The GUI displays the current status of the cluster. To display more details about an item, select the item and click Properties . Figure 4-1 shows an example of the GUI.

Figure 4-1. Cluster Status GUI

In the CLI, enter an argument with =value to assign a value or alone (without a =value) to display the current setting. For example, the following displays the name of the cluster and the number of times the configuration has been changed:

# sgicm-config-cluster-cmd --cluster

cluster:
  name = SGI High Availability cluster
  config_viewnumber = 14

The names of the device files for filesystems to store quorum information must be the same on all cluster members. You must do the following:

SGI recommends that the two shared quorum partitions should be on different Fibre Channel controllers; ideally, they should be on separate Fibre Channel controllers at the front end, on separate HBAs on the Altix, and on separate RAID logical units (LUNs) or RAID arrays if possible. They should be at least 10 MB in size and the partition type must be Linux. For more information, see Appendix B, “Setting the Partition Type to Linux”.

You can use the lspci command to see the Fibre Channel cards and you can use the hwinfo --disk command to see information about the shared disks. For example:

# lspci
...
0000:05:02.0 Fibre Channel: QLogic Corp. QLA2300 64-bit Fibre Channel Adapter (rev 01)
...

# hwinfo --disk --short

disk:                                                           
  /dev/sda             SGI ST373307LC
  /dev/sdb             SGI TP9100 FFX2
  /dev/sdc             SGI TP9100 FFX2
  /dev/sdd             SGI TP9100 FFX2
  /dev/sde             SGI TP9100 FFX2
  /dev/sdf             SGI TP9100 FFX2
  /dev/sdg             SGI TP9100 FFX2
  /dev/sdh             SGI TP9100 FFX2

For more information, see the hwinfo(8) man page.

Partition 1 from the disk at Fibre Channel target 2 and partition 1 from the disk at Fibre Channel target 3 will be used for storing shared state. To assure that the devices have the same name on both cluster members, you could create symlinks to the block device files:

# ln -s /dev/sdb1  /dev/shared1
# ln -s /dev/sdc1  /dev/shared2

You should add these symlink commands to the /usr/lib/clumanager/create_device_links script on the cluster member. For more information, see the ln(1) man page.

The shared quorum partitions will then be referred to as /dev/shared1 and /dev/shared2. These partitions will be the primary and shadow, respectively, in the following examples.

To define the shared quorum partitions in the GUI configuration window, select the following from the Cluster Configuration window:

In the CLI:

sgicm-config-cluster-cmd --sharedstate \
                         --type=raw \
                         --rawprimary=path1 \
                         --rawshadow=path2

For example:

# sgicm-config-cluster-cmd --sharedstate --type=raw \
--rawprimary=/dev/shared1 --rawshadow=/dev/shared2

You should perform this step before defining the cluster (“Step 2: Create the Cluster”).

Do not modify the /etc/init.d/clumanager script.

To create the cluster in the GUI, type the cluster name in the Cluster Name field in cluster configuration window. The default cluster name is SGI High Availability cluster.

In the CLI:

sgicm-config-cluster-cmd --cluster --name "clustername"

To define a member in the GUI, do the following in the cluster configuration window:

In the CLI:

sgicm-config-cluster-cmd --add_member --name=membername

For each member, you must provide information about its power controller. The SGI Cluster Manager supports the L2 system controller using either serial cables or Ethernet cables.

---

Note: The Serial and Network power controllers shown in the GUI refer to third-party products that are not supported by SGI Cluster Manager. (Although the other machine in the cluster will appear in the Owner field automatically, this value is not used.) Do not confuse network-based power controllers with the l2network L2 Ethernet connection.

---

In the GUI Cluster Configuration window, select the member and click Add Child. The fields for SGI controllers are as follows:

Figure 4-2 shows an example for an L2 using the Ethernet network and Figure 4-3 shows an example in the GUI for an L2 using serial cables.

Figure 4-2. Power Controller Information for an L2 Using an Ethernet Network

---

Note: You can use a hostname in place of the IP address when configuring l2network reset provided that name resolution is in place; that is, the name is in /etc/hosts on all of the servers or is otherwise available via gethostbyname(2).

