1

I need to configure a cluster with shared storage that can be moved from Node A to Node B and vica versa.

In case of failure of Node A, Node B should take over the IP address associated with the NFS service, take ownership of the shared disk, mount it and start the NFS server.

I am using SUSE Linux 11.4.

So far I am using the HA cluster package and NFS. NFS is sharing the drive from Node A, but if Node A goes down Node B stops working.

jasonwryan
  • 71,734
  • 34
  • 193
  • 226
mystery
  • 11
  • 2
  • What do you mean by "Node B stops working"? In an active-passive cluster, which you seem to describe, Node A provides service while Node B does nothing, being a backup. If Node A fails, Node B should start providing the service. Is this what you have problems with? – Ferenc Wágner Oct 31 '15 at 08:33
  • Also, I recommend following [Clusters from Scratch](http://clusterlabs.org/doc/en-US/Pacemaker/1.1/html/Clusters_from_Scratch/) substituting NFS for Apache, and reporting where you get stuck. – Ferenc Wágner Oct 31 '15 at 08:35
  • If Node A fails, Node B should start providing the service. this is what i want to achieve. actually when i am shutting down Node A and look at Node B it doesnt work. what i did is using NFS shared the mount device of Node A on Node B thats why when Node A goes down Node B stops working what i want to achieve is when Node A goes down mount point should be activated on Node B – mystery Oct 31 '15 at 08:54

1 Answers1

1

You might want to look at drbd, which provides RAID-1 like disk redundancy and sharing over the network.

Drbd is a block device which is designed to build high availability clusters by providing a virtual shared device which keeps disks in nodes synchronised using TCP/IP. This simulates RAID 1 but avoiding the use of uncommon hardware (shared SCSI buses or Fibre Channel). It is currently limited to fail-over HA clusters.

If you're actually running lots of virtualised servers rather than a beowulf-style HPC cluster, you might also want to take a look at google's ganeti, it can use LVM and DRBD to provide high-availability VMs that automatically migrate to other nodes if and when required.

Ganeti is a virtual server cluster management tool built on top of the Xen and KVM virtualization technologies and other free software. It provides an automated environment to manage highly available virtual machine instances.

It is designed to facilitate cluster management of virtual servers using commodity hardware, and to provide fast and simple recovery after physical failures. It can take care of operations such as redundant disk creation and management, operating system installation (in cooperation with OS-specific install scripts), startup, shutdown, and failover of instances between physical systems.

I haven't used ganeti in production but i experimented a lot with it a few years back before taking a job to work on an openstack cluster (using openstack infrastructure to build and run virtual HPC clusters for researchers at universities around the nation (australia)). Ganeti worked really well in my testing, I really admired how it was built and worked.

If you look around, you can even find an add-on module that lets it use ZFS iscsi exports instead of LVM+DRBD for the VM disks.

cas
  • 1
  • 7
  • 119
  • 185