Repository: cloudstack Updated Branches: refs/heads/master 17164c3c8 -> a958730b3
vGPU Service offering test automation after including review comments Project: http://git-wip-us.apache.org/repos/asf/cloudstack/repo Commit: http://git-wip-us.apache.org/repos/asf/cloudstack/commit/a958730b Tree: http://git-wip-us.apache.org/repos/asf/cloudstack/tree/a958730b Diff: http://git-wip-us.apache.org/repos/asf/cloudstack/diff/a958730b Branch: refs/heads/master Commit: a958730b3957e64ccc0d704adee50b122e17afad Parents: 17164c3 Author: sailajamada <[email protected]> Authored: Thu Sep 4 14:49:41 2014 +0530 Committer: sailajamada <[email protected]> Committed: Thu Sep 4 14:52:03 2014 +0530 ---------------------------------------------------------------------- .../component/test_deploy_vgpu_vm.py | 305 ++++++++++++++----- 1 file changed, 235 insertions(+), 70 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/cloudstack/blob/a958730b/test/integration/component/test_deploy_vgpu_vm.py ---------------------------------------------------------------------- diff --git a/test/integration/component/test_deploy_vgpu_vm.py b/test/integration/component/test_deploy_vgpu_vm.py index eb0bade..081ab4b 100644 --- a/test/integration/component/test_deploy_vgpu_vm.py +++ b/test/integration/component/test_deploy_vgpu_vm.py @@ -73,7 +73,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): for ghost in hosts : if ghost.hypervisorversion >= "6.2.0": - #sshClient = SshClient(host=ghost.ipaddress, port=22, user='root',passwd=cls.testdata["host_password"]) + sshClient = SshClient(host=ghost.ipaddress, port=22, user='root',passwd=cls.testdata["host_password"]) if ghost.hypervisorversion == "6.2.0": @@ -136,6 +136,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): ) cls.template = get_windows_template(cls.apiclient, cls.zone.id ,ostype_desc="Windows 8 (64-bit)") + #cls.template = get_windows_template(cls.apiclient, cls.zone.id ,ostype_desc="Windows Server 2012 (64-bit)") if cls.template == FAILED: cls.template = Template.register( @@ -171,7 +172,6 @@ class TestvGPUWindowsVm(cloudstackTestCase): timeout = timeout - 1 - """ Create Service Offerings for Both K1 and K2 cards to be used for VM life cycle tests """ @@ -279,6 +279,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): self.apiclient, self.testdata["vgpu"]["service_offerings"][gtype] ) + self.debug("service offering details are:%s"%self.service_offering) list_service_response = ServiceOffering.list( self.apiclient, id=self.service_offering.id @@ -290,7 +291,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): self.assertEqual( list_service_response[0].displaytext, self.testdata["vgpu"]["service_offerings"][gtype]["displaytext"], - "Check server displaytext in createServiceOfferings" + "Check server display text in createServiceOfferings" ) self.assertEqual( list_service_response[0].name, @@ -315,7 +316,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): self.assertEqual( list_service_response[0].serviceofferingdetails.vgpuType, gtype, - "Failed To Create Service Offering . Check vGPU Service Offering list for K2 passthrough" + "Failed To Create Service Offering . Check vGPU Service Offering list" ) self.assertEqual( list_service_response[0].displaytext, @@ -346,6 +347,13 @@ class TestvGPUWindowsVm(cloudstackTestCase): Get vGPU type model """ vgpu_type_model = ssh_client.execute("xe vgpu-list vm-name-label=" + vminstancename + " params=type-model-name --minimal") + self.debug("vgpu type model is %s and value is %s and length is %s"%(vgpu_type_model,vgpu_type_model[0],len(vgpu_type_model[0]))) + if vgputype is None: + if len(vgpu_type_model[0])==0: + self.debug("This is non GPU instance") + return + else: + self.fail("Non vGPU VM has GPU cards @ host:%s"%vminstancename) self.assertNotEqual( len(vgpu_type_model), 0, @@ -360,14 +368,16 @@ class TestvGPUWindowsVm(cloudstackTestCase): ) """ Check whether the vgputype in the service offering is same as the one obtained by listing the vgpu type model """ + self.debug("VM vGPU type is: %s and vGPU type on XenServer is : %s"%(list_service_offering_response[0].serviceofferingdetails.vgpuType,vgpu_type_model[0])) self.assertEqual( list_service_offering_response[0].serviceofferingdetails.vgpuType, vgpu_type_model[0], - "Vm does not have the correct GPU resources, verified on the host" + "VM does not have the correct GPU resources, verified on the host" ) """ Check whether the VM is deployed with the card which was mentioned in the service offering """ + self.assertEqual( vgputype, list_service_offering_response[0].serviceofferingdetails.vgpuType, @@ -425,7 +435,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): "Running", msg="VM is not in Running state" ) - if vgput != "None": + if vgput != "nonvgpuoffering": self.assertEqual( vm.vgpu, vgput, @@ -453,6 +463,30 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + def check_for_vm(self,vgpucard,vgpuofferingid,vmid): + list_vm_response = list_virtual_machines( + self.apiclient, + id=vmid + ) + + if isinstance(list_vm_response, list): + vm = list_vm_response[0] + if vm.state == 'Running': + self.debug("VM state: %s" % vm.state) + else: + self.fail("Failed to start VM (ID: %s)" % vm.id) + if vm.serviceofferingid == vgpuofferingid: + self.debug("VM service offering id : %s" % vm.serviceofferingid) + else: + self.fail("Service Offering is not matching " % vm.id) + if vgpucard !