---

Figure 4-3. Power Controller Information for an L2 Using Serial Cables

In the CLI:

sgicm-config-cluster-cmd --member=membername \
                         --add_powercontroller \
                         --type=l2|l2network \
                               required only for l2network:
                               --ipaddress=L2_IPaddress_or_hostname \
                               --password=L2_password_(if_defined) \
                               --partition=Altix_partition_ID \
                          --device=/dev/ttyIOCx\
                          --partition=n

You can optionally set a password for the L2 to prevent unauthorized access to L2 functions via Ethernet. If you choose to use this security feature, SGI Cluster Manager must know the password in order to access L2 functionality. For more information, see SGI L1 and L2 Controller Software User's Guide.

For example, the following defines an L2 using the Ethernet method (therefore there is no --device argument):

# sgicm-config-cluster-cmd --member=member1  --add_powercontroller \
--type=l2network --ipaddress=192.168.0.100 --partition=3 --password=foo

For example, the following defines an L2 using the serial cable method:

# sgicm-config-cluster-cmd --member=member1  --add_powercontroller \
--type=l2 --device=/dev/ttyIOC0

For hardware information, see “Power Control” in Chapter 2.

You can modify the time it takes to detect a member failure, known as the failover speed.

---

Note: The default failover speed differs depending upon which tool (GUI or CLI) you use to define the cluster. You cannot change the value for failover speed while the cluster daemons are running.

---

Failover Speed and the GUI

In the GUI, you can supply the failover speed directly:

1. In the Cluster Configuration window, select the following:

   Cluster -> Daemon properties

2. Select the clumembd tab

3. Use the sliding bar to adjust failover speed, as shown in Figure 4-4. The GUI provides 15 seconds as the default failover speed value.

4. You can choose to enable either broadcast heartbeating or multicast heartbeating.

Figure 4-4. Adjusting Failover Speed

Failover Speed and the CLI

The clumembd daemon lets you specify the failover speed indirectly by defining the heartbeat interval and the timeout, from which the failover speed is automatically calculated:

• interval specifies the heartbeat interval, which is the number of microseconds before a heartbeat is sent to all other members in the cluster. The default value is 500000 (0.5 seconds).

• tko_count specifies the heartbeat timeout, which is the number of heartbeats missed before a member is declared as failed. The default value is 20.

---

Note: The GUI does not let you display or set the heartbeat interval or the heartbeat timeout individually.

---

The failover speed is calculated as follows:

interval_value * tko_count_value = failover_speed

Therefore, the default member failure detection time is 10 seconds (0.5 * 20 = 10).

Table 4-1 shows the failure detection times and parameter values that are supported.

Table 4-1. Supported Failure Detection Times and Parameter Values

Failover Speed (in seconds)	interval (in microseconds)	tko_count
30	1000000	30
25	1000000	25
20	1000000	20
15	750000	20
10	500000	20
5	330000	15

For example, the following command displays the heartbeat interval and tko_count values:

# sgicm-config-cluster-cmd --clumembd clumembd: loglevel = 5 interval = 500000 tko_count = 20 thread = yes broadcast = no multicast = yes multicast_ipaddress = 225.0.0.11

The failover speed is therefore 10 seconds. The following command changes the failover speed 15 seconds:

# sgicm-config-cluster-cmd --clumembd --interval=750000 --tko_count=20

---

Note: You cannot change the values for interval and tko_count while the cluster daemons are running.

---

For more information about using the command-line interface, see sgicm-config-cluster-cmd man page.

There are two types of tiebreakers:

In the GUI:

Figure 4-5 shows an example of the cluquorumd window.

Figure 4-5. Tiebreakers

In the CLI:

sgicm-config-cluster-cmd --cluquorumd \
                         --tiebreaker_ip=IPaddress \
                         --pinginterval=seconds

The failover domain is optional; if a failover domain is not defined, the service will be started on any member. For more information, see “Failover Domains” in Chapter 1.

In the GUI Cluster Configuration window:

For information about the failover and failback options, see “Failover Domains” in Chapter 1.

Figure 4-6 shows an example.

