(1M) command to view mirror status.
# metastat mirror |
If the disk used to mirror the user-defined file system is physically connected to more than one node (multiported), enable the localonly property of the raw disk device group for the disk used to mirror the user-defined file system.
You must enable the localonly property to prevent unintentional fencing of a node from its boot device if the boot device is connected to multiple nodes.
If necessary, use the scdidadm -L command to display the full device ID (DID) pseudo-driver name of the raw disk device group.
In the following example, the raw disk device group name dsk/d4 is part of the third column of output, which is the full DID pseudo-driver name.
# scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 # scconf -c -D name=dsk/d2,localonly=true |
For more information about the localonly property, refer to the scconf_dg_rawdisk(1M) man page.
Use the scconf(1M) command to enable the localonly property.
# scconf -c -D name=rawdisk_groupname,localonly=true |
Specifies the name of the raw disk device group
The following example shows creation of mirror d4 to mirror /home, which resides on c0t0d0s4. Mirror d4 consists of submirror d14 on partition c0t0d0s4 and submirror d24 on partition c2t2d0s4. The /etc/vfstab file entry for /home is updated to use the mirror name d4. Disk c2t2d0 is a multiported disk, so the localonly property is enabled.
(Create the mirror:) # metainit -f d14 1 1 c0t0d0s4 d14: Concat/Stripe is setup # metainit -f d24 1 1 c2t2d0s4 d24: Concat/Stripe is setup # metainit d4 -m d14 d4: Mirror is setup (Edit the /etc/vfstab file:) # vi /etc/vfstab #device device mount FS fsck mount mount #to mount to fsck point type pass at boot options # /dev/md/dsk/d4 /dev/md/rdsk/d4 /home ufs 2 no global (Attach the second submirror:) # metattach d4 d24 d4: Submirror d24 is attached (View the sync status:) # metastat d4 d4: Mirror Submirror 0: d14 State: Okay Submirror 1: d24 State: Resyncing Resync in progress: 15 % done ... (Identify the DID name of the mirrored disk's raw disk device group:) # scdidadm -L ... 1 phys-schost-3:/dev/rdsk/c2t2d0 /dev/did/rdsk/d2 (Enable the localonly property of the mirrored disk's raw disk device group:) # scconf -c -D name=dsk/d2,localonly=true |
To create a diskset, go to "How to Create a Diskset".
Perform this procedure for each diskset in the cluster.
If configured with exactly two disk strings, the diskset must connect to exactly two nodes and use exactly two mediator hosts, which must be the same two hosts used for the diskset. Refer to "Mediators Overview" for details on setting up mediators.
If configured with more than two disk strings, ensure that for any two disk strings S1 and S2, the sum of the number of disks on those strings exceeds the number of disks on the third string S3. Stated as a formula, the requirement is that count(S1) + count(S2) > count(S3).
If the /.rhosts file is not used to list the cluster node names, ensure that root is a member of group 14.
# vi /etc/group ... sysadmin::14:root ... |
Make sure the local metadevice state database replicas exist.
If necessary, refer to the procedure "How to Create Metadevice State Database Replicas".
Become superuser on the cluster node that will master the diskset.
Create the diskset.
This command also registers the diskset as a Sun Cluster disk device group.
# metaset -s setname -a -h nodename1 nodename2 |
Specifies the diskset name
Adds (creates) the diskset
Specifies the name of the primary node to master the diskset
Specifies the name of the secondary node to master the diskset
Verify the status of the new diskset.
# metaset -s setname |
The following command creates two disksets, dg-schost-1 and dg-schost-2, assigning the nodes phys-schost-1 and phys-schost-2 as the potential primaries.
# metaset -s dg-schost-1 -a -h phys-schost-1 phys-schost-2 # metaset -s dg-schost-2 -a -h phys-schost-1 phys-schost-2 |
To add drives to the diskset, go to "How to Add Drives to a Diskset".
When a disk drive is added to a diskset, Solstice DiskSuite repartitions it as follows so that the metadevice state database for the diskset can be placed on the drive.
A small portion of each drive is reserved in slice 7 for use by Solstice DiskSuite software. The remainder of the space on each drive is placed into slice 0.
Drives are repartitioned when they are added to the diskset only if slice 7 is not set up correctly.
