Adding building blocks to an existing ESS cluster
A building block is made up of a pair of Power® 8 servers and one or more storage enclosures. When adding building blocks in an ESS cluster, perform the following steps. You can add more than one building blocks at a time.
Attention: Be careful when running this procedure on a production system. Failure to
follow the listed procedure might result in re-deployment of a running I/O server node, loss of
quorum, and/or data loss. Be especially carefully before running gssdeploy -d or
gsscheckdisks. It is always a good practice to back up the xCAT database prior to
starting this procedure and for the customer to back up their data, if possible.
- Power on and boot the new Power 8 servers within the building block(s). Do not power up the associated storage enclosures.
For example, a GL3C building block is made up of two Power 8 (PPC64LE) servers and three 4U106 enclosures. At this step, power on only the servers but do not apply power to the enclosures.
- Add the new building block to the /etc/hosts file.
- Find the new building block serial numbers. The subnet and mask are typically 10.0.0.1/24 by
default. However, check the customer FSP
network.
/var/tmp/gssdeploy -f subnet/mask
- Find rack positions that are used for
GUI.
/var/tmp/gssdeploy -i
- Update gssdeploy.cfg to make the following changes.
- Change
DEPLOYMENT_TYPE
toADD_BB
. - Change
GSS_GROUP
to something other thangss_ppc64
orces_ppc64
. For example,new_bb
. - Add the new serial numbers and node names to the gssdeploy.cfg file.
- Change
- Run gssdeploy -o to add the new building block to
xCAT.
/var/tmp/gssdeploy -o
- Run gssprecheck on the new node group and address any
issues.
/opt/ibm/gss/tools/samples/gssprecheck -G new_bb --install --file /var/tmp/gssdeploy.cfg
- Run gssdeploy -d to deploy the new building
block.
/var/tmp/gssdeploy -d
- After about 30 minutes, run health checks on the new building block and address any issues.
- gssstoragquickcheck -G new_bb
- gssfindmissingdisks -G new_bb
- GSSENV=INSTALL gsscheckdisks -G new_bb --encl all --iotest a --write-enable
- Perform the steps in this procedure: Set up the high-speed network.
- Run gssinstallcheck on the new group. Ignore any
GPFS related settings for
now.
xdsh new_bb "/opt/ibm/gss/tools/bin/gssinstallcheck -N localhost" | xcoll -n
- Add the new nodes to the
cluster.
gssaddnode -G new_bb --suffix=-hs --accept-license --nodetype gss --cluster-node low_speed_name_of_cluster_node
- Create recovery
groups.
mmvdisk server configure --nc new_node_class --recycle one ; mmvdisk rg create --rg newrg1>,<newrg2 --nc new_node_class
- Create vdisks and NSDs. Make sure that you match the block size and the RAID
code.
This is an example command. For best practices, see mmvdisk documentation in IBM Spectrum Scale RAID: Administration.mmvdisk vs define --vs new_vdisk_set --rg newrg1,newrg2 --code 8+2p --bs 8M --ss 100% --nsd-usage dataAndMetadata --sp system
- Reboot both new I/O server
nodes.
xdsh new_bb "systemctl reboot"
When both nodes are back up, you can start GPFS and move on to the next step.
- Add NSDs to an existing file system. For
example:
mmvdisk fs add --file-system existing_filesystem --vdisk-set new_vdisk_set
- Run restripe if needed.
mmresripefs -b
- Update the performance monitoring list by using the mmchnode command. For more information, see mmchnode command.
- Put the new nodes back in the
gss_ppc64
node group and delete the temporary group, and then comment out theGSS_GROUP
line ingssdeploy.cfg
.For example, add nodes togss_ppc64
as follows.chdef -t group gss_ppc64 -o gssio3,gssio4 -p
- Update call home for new building
blocks.
gsscallhomeconf -E ems1 -N ems1,gss_ppc64 --suffix=-hs --register=all
Note: Now that you have additional building blocks, ensure that file system metadata
replication is enabled (-m 2). The new pair of NSDs should be in a new failure group. The maximum
number of failure groups is 5.