This section describes the availability of cluster system in the following environments in Linux.
Cluster system in the physical environment
Cluster system between guest OSes with the Host OS failover function(KVM)
Cluster system between guest OSes on multiple host OSes(KVM)
Cluster system between guest OSes on one host OS(KVM)
Cluster system between guest OSes on multiple ESX hosts(VMware)
Cluster system between guest OSes on one ESX host(VMware)
Monitoring target | Physical server | KVM | VMware | |||
|---|---|---|---|---|---|---|
Cluster system between guest OSes with the Host OS failover function | Cluster system between guest OSes on multiple host OSes | Cluster system between guest OSes on one host OS | Cluster system between guest OSes on multiple ESX hosts | Cluster system between guest OSes on one ESX host | ||
(1) Unit | Y | Y | Y | N | Y | N |
(2) Shared disk and path of disk access | Y | Y | Y | N | Y | N |
(3) Public LAN | Y | Y | Y | N | Y | N |
(4) OS (physical, host OS/ESX host) | Y | Y | Y*1 | N | Y*2 | N |
(5) OS (guest OS) | - | Y | Y | Y | Y*2 | N |
(6) Service (cluster application) | Y | Y | Y | Y | Y | Y |
Service continuity when an error occurs Y: Available, N: Unavailable
*1 When a system disk error occurs in the host OS, the guest OS is switched automatically. When a host OS error occurs, it becomes the LEFTCLUSTER state.
*2 If a hang-up is detected in a guest OS and the guest OS cannot be switched to the standby system, the guest OS becomes the LEFTCLUSTER state.
Figure 1.11 Physical environment
Figure 1.12 Virtual environment
For VMware environment, read "host OS" as "ESX host" or "ESXi host".
How to detect an error in the following targets to be monitored
Unit
For PRIMEQUEST, the asynchronous monitoring linked with Management Board (MMB) immediately detects a panic or a reset triggered by an error in CPU, memory, or others, and the service is switched to the standby system. For PRIMERGY, an error is detected by the heartbeat monitoring, and the service is switched to the standby system.
Shared disk and path of disk access
Combining with the volume management function (GDS), the system detects a failure of a disk access or disk access path (monitored by Gds resource) and the service is switched to the standby system when the disk cannot be accessed or an error occurs in the entire communication path of disk access.
Public LAN
Combining with the network multiplexing function (Global Link Services. Hereinafter, GLS), the system detects a failure of network adapter or a path in the public LAN (monitored by Gls resource) and the service is switched to the standby system when an error occurs in the entire communication path of network.
OS (physical and host OS/ESX host)
For the cluster system in the physical environment and the cluster system between guest OSes with the Host OS failover function (KVM), a hang-up of the OS is detected by the fixed-cycle monitoring of cluster interconnect (LAN) and the service is switched to the standby system.
For the cluster system between guest OSes on multiple host OSes (KVM), when an error occurs in the host OS, the target guest OS becomes the LEFTCLUSTER state.
Note: At the time of a system disk failure, the service can be switched automatically by changing the system disk to ext3 or ext4 and also specifying "errors=panic" to the mount option.
For the cluster system between guest OSes on multiple ESX hosts (VMware), when an error occurs in the ESX host, a hang-up of the target OS is detected and the service is switched to the standby system.
For the cluster system between guest OSes on one host OS (KVM, VMware), an error in the host OS/ESX host cannot be detected because these are single hosts.
OS (guest OS)
For the cluster system in the KVM environment, a hang-up of the guest OS is detected by the fixed-cycle monitoring of cluster interconnect (LAN) and the service is switched to the standby system.
For the cluster system between guest OSes on multiple ESX hosts (VMware), set the I/O fencing function. If cluster interconnects and other networks cannot communicate with other nodes, the service is switched to the standby system.
For the cluster system between guest OSes on one ESX host (VMware), when an error occurs in a guest OS, the guest OS becomes the LEFTCLUSTER state.
Service (cluster application)
When a resource error of the cluster application occurs, the service is switched to the standby system.