Any existing data on the disks is lost by the repartitioning.
If slice 7 starts at cylinder 0, and the disk is large enough to contain a state database replica, the disk is not repartitioned.
Become superuser on the node.
Ensure that the diskset has been created.
If necessary, refer to "How to Create a Diskset".
List the device ID (DID) mappings.
# scdidadm -L |
Choose drives that are shared by the cluster nodes that will master or potentially master the diskset.
Use the full DID pseudo-driver names when adding drives to a diskset.
The first column of output is the DID instance number, the second column is the full path (physical path), and the third column is the full DID pseudo-driver name (pseudo path). A shared drive has more than one entry for the same DID instance number.
In the following example, the entries for DID instance number 2 indicate a drive that is shared by phys-schost-1 and phys-schost-2, and the full DID name is /dev/did/rdsk/d2.
1 phys-schost-1:/dev/rdsk/c0t0d0 /dev/did/rdsk/d1 2 phys-schost-1:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 2 phys-schost-2:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 3 phys-schost-1:/dev/rdsk/c1t2d0 /dev/did/rdsk/d3 3 phys-schost-2:/dev/rdsk/c1t2d0 /dev/did/rdsk/d3 ... |
Take ownership of the diskset.
# metaset -s setname -t |
Specifies the diskset name
Takes ownership of the diskset
Add the drives to the diskset.
Use the full DID pseudo-driver name.
Do not use the lower-level device name (cNtXdY) when adding a drive to a diskset. Because the lower-level device name is a local name and not unique throughout the cluster, using this name might prevent the metaset from being able to switch over.
# metaset -s setname -a DIDname |
Adds the disk drive to the diskset
Verify the status of the diskset and drives.
# metaset -s setname |
The metaset command adds the disk drives /dev/did/dsk/d1 and /dev/did/dsk/d2 to the diskset dg-schost-1.
# metaset -s dg-schost-1 -a /dev/did/dsk/d1 /dev/did/dsk/d2 |
To repartition drives for use in metadevices, go to "How to Repartition Drives in a Diskset".
The metaset(1M) command repartitions drives in a diskset so that a small portion of each drive is reserved in slice 7 for use by Solstice DiskSuite software. The remainder of the space on each drive is placed into slice 0. To make more effective use of the disk, use this procedure to modify the disk layout. Allocating space to slices 1 through 6 enables you to use these slices when setting up metadevices.
Become superuser on the cluster node.
Use the format(1M) command to change the disk partitioning for each drive in the diskset.
When you repartition a drive, you must meet the following conditions to prevent the metaset(1M) command from repartitioning the disk.
Create a partition 7 starting at cylinder 0 that is large enough to hold a state database replica (approximately 2 Mbytes).
Set the Flag field in slice 7 to V_UNMT (unmountable) and do not set it to read-only.
Do not allow slice 7 to overlap any other slice on the disk.
Refer to the format(1M) man page for details.
To define metadevices by using an md.tab file, go to "How to Create an md.tab File".
Create an /etc/lvm/md.tab file for each node in the cluster.
When using Solstice DiskSuite software, ensure that local metadevices have names that are distinct from the device ID (DID) names used to form disksets. For example, if the DID name /dev/did/dsk/d3 is used in a diskset, do not use the name /dev/md/dsk/d3 in configuring local metadevices.
Become superuser on the cluster node.
List the DID mappings for reference when creating your md.tab file.
Use the full DID pseudo-driver names in the md.tab file in place of the lower-level device names (cNtXdY).
# scdidadm -L |
In the following example, the first column of output is the DID instance number, the second column is the full path (physical path), and the third column is the full DID pseudo-driver name (pseudo path).
1 phys-schost-1:/dev/rdsk/c0t0d0 /dev/did/rdsk/d1 2 phys-schost-1:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 2 phys-schost-2:/dev/rdsk/c1t1d0 /dev/did/rdsk/d2 3 phys-schost-1:/dev/rdsk/c1t2d0 /dev/did/rdsk/d3 3 phys-schost-2:/dev/rdsk/c1t2d0 /dev/did/rdsk/d3 ... |
Create an /etc/lvm/md.tab file and edit it by hand with your preferred text editor.
Refer to the Solstice DiskSuite documentation for details on creating an md.tab file.
