4 Node Cluster( N:1 switchover)

[Rizwan Mohammed]

Hi

We are planning to configure 4 node cluster in this ,3 node will have active packages running and 4th node will be standby for all the 3 nodes

my question
1. Do we need to configure Lock disk for the 4 node cluster.

2. Can we configure in such a manner if the package fails on node1(it can be any node) to takeover to Node 4 and in the same time.Node 4 should not take over the package if the other node fails too ( in this case if two nodes fails )

3.What are the precautionary measure to be taken care for this type of configuration

If any one has configure this type.

[avsrini]

Hi Rizwan,

1. For 4 node cluster, it is not compulsary for a Lock Disk.

2. Yes, you can configure the 4th node not to fail-over more than one package. You can get the details of how to configure this, from the Configuration Manual at docs.hp.com.

3. You have to make sure that if more than one package is failing, then it can be started on some other node, other than node 4.

If you give us more info's like, how many packages totally are you planning to run, Disk configurations etc. then we will be able to help you in detail.

Happy clustering.
Srini.

[Vincent Farrugia]

Hello,

1. It is not compulsory, but HIGHLY RECOMMENDED.

2. Yes you can. In the package configuration file, select the MIN_PACKAGE_NODE instead of CONFIGURED_NODE policy (the default). This way, the packages will failover to the server which has the minimum number of packages running at the moment.

3. Precautionary measures... well since your servers will run more or less the same number of packages at any one time, you won't need extra CPUs/RAM like in the case where a great number of packages may be running on a single server.

HTH,
Vince

[Rizwan Mohammed]

Hi

Thanks for reply ,it is considered that ,one package will be running in each machines ,and if one fails the standby will takeover .

2. same time if other machine fails ,this package can be considered to be down and it should not be transfered to other machines ,is this type of configuration can be made.

3.my only worry is if two machines goes down ,is there would be major cluster formation problem ,as if it is not configured for LOCK disk

some details of configuration planned are

1. VGs names will be uniquely configured in all the machines.
2. all the disk will be seen in the 4 node,this will be a standby
3.there will be private HB for all four nodes and 2 redundant link for data

is some thing else need to be included ?

[Vincent Farrugia · Accepted solution]

Hello,

It is HIGHLY RECOMMENDED to do a cluster-lock disk. Even on a 4-node cluster.

If you want a certain package to fail completely when 2 servers fail, you just enter the names of the two servers instead of the 4 servers in the package configuration script. You will have to configure it as CONFIGURED_NODE though in this setup, not MIN_PACKAGE_NODE like I told you before.

HTH,
Vince

[Stephen Doud]

1) The cluster-lock disk is supported on up to 4 nodes. It is used only when an equal number of nodes remain active on either side of a heartbeat LAN outage. For instance, if a LAN break occured between two nodes and the other 2 nodes.
The cluster lock disk is NOT used when a break occurs between an unequal number of nodes (when one node has left the cluster previously) - say 1 node versus 3 nodes. In that situation, the majority (3 nodes) would form a new cluster and the lone 1-node would programatically reboot itself.

Though it is not required to configure a cluster lock disk for a 4-node cluster (in light of the fact that it would take a network outage between 2 nodes and the other or between 1 node and the other (both are a low-probability of occurence... it is recommended simply because the possibility exists.

This topic is discussed in detail in the Managing MC/ServiceGuard manual located at:
http://docs.hp.com/hpux/ha/index.html#ServiceGuard

2) Others have answered items 2 and 3.

[Rizwan Mohammed]

Hi

If i dont configured as minpackagenode ,when 2 nodes fails the 2 package will be shifted to the standby node ,i need to shift only one machine failure package at time ,if two machines fails i can take one machine in down state.

i guess i need to configure as min node package insetead of configured node.

Rizwan

[Stephen Doud]

setting FAILOVER_POLICY to MIN_PACKAGE_NODE does a sort of load balancing (see the explanation in the package configuration file and the manual). After a cluster reformation, with a package that must be adopted, the server with the fewest number of packages would take over the orphaned package (assuming it was designated as an adoptive node for that package, and proper package switching was enabled).

If you don't like the effect of MIN_PACKAGE_NODE , you still have the option of writing test statements in the package control script (customer_defined_run_cmds) which determines the state of the available nodes and packages and makes decisions which node should start a package.

-s.