Host's pending resources tracking was fixed. Previously the required memory was not freed on the destination host during live migration when the migration was cancelled, leading eventually to an apparent memory exhaustion on the destination host - preventing VMs to be launched there.
Resources are now tracked correctly
Description of problem:
After migration canceling, host pending resources (pending_vmem_size and pending_vcpus_count) are not cleared. eventually causing to insufficiant memory when trying to create new VM or migrating VM.
Version-Release number of selected component (if applicable):
engine (3.5):
rhevm-3.5.0-0.21.el6ev.noarch
hosts (7.1):
qemu-kvm-rhev-1.5.3-60.el7_0.11.x86_64
vdsm-4.16.7.6-1.el7ev.x86_64
libvirt-1.2.8-9.el7.x86_64
sanlock-3.2.2-1.el7.x86_64
How reproducible:
Consistently
Steps to Reproduce:
1. Via RHEV UI, virtual machine tab, create new VM and install OS on it.
2. Right click on created VM and select "Migrate".
3. Right after migration started, right click on VM and selet "cancel migration"
Actual results:
Afetr migration canceled, pending mem and pending cpu should be released.
Expected results:
Pending cpu and mem are not released after migration canceling.
Additional info:
Host and engine logs attached (migration started at 2014-12-02 16:28:18)