Red Hat Bugzilla – Bug 970520
[RFE] prevent VMS from pausing when using allocated disks
Last modified: 2015-02-16 14:37:31 EST
Created attachment 756660 [details]
hook to set errorpolicy to report for all disks or selected ones
Description of problem:
when the storage domain of one of the disks of a VM experiences failure,
the VM will go on pause .
while this might be helpful when using thin provisioning, behaviour should be configurable when using preallocated disks, so the policy changes to report to OS and have the VM handle the lack of storage ( for instance using mdadm or ha-lvm with disks from other storage domains )
Version-Release number of selected component (if applicable):
have a storage domain fail
Steps to Reproduce:
1.have a storage domain fail
all vm with disks in this SD will go pausing
give the ability to report the I/O error
find attached a proposed before_vm_start hook that will change the errorpolicy to report upon launching vm
having this makes it possible to
-set red hat clusters using virtual machines
-use a Business Disaster Recovery solution using raid software (like mdadm ) for VMS , as long as they are presented with disks from 2 differents storage domains (one from both site)
what's the implication - will the disk move to error, or blocks trying to be written to will be marked 'bad blocks', which will stay that way when the storage comes back?
when we let OS handle the I/O error, either using mdadm or lvm, disk from the failed storage domain will be marked as such and OS will handle properly the situation with no impact at lvm level.
when disk gets back, standard OS procedures will be applied to sync back data.
Ayal - do you see a disk level field for this, a supported custom level property at disk level (now that we have device level custom properties), or just the external hook approach?
(In reply to Itamar Heim from comment #3)
> Ayal - do you see a disk level field for this, a supported custom level
> property at disk level (now that we have device level custom properties), or
> just the external hook approach?
Should be advanced per VM disk level field, similar to cache mode
*** This bug has been marked as a duplicate of bug 1024428 ***