Created attachment 995229 [details] engine log Description of problem: Via UI create vm with one cpu and two numa nodes, because we have more cpu's that numa nodes, UI will fail to add cpu's to numa nodes, and when I start vm it fail with message: "VM test_one_numa is down with error. Exit message: invalid argument: Failed to parse bitmap ''" Version-Release number of selected component (if applicable): rhevm-3.5.0-0.31.el6ev.noarch How reproducible: Always Steps to Reproduce: 1. Create vm with one cpu and two numa nodes 2. Start vm 3. Actual results: Start of vm failed with error message Expected results: Start of vm success Additional info: Under org.ovirt.engine.core.vdsbroker.vdsbroker.CreateVDSCommand I see that: guestNumaNodes=[{cpus=, nodeIndex=0, memory=512}, {cpus=, nodeIndex=1, memory=512}] so numa nodes not received any cpus
we'll supply a CandoAction and a validation whenever setting the # of nodes we want to create # of nodes <= # of cpus its best if UI validation will be 2 ways - on number of cpus and on number of nodes. that way we'll make sure the numa tunning dialog is initiated with right values. Doron - need Z flags probably
Problem verified and recreated. Diagnostics. ----------- The command sent to vdsm is valid and correct. Using REST api, the stored vm_numa_nodes state is valid: <vm_numa_nodes> <vm_numa_node href= "/api/vms/5684a623-1f2f-46db-8570-945469314ac1/numanodes/cccbe0e8-b495-4c0c-85fa-75049243df2c" id="cccbe0e8-b495-4c0c-85fa-75049243df2c"> <index>0</index> <memory>256</memory> <vm href= "/api/vms/5684a623-1f2f-46db-8570-945469314ac1" id="5684a623-1f2f-46db-8570-945469314ac1"/> <numa_node_pins> <numa_node_pin pinned="true" index="1"> <host_numa_node id="6216e294-eaff-4ac4-a13f-612696729b98"/> </numa_node_pin> </numa_node_pins> </vm_numa_node> <vm_numa_node href= "/api/vms/5684a623-1f2f-46db-8570-945469314ac1/numanodes/4eb89f0f-47cb-46aa-9901-30f71fc13a66" id="4eb89f0f-47cb-46aa-9901-30f71fc13a66"> <index>1</index> <memory>256</memory> <vm href= "/api/vms/5684a623-1f2f-46db-8570-945469314ac1" id="5684a623-1f2f-46db-8570-945469314ac1"/> <numa_node_pins> <numa_node_pin pinned="true" index="1"> <host_numa_node id="6216e294-eaff-4ac4-a13f-612696729b98"/> </numa_node_pin> </numa_node_pins> </vm_numa_node> </vm_numa_nodes> Looking at vdsm logs. The problem is found as invalid section in domain.xml <cpu match="exact"> <model>Nehalem</model> <topology cores="1" sockets="16" threads="1"/> <numa> <cell cpus="" memory="262144"/> <cell cpus="" memory="262144"/> </numa> </cpu> Notice cpus="" which should be cpus="0" and cputs="1" conclusion: ----------- We need to fix this bug at vdsm level.
Following discussion with dfediuk. We will prevent such configuration in canDoAction() method, and GUI.
Verified on ovirt-engine-3.6.0-0.0.master.20150519172219.git9a2e2b3.el6.noarch Receive message: Cannot edit VM. Assigned 2 NUMA nodes for 1 CPU cores. Cannot assign more NUMA nodes than CPU cores.