The following sample md.tab file defines the metadevices for the diskset named dg-schost-1. The ordering of lines in the md.tab file is not important.
dg-schost-1/d0 -t dg-schost-1/d1 dg-schost-1/d4 dg-schost-1/d1 -m dg-schost-1/d2 dg-schost-1/d2 1 1 /dev/did/rdsk/d1s4 dg-schost-1/d3 1 1 /dev/did/rdsk/d55s4 dg-schost-1/d4 -m dg-schost-1/d5 dg-schost-1/d5 1 1 /dev/did/rdsk/d3s5 dg-schost-1/d6 1 1 /dev/did/rdsk/d57s5 |
The sample md.tab file is constructed as follows.
The first line defines the trans metadevice d0 to consist of a master (UFS) metadevice d1 and a log device d4. The -t signifies this is a trans metadevice. The master and log devices are specified by their position after the -t flag.
dg-schost-1/d0 -t dg-schost-1/d1 dg-schost-1/d4 |
The second line defines the master device as a mirror of the metadevices. The -m in this definition signifies a mirror device, and one of the submirrors, d2, is associated with the mirror device, d1.
dg-schost-1/d1 -m dg-schost-1/d2 |
The fifth line similarly defines the log device, d4, as a mirror of metadevices.
dg-schost-1/d4 -m dg-schost-1/d5 |
The third line defines the first submirror of the master device, d2, as a one-way stripe.
dg-schost-1/d2 1 1 /dev/did/rdsk/d1s4 |
The fourth line defines the second submirror of the master device, d3.
dg-schost-1/d3 1 1 /dev/did/rdsk/d55s4 |
Finally, the log device submirrors, d5 and d6, are defined. In this example, simple metadevices for each submirror are created.
dg-schost-1/d5 1 1 /dev/did/rdsk/d3s5 dg-schost-1/d6 1 1 /dev/did/rdsk/d57s5 |
If you have existing data on the disks that will be used for the submirrors, you must back up the data before metadevice setup and restore it onto the mirror.
To activate the metadevices defined in the md.tab files, go to "How to Activate Metadevices".
Ensure that you meet the following requirements.
You have ownership of the diskset on the node where the command will be executed.
The md.tab files are located in the /etc/lvm directory.
Become superuser on the cluster node.
Take ownership of the diskset.
# metaset -s setname -t |
Specifies the diskset name
Takes ownership of the diskset
Activate the diskset's metadevices, which are defined in the md.tab file.
# metainit -s setname -a |
Activates all metadevices in the md.tab file
For each master and log device, attach the second submirror (submirror2).
When the metadevices in the md.tab file are activated, only the first submirror (submirror1) of the master and log devices is attached, so submirror2 must be attached by hand.
# metattach mirror submirror2 |
Repeat Step 4 and Step 5 for each diskset in the cluster.
If necessary, run the metainit(1M) command from another node that has connectivity to the disks. This step is required for cluster-pair topologies, where the disks are not accessible by all nodes.
Check the status of the metadevices.
# metastat -s setname |
In the following example, all metadevices defined in the md.tab file for diskset dg-schost-1 are activated. Then the second submirrors of master device dg-schost-1/d1 and log device dg-schost-1/d4 are activated.
# metainit -s dg-schost-1 -a # metattach dg-schost-1/d1 dg-schost-1/d3 # metattach dg-schost-1/d4 dg-schost-1/d6 |
If your cluster contains disksets configured with exactly two disk enclosures and two nodes, those disksets require mediators. To add mediator hosts, go to "Mediators Overview". Otherwise, to create a cluster file system, go to "How to Add Cluster File Systems".
A mediator, or mediator host, is a cluster node that stores mediator data. Mediator data provides information on the location of other mediators and contains a commit count that is identical to the commit count stored in the database replicas. This commit count is used to confirm that the mediator data is in sync with the data in the database replicas.
Mediators are required for all Solstice DiskSuite disksets configured with exactly two disk strings and two cluster nodes. A disk string consists of a disk enclosure, its physical disks, cables from the enclosure to the node(s), and the interface adapter cards. The use of mediators enables the Sun Cluster software to ensure that the most current data is presented in the instance of a single-string failure in a dual-string configuration. The following rules apply to dual-string configurations using mediators.
