解决多张gpu卡情况下测试失败问题

oec-hardware-1.1.5-6-fix-bug-gpu.patch

@@ -0,0 +1,36 @@
+diff -Naur rpm/tests/compatible/gpu/nvidia_gpu.py oech/tests/compatible/gpu/nvidia_gpu.py
+--- rpm/tests/compatible/gpu/nvidia_gpu.py	2024-08-29 19:48:19.472522063 +0800
++++ oech/tests/compatible/gpu/nvidia_gpu.py	2024-08-30 16:21:51.821261220 +0800
+@@ -73,6 +73,11 @@
+ 
+         self.logger.info("Set default test gpu as %s." % id_num)
+ 
++    def clean_default_gpu(self):
++        if 'CUDA_VISIBLE_DEVICES' in os.environ:
++            del os.environ['CUDA_VISIBLE_DEVICES']
++            self.logger.info("Clean default test gpu.")
++
+     def test_pressure(self):
+         """
+         Set pressure for gpu to test
+@@ -87,10 +92,7 @@
+ 
+         os.chdir("/opt/gpu-burn")
+         cmd = self.command.run_cmd(
+-            "nvidia-smi -q | grep -i -A1 '%s' | grep 'Product Name' | cut -d ':' -f 2" % pci_num)
+-        device_name = cmd[0].strip()
+-        cmd = self.command.run_cmd(
+-            "./gpu_burn -l | grep -i '%s' | cut -d ':' -f 1 | awk '{print $2}'" % device_name)
++            "nvidia-smi -q | grep -i -A20 '%s' | grep 'Minor Number' | cut -d ':' -f 2" % pci_num)
+         run_id = cmd[0].strip()
+         cmd = getstatusoutput(
+             'nohup ./gpu_burn -i%s 10 &> %s &' % (run_id, self.gpu_burn))
+@@ -185,6 +187,8 @@
+                     result = False
+                     self.logger.error("Test Vulkan failed.")
+ 
++            self.clean_default_gpu()
++
+         except Exception as e:
+             self.logger.error(
+                 "Failed to run the script because compiling or setting variables: %s" % e)

oec-hardware.spec

@@ -6,7 +6,7 @@
 Name:           oec-hardware
 Summary:        openEuler Hardware Compatibility Test Suite
 Version:        1.1.5
-Release:        5
+Release:        6
 Group:          Development/Tools
 License:        Mulan PSL v2
 URL:            https://gitee.com/openeuler/oec-hardware
@@ -18,6 +18,7 @@ Patch0002:      oec-hardware-1.1.5-2-fix-bug.patch
 Patch0003:      oec-hardware-1.1.5-3-functional-optimization-fix-bug.patch
 Patch0004:      oec-hardware-1.1.5-4-fix-gpu-testcase-bug.patch
 Patch0005:      oec-hardware-1.1.5-5-fix-cpufreq-userspace.patch
+Patch0006:      oec-hardware-1.1.5-6-fix-bug-gpu.patch
 
 Buildroot:      %{_tmppath}/%{name}-%{version}-root
 BuildRequires:  gcc
@@ -46,6 +47,7 @@ openEuler Hardware Compatibility Test Server
 %patch3 -p1
 %patch4 -p1
 %patch5 -p1
+%patch6 -p1
 
 %build
 %ifarch x86_64 aarch64
@@ -93,6 +95,10 @@ sed -i 's#grep openeulerversion /etc/openEuler-latest#grep %{vendor_lowercase}ve
 rm -rf /var/lock/oech.lock
 
 %changelog
+* Fri Aug 30 2024 gaochuanji <gaochuanji@inspur.com> - 1.1.5-6
+- Resolve the issue of the gpu_burn test failed when multiple gpus
+- Resolve the issue of first gpu test success and others failed when multiple gpus
+
 * Tue Aug 13 2024 gaochuanji <gaochuanji@inspur.com> - 1.1.5-5
 - Resolve the issue of failed userspace testing for cpufreq on certain models