After more investigation, I found out that  changing template's "OS Type" 
assignment affects how a VM is started as PV or HVM guest on either XenServer 
6.5 or 7.0 clusters:

On XenServer 6.5 clusters,  64 bit RHEL guests with versions up to 6.7 (last 
6.x version listed by ACS 4.9.3.0) are always started as PV guest, and they all 
run properly.
On XenServer 7.0 clusters,  64 bit RHEL guests with versions up to 6.4 are 
started as PV guests and run properly, guests with version 6.5 onwards are 
started as HVM guest, and these won't start.

So, the simple work around is to assign "RHEL 6.4 (64bit)" to all my rhel 6.x 
templates so that guests using these templates are always started as PV guest 
regardless the version of XenServer cluster they are running on.  

IMHO, I think this is a bug in CloudStack, that is all rhel 6.x guests should 
be started as PV guests regardless Xen hypervisor versions (I only verified on 
XenServer 6.5 and 7.0).

Yiping

On 4/11/18, 2:31 PM, "Yiping Zhang" <yzh...@marketo.com> wrote:

    Hi, list:
    
    
    I am in the process of upgrading my hypervisors clusters from XenServer 6.5 
to 7.0 for all my ACS instances.
    
    My XS 7.0 clusters are patched up to XS70E050.
    
    During cluster rolling upgrade, VM instances are live migrated several 
times, eventually all of them running on XS 7.0 hosts.
    However, out of a hundred or so instances, a few of rhle 6.x VM got stuck 
at “Plex86/Bochs VGABios …” screen with at least one vCPU at 100%.
    Comparing VM parameters between running and stuck instances, I can see that 
running instances are PV guests while stuck instances are HVM guests.
    RHEL 7.x guest are always in HVM mode and always runs on both XS 6.5 and 
7.0 hosts.
    
    Running RHEL6.x guests:
    
                   HVM-boot-policy ( RW):
    
                   HVM-boot-params (MRW):
    
             HVM-shadow-multiplier ( RW): 1.000
    
                           PV-args ( RW): graphical utf8
    
                     PV-bootloader ( RW): pygrub
    
    Stuck RHEL 6.x guests:
    
                   HVM-boot-policy ( RW): BIOS order
    
                   HVM-boot-params (MRW): order: dc
    
             HVM-shadow-multiplier ( RW): 1.000
    
                           PV-args ( RW):
    
                     PV-bootloader ( RW):
    
    Why does CloudStack convert just a few rhel 6.x instances into HVM mode, 
while leaving most in PV mode?
    How would I force them back to PV guests?
    
    Thanks for all the help
    
    Yiping
    

Reply via email to