[
https://issues.apache.org/jira/browse/CLOUDSTACK-9767?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15848207#comment-15848207
]
Dag Sonstebo commented on CLOUDSTACK-9767:
------------------------------------------
Initial analysis (from AP):
At the storage level in VolumeOrchestrator, If there is a cross pool migration
then, there is a check for the operation type and depending on the type which
could be HA or a non HA the decision is moved to whether storage migration is
enabled or not.
One issue that pops up here is the use of 'recreate.systemvm.enabled' and it
seems it applies it to any operation and does not specifically enforce it for
system vms.
The definition for this global config is: "If true, will recreate system vm
root disk whenever starting system vm"
It is possible that it ends up applying this setting for all kind of VMS, this
can be checked by setting it to false.
Looking at all the paths from where HA is invoked is much more complicated. For
VolumeOrchestrator to recognise an operation as a ha operation, vm operation
param indicating ha (VirtualMachineProfile.Param.HaOperation) needs to be
set.It is possible that it is being missed out on a particular flow of events,
resulting in the ha-operation being considered as a non-HA operation by the
volume orchestration layer.
In short the behaviour documented as expected is what it should be. This is
what is implemented where the global setting is getting used. But there are
other issues and variables that are affecting this and cause of
inconsistencies. This will require deeper investigation to narrow down to exact
cause that will include further testing.
> enable.ha.storage.migration not honored during HA failover
> ----------------------------------------------------------
>
> Key: CLOUDSTACK-9767
> URL: https://issues.apache.org/jira/browse/CLOUDSTACK-9767
> Project: CloudStack
> Issue Type: Bug
> Security Level: Public(Anyone can view this level - this is the
> default.)
> Affects Versions: 4.9.1.0
> Reporter: Dag Sonstebo
> Priority: Minor
>
> Environment:
> - CloudStack 4.9.1
> - KVM (CentOS 6.2)
> - 'cluster capacity' was 'full' to force failover
> Assumption is that HA will not be executed beyond the clusters if
> 'enable.ha.storage.migration=false' is set.
> Findings:
> A) If 'enable.storage.migration=true' is set, HA was executed beyond clusters
> when the HA test is conducted with KVM hosts
> B) If 'enable.storage.migration=false' is set, HA was not executed beyond
> clusters as expected when the HA test is conducted with KVM hosts.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