Disksets must be configured with exactly two mediator hosts, and those two mediator hosts must be the same two cluster nodes used for the diskset.
A diskset cannot have more than two mediator hosts.
Mediators cannot be configured for disksets that do not meet the two-string and two-host criteria.
These rules do not require that the entire cluster have exactly two nodes. Rather, they only require that those disksets that have two disk strings must be connected to exactly two nodes. An N+1 cluster and many other topologies are permitted under these rules.
Perform this procedure if your configuration requires mediators.
Become superuser on the node currently mastering the diskset to which you want to add mediator hosts.
Run the metaset(1M) command to add each node with connectivity to the diskset as a mediator host for that diskset.
# metaset -s setname -a -m mediator_host_list |
Specifies the diskset name
Adds to the diskset
Specifies the name of the node to add as a mediator host for the diskset
See the mediator(7) man page for details about mediator-specific options to the metaset command.
The following example adds the nodes phys-schost-1 and phys-schost-2 as mediator hosts for the diskset dg-schost-1. Both commands are run from the node phys-schost-1.
# metaset -s dg-schost-1 -a -m phys-schost-1 # metaset -s dg-schost-1 -a -m phys-schost-2 |
To check the status of mediator data, go to "How to Check the Status of Mediator Data".
Run the medstat command.
# medstat -s setname |
Specifies the diskset name
If the value in the Status field is Bad, repair the affected mediator host by following the procedure "How to Fix Bad Mediator Data".
See the medstat(1M) man page for more information.
If the mediator data for one of the mediator hosts is bad, go to "How to Fix Bad Mediator Data" to fix the problem. Otherwise, to create a cluster file system, go to "How to Add Cluster File Systems".
Perform this procedure to repair bad mediator data. To determine the status of mediator data, follow the procedure "How to Check the Status of Mediator Data".
Become superuser on the node that owns the affected diskset.
Remove the mediator host(s) with bad mediator data from all affected disksets.
# metaset -s setname -d -m mediator_host_list |
Specifies the diskset name
Deletes from the diskset
Specifies the name of the node to remove as a mediator host for the diskset
Restore the mediator host.
# metaset -s setname -a -m mediator_host_list |
Adds to the diskset
Specifies the name of the node to add as a mediator host for the diskset
See the mediator(7) man page for details about mediator-specific options to the metaset command.
To create a cluster file system, go to "How to Add Cluster File Systems".
The following example helps to explain the process for determining the number of disks to place in each diskset when using Solstice DiskSuite software. This example assumes you are using three storage devices. In this example, existing applications are running over NFS (two file systems of 5 Gbytes each) and two Oracle databases (one 5 Gbytes and one 10 Gbytes).
The following table shows the calculations used to determine the number of drives needed in the sample configuration. If you have three storage devices, you would need 28 drives that would be divided as evenly as possible among each of the three storage devices. Note that the 5-Gbyte file systems were given an additional Gbyte of disk space because the number of disks needed was rounded up.
Table A-2 Determining Drives Needed for a Configuration
Use |
Data |
Disk Storage Needed |
Drives Needed |
---|---|---|---|
nfs1 |
5 Gbytes |
3x2.1 Gbyte disks * 2 (Mirror) |
6 |
nfs2 |
5 Gbytes |
3x2.1 Gbyte disks * 2 (Mirror) |
6 |
oracle1 |
5 Gbytes |
3x2.1 Gbyte disks * 2 (Mirror) |
6 |
oracle2 |
10 Gbytes |
5x2.1 Gbyte disks * 2 (Mirror) |
10 |
The following table shows the allocation of drives among the two disksets and four data services.
Table A-3 Division of Disksets
Diskset |
Data Services |
Disks |
Storage Device 1 |
Storage Device 2 |
Storage Device 3 |
---|---|---|---|---|---|
dg-schost-1 |
nfs1/oracle1 |
12 |
4 |
4 |
4 |
dg-schost-2 |
nfs2/oracle2 |
16 |
5 |
6 |
5 |
Initially, four disks on each storage device (a total of 12 disks) are assigned to dg-schost-1, and five or six disks on each (a total of 16) are assigned to dg-schost-2.
No hot spare disks are assigned to either diskset. A minimum of one hot spare disk per storage device per diskset enables one drive to be hot spared (restoring full two-way mirroring).