="nonvgpuoffering": + if vm.vgpu == vgpucard: + self.debug("VM vGPU card is : %s" % vm.vgpu) + else: + self.fail("Failed to start VM (ID: %s) with %s vGPU card " % (vm.id,vgpucard)) + return(list_vm_response[0]) + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") def destroy_vm(self): """Destroy Virtual Machine """ @@ -528,6 +562,49 @@ class TestvGPUWindowsVm(cloudstackTestCase): ) return + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + def serviceoffering_upgrade(self,sourcetype,sourcemodel,desttype,destmodel): + + self.sourcevgpuoffering=self.vgpu_serviceoffering_creation(sourcetype,sourcemodel) + self.deploy_vGPU_windows_vm(self.sourcevgpuoffering.id,sourcetype) + time.sleep(self.testdata["vgpu"]["sleep"]) + + vm=self.check_for_vm(sourcetype,self.sourcevgpuoffering.id,self.virtual_machine.id) + self.check_for_vGPU_resource(vm.hostid,vm.instancename,vm.serviceofferingid,vm.vgpu) + + self.stop_vm() + + self.destvgpuoffering=self.vgpu_serviceoffering_creation(desttype,destmodel) + cmd = changeServiceForVirtualMachine.changeServiceForVirtualMachineCmd() + cmd.id = vm.id + cmd.serviceofferingid = self.destvgpuoffering.id + self.apiclient.changeServiceForVirtualMachine(cmd) + + self.debug("Starting VM - ID: %s" % vm.id) + self.start_vm() + time.sleep(self.testdata["vgpu"]["sleep"]) + + # Ensure that VM is in running state + vm=self.check_for_vm(desttype,self.destvgpuoffering.id,vm.id) + self.check_for_vGPU_resource(vm.hostid,vm.instancename,vm.serviceofferingid,vm.vgpu) + self.delete_vgpu_service_offering(self.destvgpuoffering) + self.delete_vgpu_service_offering(self.sourcevgpuoffering) + self.destroy_vm() + return + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + def deploy_vm(self,type,model): + + self.vgpuoffering=self.vgpu_serviceoffering_creation(type,model) + if self.vgpuoffering is not None: + self.deploy_vGPU_windows_vm(self.vgpuoffering.id,type) + time.sleep(self.testdata["vgpu"]["sleep"]) + vm=self.check_for_vm(type,self.vgpuoffering.id,self.virtual_machine.id) + self.check_for_vGPU_resource(vm.hostid,vm.instancename,vm.serviceofferingid,vm.vgpu) + self.destroy_vm() + self.delete_vgpu_service_offering(self.vgpuoffering) + return + def test_01_list_vgpu_host_details(self): """ list vGPU host details """ @@ -614,17 +691,15 @@ class TestvGPUWindowsVm(cloudstackTestCase): @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_01_create_deploy_windows_vm_with_k100_vgpu_service_offering(self): + def test_02_create_deploy_windows_vm_with_k100_vgpu_service_offering(self): """Test to create and deploy vm with K100 vGPU service offering""" + k100capacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K100") - self.debug("k100capacity is:%s"%(k100capacity)) - self.debug("k100gpuhosts is:%s"%(self.k100gpuhosts)) + if (self.k100gpuhosts == 0) or (k100capacity == 0): raise unittest.SkipTest("No XenServer available with K100 vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K100","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K100") - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + + self.deploy_vm("GRID K100","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -636,12 +711,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k120qgpuhosts == 0) or (k120qcapacity==0): raise unittest.SkipTest("No XenServer available with K120Q vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K120Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K120Q") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("GRID K120Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -653,12 +723,8 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k140qgpuhosts == 0) or (k140qcapacity ==0): raise unittest.SkipTest("No XenServer available with K140Q vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K140Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K140Q") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("GRID K140Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -670,12 +736,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k1passthroughgpuhosts == 0) or (k1passcapacity==0): raise unittest.SkipTest("No XenServer available with K1 passthrough installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("passthrough","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"passthrough") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("passthrough","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -683,16 +744,11 @@ class TestvGPUWindowsVm(cloudstackTestCase): """Test to create and deploy vm with K2 pasthrough vGPU service offering""" k2passcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","passthrough") - + if (self.k2passthroughgpuhosts == 0) or (k2passcapacity==0): raise unittest.SkipTest("No XenServer available with K2 passthrough installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("passthrough","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"passthrough") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("passthrough","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -704,12 +760,8 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k260qgpuhosts == 0) or (k260qcapacity == 0): raise unittest.SkipTest("No XenServer available with K260Q vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K260Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K260Q") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("GRID K260Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -721,12 +773,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k240qgpuhosts == 0) or (k240qcapacity==0): raise unittest.SkipTest("No XenServer available with K240Q vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K240Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K240Q") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("GRID K240Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -738,12 +785,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k220qgpuhosts == 0) or (k220qcapacity==0): raise unittest.SkipTest("No XenServer available with K220Q vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K220Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K220Q") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) + self.deploy_vm("GRID K220Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -755,13 +797,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): if (self.k200gpuhosts == 0) or (k200capacity==0): raise unittest.SkipTest("No XenServer available with K200 vGPU Drivers installed") - self.vgpuoffering=self.vgpu_serviceoffering_creation("GRID K200","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") - if self.vgpuoffering is not None: - self.deploy_vGPU_windows_vm(self.vgpuoffering.id,"GRID K200") - self.delete_vgpu_service_offering(self.vgpuoffering) - self.destroy_vm() - self.delete_vgpu_service_offering(self.vgpuoffering) - + self.deploy_vm("GRID K200","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") @@ -870,7 +906,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_13_validate_deployed_vGPU_windows_vm(self): + def test_12_validate_deployed_vGPU_windows_vm(self): """ Test deploy virtual machine """ self.debug("Check if deployed VMs are in running state?") @@ -890,7 +926,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_14_stop_vGPU_windows_vm(self): + def test_13_stop_vGPU_windows_vm(self): """ Test stop virtual machine """ if(self.vm_k1_card): @@ -902,7 +938,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_15_start_vGPU_windows_vm(self): + def test_14_start_vGPU_windows_vm(self): """ Test start virtual machine """ if(self.vm_k1_card): @@ -918,7 +954,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_16_restore_vGPU_windows_vm(self): + def test_15_restore_vGPU_windows_vm(self): """Test restore Virtual Machine """ if(self.vm_k1_card): @@ -942,7 +978,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_17_reboot_vGPU_windows_vm(self): + def test_16_reboot_vGPU_windows_vm(self): """ Test reboot virtual machine """ if(self.vm_k1_card): @@ -966,7 +1002,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_18_destroy_vGPU_windows_vm(self): + def test_17_destroy_vGPU_windows_vm(self): """Test destroy Virtual Machine """ if(self.vm_k1_card): @@ -984,7 +1020,7 @@ class TestvGPUWindowsVm(cloudstackTestCase): return @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") - def test_19_recover_vGPU_windows_vm(self): + def test_18_recover_vGPU_windows_vm(self): """Test recover Virtual Machine """ if(self.vm_k1_card): @@ -1002,6 +1038,135 @@ class TestvGPUWindowsVm(cloudstackTestCase): return + def test_19_destroy_vGPU_windows_vm_after_recover(self): + """Test destroy Virtual Machine + """ + if(self.vm_k1_card): + self.vm_k1_card.delete(self.apiclient) + self.vm_k1_card.getState( + self.apiclient, + "Destroyed") + + if(self.vm_k2_card): + self.vm_k2_card.delete(self.apiclient) + self.vm_k2_card.getState( + self.apiclient, + "Destroyed") + + return + + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_20_nonvgpuvm_k2vgpuvm_offline(self): + """Test to change service from non vgpu to vgpu K200""" + k200capacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K200") + if (self.k200gpuhosts == 0) or (k200capacity==0): + raise unittest.SkipTest("No