For the following cases, GDS cannot determine the disk containing the latest data. In this case, a user must select the copy source disk for the synchronization copying to take the restoration procedure.
Both nodes are stopped after the operation with an INVALID status slice, and the cluster application is forcibly started in the other node with the INVALID status slice.
Both nodes are stopped after the standby node was stopped, and the cluster application is forcibly started with the node that was on standby.
After the communication between the nodes is disabled due to the cluster interconnect failure or slowdown of the server, the operational node is stopped and switched.
In the above cases, starting the netmirror volume and the synchronization copying are inhibited.
The restoration procedure for the above a and b is as follows.
Restoration Procedure
Start the stopped node.
Stop RMS on both nodes.
Restore the iSCSI device.
Take this procedure on the node where the cluster application is forcibly started last time.
# /etc/opt/FJSVsdx/bin/sdxiscsi_ctl -F -e init |
The disk of the stopped node must be recognized on the operating node. For -d option of the sdxfix command, specify the disk of the stopped node.
Take this procedure on the operating node.
# sdxfix -D -c class_name -d disk_name -x Devlabel |
-x Devlabel is the option to specify the restoration of the device information. Use this option only for this restoration procedure.
Decide the copy source slice for the synchronization copying.
See
When selecting the slice of a node on which a cluster application has last started as a copy source for the synchronization copying, see "Information: Verifying a node on which a cluster application has last started."
When verifying the contents of each slice and then deciding the copy source slice for the synchronization copying, see "Information: Verifying the contents of the slices that compose the netmirror volume."
If the copy source slice for the synchronization copying is in INVALID, restore the slice state. For -d option of the sdxfix command, specify the copy source disk for the synchronization copying.
Take this procedure on either of both nodes.
# sdxfix -V -c class_name -v volume_name -d disk_name -x NoRdchk |
-x NoRdchk is the option to skip the read check of the slice. Use this option only for this restoration procedure.
Cancel the lock volume of the netmirror volume.
Take this procedure on both nodes.
# sdxattr -V -c class_name -v volume_name -a lock=off |
Delete the iSCSI device information.
Take this procedure on both nodes.
# rm /var/opt/FJSVsdx/log/.sdxnetmirror_disable.db |
Restore the status of the copy destination disk for the synchronization copying.
Take this procedure on either of both nodes.
For -d option of the sdxswap command, within the disks in the netmirror group, specify the copy destination disk for the synchronization copying.
# sdxswap -O -c class_name -d disk_name |
Start RMS on both nodes.
In this case, the synchronization copying of the netmirror volume is performed.
Restoration Procedure
Decide the copy source slice for the synchronization copying.
See
When selecting the slice of a node on which a cluster application has last started as a copy source for the synchronization copying, see "Information: Verifying a node on which a cluster application has last started."
When verifying the contents of each slice and then deciding the copy source slice for the synchronization copying, see "Information: Verifying the contents of the slices that compose the netmirror volume."
If the copy source slice for the synchronization copying is not in INVALID, restore the iSCSI device information. For -d option of the sdxnetdisk command, specify the copy source disk for the synchronization copying.
Take this procedure on either of both nodes.
# /etc/opt/FJSVsdx/bin/sdxnetdisk -F -e devinfo -c class_name -v volume_name -d disk_name |
If the copy source slice for the synchronization copying is in INVALID, take the following steps.
3-1) Stop RMS on both nodes.
3-2) Restore the slice state.
For -d option of the sdxfix command, specify the copy source disk for the synchronization copying.
Take this procedure on either of both nodes.
# sdxfix -V -c class_name -v volume_name -d disk_name -x NoRdhk |
-x NoRdchk is the option to skip the read check of the slice. Use this option only for this restoration procedure.
Perform the synchronization copying.
Take this procedure on either of both nodes.
# sdxcopy -B -c class_name -v volume_name |
Restart the service if necessary.
Execute the following command on both nodes, and compare the output results of both nodes.
# /etc/opt/FJSVsdx/bin/sdxiscsi_ctl -L -e time -c class_name |
[Execution example]
Node 1
# /etc/opt/FJSVsdx/bin/sdxiscsi_ctl -L -e time -c class1
Wed Oct 26 19:13:16 2016 |
Node 2
# /etc/opt/FJSVsdx/bin/sdxiscsi_ctl -L -e time -c class1
Tue Oct 11 11:34:56 2016 |
The output displays the date and time the Gds resource of the class specified in the argument last started on the node where the command was executed.
In the above example, the last started cluster application was on Node 1.
Take the following steps to change the state of the slice whose contents you want to verify to STOP, and to change the state of the other slice to INVALID. After that, verify the contents of the volume and you can verify the contents of the slice.
1) Stop RMS on both nodes.
2) If the slice whose contents you want to verify is in INVALID, change the slice status.
For -d option of the sdxnetdisk command, specify the disk of the slice in INVALID.
Take this procedure on either of both nodes.
# /etc/opt/FJSVsdx/bin/sdxnetdisk -F -e slice -c class_name -v volume_name -d disk_name |
After executing this command, the slice whose contents you want to verify becomes STOP, and the other slice becomes INVALID.
3) Start the volume.
Take this procedure on either of both nodes.
# sdxvolume -N -c class_name -v volume_name -e unlock,mode=ro |
4) Verify the contents of the volume.
5) Stop the volume.
Perform this step on the node on which 3) of this procedure has been performed.
# sdxvolume -F -c class_name -v volume_name |
