HP Serviceguard for Linux Version A.11.19 Release Notes, April 2009

increases. See “About HPVM Virtual Machines and Cluster Re-formation Time”
(page 27).
• VMware nodes. In the case of VMware nodes, Persistent Reservations (PR) handle
the I/O issues. This entails the following restriction:
— Serviceguard does not support configurations with more than one VMware
virtual-machine node on the same host (“cluster-in-a-box”) because such
configurations are not supported by PR.
IMPORTANT: See “About Persistent Reservations” in Chapter 3 of the latest
version of Managing Serviceguard for Linux for more information about PR and
VMware nodes.
For more information about using VMware virtual machines as Serviceguard
cluster nodes, see Using Serviceguard for Linux with VMware Virtual Machines at
the address given under “For More Information” (page 28).
About HPVM Virtual Machines and Cluster Re-formation Time
General case
Whenever a node fails and the cluster re-forms, Serviceguard must wait a certain
amount of time to allow I/O from the failed node to be written out to the target storage
device. Only after that time has elapsed can Serviceguard allow an adoptive node access
to that device; otherwise data corruption could occur. The amount of time Serviceguard
waits is calculated by Serviceguard and is not user-configurable.
Additional time needed for HPVM nodes
The above is true whether or not the cluster includes HPVM virtual machines, but
using HPVM nodes increases the amount of time Serviceguard needs to wait before it
is safe to allow another node access to the same storage. This additional wait can increase
cluster re-formation time by as much as 70 seconds.
NOTE: This discussion applies only to HPVM nodes, not to VMware nodes.
Additional timeout calculation and hpvminfo
How much additional time Serviceguard needs to wait depends in part on whether or
not you have installed HP software including the hpvminfo command on the HPVM
node. For installation information, see “For More Information” (page 28).
Serviceguard uses information it derives from hpvminfo to set the timeout to the
optimal value. If any HPVM node does not have the hpvminfo software, Serviceguard
may not be able to obtain the information it needs to set the optimal timeout, and in
that case it sets the additional timeout to the maximum value, 70 seconds.
What’s in this Release 27