XenServer available with K200 vGPU Drivers installed. Skipping non gpu to K200 Offering Upgrade") + + self.serviceoffering_upgrade("nonvgpuoffering","None","GRID K200","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") + + return + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_21_K2_vgpuvm_vgpuvm_offline(self): + """Test to change service from vgpu K200 to vgpu K240Q""" + k240qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K240Q") + k200capacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K200") + if (self.k240qgpuhosts == 0) or (self.k200gpuhosts == 0) or (k240qcapacity==0) or (k200capacity==0): + raise unittest.SkipTest("No XenServer available with K240Q,K200Q vGPU Drivers installed. Skipping K200 to K240Q Offering Upgrade") + + self.serviceoffering_upgrade("GRID K200","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K240Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") + + return + + + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_22_K1_vgpuvm_vgpuvm_offline(self): + """Test to change service from K1 vgpu K120Q to K1 vgpu K140Q""" + + k140qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K140Q") + k120qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K120Q") + + if (self.k140qgpuhosts == 0) or (self.k120qgpuhosts == 0) or (k140qcapacity==0) or (k120qcapacity==0): + raise unittest.SkipTest("No XenServer available with K140Q,K120Q vGPU Drivers installed. Skipping K200 to K240Q Offering Upgrade") + + self.serviceoffering_upgrade("GRID K120Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K140Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") + + return + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_23_nonvgpuvm_k1vgpuvm_offline(self): + """Test to change service from non vgpu to vgpu K100""" + + k100capacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K100") + if (self.k100gpuhosts == 0) or (k100capacity == 0): + raise unittest.SkipTest("No XenServer available with K100 vGPU Drivers installed. Skipping non gpu to K100 Offering Upgrade") + self.serviceoffering_upgrade("nonvgpuoffering","None","GRID K100","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") + + return + + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_24_K2_vgpuvm_nonvgpuvm_offline(self): + """Test to change service from non vgpu to vgpu K240Q""" + + k240qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K240Q") + if (self.k240qgpuhosts == 0) or (k240qcapacity==0): + raise unittest.SkipTest("No XenServer available with K240Q vGPU Drivers installed. Skipping K2 vgpu 240Q to nonvgpu Offering Upgrade") + + self.serviceoffering_upgrade("GRID K240Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","nonvgpuoffering","None") + + return + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_25_K1_vgpuvm_nonvgpuvm_offline(self): + """Test to change service from non vgpu to vgpu K140Q""" + + k140qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K140Q") + + if (self.k140qgpuhosts == 0) or (k140qcapacity==0): + raise unittest.SkipTest("No XenServer available with K140Q vGPU Drivers installed. Skipping K1 vgpu 140Q to nonvgpu Offering Upgrade") + + self.serviceoffering_upgrade("GRID K140Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","nonvgpuoffering","None") + + return + + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_26_K140Q_vgpuvm_K240Q_vgpuvm_offline(self): + """Test to change service from K1 vgpu K140Q to K2 vgpu K240Q""" + + k140qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K140Q") + k240qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K240Q") + + if (self.k140qgpuhosts == 0) or (self.k240qgpuhosts == 0) or (k140qcapacity == 0) or (k240qcapacity==0): + raise unittest.SkipTest("No XenServer available with K140Q,K240Q vGPU Drivers installed. Skipping K140Q to K240Q Offering Upgrade") + + self.serviceoffering_upgrade("GRID K140Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K240Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs") + return + + @attr(tags = ['advanced', 'basic' , 'vgpu'], required_hardware="true") + + def test_27_K240Q_vgpuvm_K140Q_vgpuvm_offline(self): + """Test to change service from K2 vgpu K240Q to K1 vgpu K140Q""" + + k140qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK107GL [GRID K1] GPUs","GRID K140Q") + k240qcapacity=self.check_host_vgpu_capacity("Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K240Q") + + if (self.k140qgpuhosts == 0) or (self.k240qgpuhosts == 0) or (k140qcapacity == 0) or (k240qcapacity==0): + raise unittest.SkipTest("No XenServer available with K140Q,K240Q vGPU Drivers installed. Skipping K140Q to K240Q Offering Upgrade") + + self.serviceoffering_upgrade("GRID K240Q","Group of NVIDIA Corporation GK104GL [GRID K2] GPUs","GRID K140Q","Group of NVIDIA Corporation GK107GL [GRID K1] GPUs") + return + @classmethod def tearDownClass(self):
