[
https://issues.apache.org/jira/browse/CLOUDSTACK-8848?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
René Moser updated CLOUDSTACK-8848:
-----------------------------------
Description:
In some conditions, VR gets rebooted after a out of band migration was done on
vCenter.
*Note, new global setting in 4.5.2 "VR reboot after out of band migration" is
set to false and this looks more like a bug.*
After a VR migration to a host _and_ when the VM power state report gathering
is running, the VR (and also any user VM as well) will get into the
"PowerReportMissing" because it is not on the expected host anymore. This is
ok.
But if the VM is a VR. it will be stopped and started again. That is what we
see.
In can not be reproduced every time a migration was done, but it seems the
problem is "powerReportMissing".
I grep-ed the source and found this line related
https://github.com/apache/cloudstack/blob/master/engine/orchestration/src/com/cloud/vm/VirtualMachineManagerImpl.java#L3702
and also it seems that the graceful period might be also related,
https://github.com/apache/cloudstack/blob/master/engine/orchestration/src/com/cloud/vm/VirtualMachinePowerStateSyncImpl.java#L110
In case it is a user VM, we see in the logs that the state will be set to
power-off, but the VM keeps running. After a while a new VM power state report
is running and the state for the user VM gets updated to Running again.
was:
In some conditions, VR gets rebooted after a out of band migration was done on
vCenter.
Note, new global setting in 4.5.2 "VR reboot after out of band migration" is
set to false and this looks more like a bug.
After a VR migration to a host _and_ when the VM power state report gathering
is running, the VR (and also any user VM as well) will get into the
"PowerReportMissing" because it is not on the expected host anymore. This is
ok.
But if the VM is a VR. it will be stopped and started again. That is what we
see.
In can not be reproduced every time a migration was done, but it seems the
problem is "powerReportMissing".
I grep-ed the source and found this line related
https://github.com/apache/cloudstack/blob/master/engine/orchestration/src/com/cloud/vm/VirtualMachineManagerImpl.java#L3702
and also it seems that the graceful period might be related,
(I am not quite sure but this graceful period looks like way to optimistic to
me. 2 ping.interval are 2 x 60s and the power-state-report is running every 60s)
https://github.com/apache/cloudstack/blob/master/engine/orchestration/src/com/cloud/vm/VirtualMachinePowerStateSyncImpl.java#L110
In case it is a user VM, we see in the logs that the state will be set to
power-off, but the VM keeps running. After a while a new VM power state report
is running and the state for the user VM gets updated to Running again.
I could not yet do much debugging, logs will be appended tomorrow.
> Unexpected VR reboot after out-of-band migration
> ------------------------------------------------
>
> Key: CLOUDSTACK-8848
> URL: https://issues.apache.org/jira/browse/CLOUDSTACK-8848
> Project: CloudStack
> Issue Type: Bug
> Security Level: Public(Anyone can view this level - this is the
> default.)
> Components: VMware
> Affects Versions: 4.5.2, 4.6.0
> Reporter: René Moser
> Priority: Critical
> Fix For: 4.5.3
>
>
> In some conditions, VR gets rebooted after a out of band migration was done
> on vCenter.
> *Note, new global setting in 4.5.2 "VR reboot after out of band migration" is
> set to false and this looks more like a bug.*
> After a VR migration to a host _and_ when the VM power state report gathering
> is running, the VR (and also any user VM as well) will get into the
> "PowerReportMissing" because it is not on the expected host anymore. This is
> ok.
> But if the VM is a VR. it will be stopped and started again. That is what we
> see.
> In can not be reproduced every time a migration was done, but it seems the
> problem is "powerReportMissing".
> I grep-ed the source and found this line related
> https://github.com/apache/cloudstack/blob/master/engine/orchestration/src/com/cloud/vm/VirtualMachineManagerImpl.java#L3702
> and also it seems that the graceful period might be also related,
> https://github.com/apache/cloudstack/blob/master/engine/orchestration/src/com/cloud/vm/VirtualMachinePowerStateSyncImpl.java#L110
> In case it is a user VM, we see in the logs that the state will be set to
> power-off, but the VM keeps running. After a while a new VM power state
> report is running and the state for the user VM gets updated to Running again.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
