As you may know, I'm working on a blog series that demonstrates Atlantis USX on XenServer. For various reasons, the actual test lab is actually built within a VMware vCloud Director environment.
For the inquisitive, it looks like this:
For those unfamiliar with vCloud Director, within a vApp configuratio (as shown above) you have a virtual application comprised of virtual machines, their networks and their policies (with thanks to Bluelock). Which, I'll grant you, is very similar to a XenServer vApp which is a logical group of one or more related virtual machines (VMs) which can be managed as a single entity as defined by Citrix eDocs. Of course, the VMs within a XenServer vApp do not have to reside on one host server - but then vCloud Director abstracts all physical hosts too..but I digress.
Anyhoo, in the beginning testing went very well: initial deployment and USX Simple Volumes all fine and dandy. However pooled volumes would be deployed and then some of the XenServer hosts would simply abend. Not even with the decency of a crash dump. The cads.
Investigation brought this Citrix XenServer forum thread to light on Random Reboots after Upgrade to 6.5 - which turns out not to be a 6.5 upgrade issue, but a general issue with 6.5 if you've over committed vCPU resource to the underlying infrastructure. Granted that isn't Best Practice - but it should also not Not Work ungracefully.
So, it appears that as of the time of writing - XenServer 6.5 doesn't respond well to a lack of vCPU resource. Reading the feedback (check out very useful comment on page 3 posted on the 2nd April) your takeaway should be :-
- The issue has been reproduced by Citrix.
- A workaround is to ensure that you've not over-committed resources and set all vCPU priorities to "Normal".
- A hotfix is on its way.
Changing the Atlantis USX preferences to ensure that VM priorities are at normal resolved my particular issue.
In summary, if your XenServer 6.5 hosts are randomly crashing (which may be more noticeable after an upgrade) be aware as of the time of writing XenServer 6.5 doesn't respond well to a lack of vCPU resource - workarounds are possible while Citrix resolve the issue.