1836475
Members
2216
Online
110101
Solutions
This is what I think you are asking:
"When all paths to a disk array that is used by a package in a Serviceguard cluster fail, why does the package not fail over to the other node?"
Assuming thats the case, here's my answer:
Because Serviceguard isn't designed to do that out of the box. Losing all paths to your storage represents multiple points of failure, and Serviceguard is designed to protect against single points of failure. POresumably you are using some form of MPIO to protect against link failure (such as LVM PVlinks or Secure Path).
If you *really* want Serviceguard to failover when it loses all connections to a disk or disks, it can be configured to do that. See the following link:
http://www13.itrc.hp.com/service/cki/docDisplay.do?docLocale=en&docId=emr_na-c01028614-5
However, most people when they understand the implications of failing over in this situation choose not to set this up, as it means you *must* always TOC (i.e. panic and reboot) the failing node (if you can't talk to the disk, you can't unmount the filesystems, so the other node wouldn't be able to get access to them - so the system has to be TOC'd to ensure it is no longer attached to the disks).
HTH
Duncan
I am an HPE Employee
"When all paths to a disk array that is used by a package in a Serviceguard cluster fail, why does the package not fail over to the other node?"
Assuming thats the case, here's my answer:
Because Serviceguard isn't designed to do that out of the box. Losing all paths to your storage represents multiple points of failure, and Serviceguard is designed to protect against single points of failure. POresumably you are using some form of MPIO to protect against link failure (such as LVM PVlinks or Secure Path).
If you *really* want Serviceguard to failover when it loses all connections to a disk or disks, it can be configured to do that. See the following link:
http://www13.itrc.hp.com/service/cki/docDisplay.do?docLocale=en&docId=emr_na-c01028614-5
However, most people when they understand the implications of failing over in this situation choose not to set this up, as it means you *must* always TOC (i.e. panic and reboot) the failing node (if you can't talk to the disk, you can't unmount the filesystems, so the other node wouldn't be able to get access to them - so the system has to be TOC'd to ensure it is no longer attached to the disks).
HTH
Duncan
I am an HPE Employee