Figure 4-6. Failover Domain

In the CLI:

sgicm-config-cluster-cmd --add_failoverdomain \
                         --name=domainname  \
                         --restricted=yes|no \
                         --ordered=yes|no \
                         --controlled=yes|no

sgicm-config-cluster-cmd --failoverdomain=domainname \
                         --add_failoverdomainnode \
                         --name=membername

The default for --restricted, --ordered , and --controlled is no.

You can specify the following for a service (the service must be disabled in order to configure it):

In the GUI Cluster Configuration window:

Figure 4-7 shows an example of configuring an NFS high-availability service.

Figure 4-7. Configuring a High-Availability Service

In the CLI:

sgicm-config-cluster-cmd --add_service \
                         --name=servicename \
                         --failoverdomain=domainname \
                         --checkinterval=seconds \
                         --servicetimeout=seconds \
                         --monitorlevel="level" \
                         --restartcount=N \
                         --userscript=pathname

---

Note: The monitoring-level string values are case-sensitive and should be either of the following: "Check for processes" "Check as client"

---

In the GUI Cluster Configuration window:

In the CLI:

sgicm-config-cluster-cmd --service=servicename \
                         --add_service_ipaddress \ 
                         --ipaddress=IPaddress \
                         --netmask=netaddress \
                         --broadcast=broadcastaddress

In the GUI Cluster Configuration window:

In the CLI:

sgicm-config-cluster-cmd --service=servicename \
                         --add_device \
                         --name=path

sgicm-config-cluster-cmd --service=servicename \
                         --device=path \
                         --mount   \
                         --mountpoint=mountpoint \
                         --fstype=xfs|cxfs \
                         --options=mountoptions \
                         --forceunmount=yes

Samba share names must be unique within the cluster.

The Samba Druid is a configuration guide that lets you create a new Samba service or add Samba to an existing service. In the GUI Cluster Configuration window, start the Samba Druid by selecting the following:

For more information, see “Samba Druid Example”.

In the CLI:

sgicm-config-cluster-cmd --service=servicename \
                         --device=path \
                         --sharename=sharename

Define the NFS export point and NFS client information.

The NFS Druid is a configuration guide that lets you create a new NFS service or add NFS export points to an existing service. In the GUI Cluster Configuration window, start the NFS Druid by selecting the following:

For more information, see “NFS Druid Example”.

In the CLI:

sgicm-config-cluster-cmd --service=servicename \
                         --device=path \
                         --add_nfsexport \
                         --name=exportdirectory

sgicm-config-cluster-cmd --service=servicename \
                         --device=path \
                         --nfsexport=exportpath \
                         --add_client \
                         --name=\* \
                         --options=options

The value * for the NFS client name means “all NFS clients.” For better security, supply a list of NFS client systems instead of the * character. For more information, see the exports(5) man page.

---

Note: In general, you must use the fsid option to set the fsid value for the export. Any number in the range 1 through 65535 will work. See the exports(5) man page for further details on the fsid option.

---

If you are using the GUI, you must explicitly save the configuration information as noted in “Cluster Configuration Tools”. Select the following from the Cluster Configuration window:

During the initial configuration, you must manually copy the /etc/cluster.xml file to the other member in the cluster whether you use the GUI or the CLI.

Each member has an /etc/cluster.xml file that contains cluster configuration information. If you make a change to this file on one member, you must copy the file to the other member using a command such as scp(1).

After making configuration changes, you must verify that the configuration files across the cluster are in synchronization. To do this, you can run the following command on each node and compare the config_viewnumber value on each, which lists the configuration file version number:

sgicm-config-cluster-cmd --cluster

The config_viewnumber value is updated each time a change is made to the configuration file.

For example, the following output from Machine1 and Machine2 shows that the configuration files are in synchronization for test-cluster because they both have the same config_viewnumber value (10):

In another example, the following output from Machine1 and Machine2 shows that the configuration files are out synchronization because they have differentconfig_viewnumber values (10 and 11):

If the config_viewnumber values are different, then configuration files are different. You should copy the configuration file with higher config_viewnumber number (which indicates the more recent configuration file) to the other member. In this case, you would copy the configuration file from Machine2 (which has the higher number of 11) to Machine1 (which has the lower number of 10).

To automatically restart the SGI Cluster Manager daemons after a reboot, do the following in the CLI:

For more information, see Chapter 5, “Administration”.