Merge pull request #432 from xcp-ng/dnt/win-145

Add more fixes and hacks for Windows tests with 9.1.145

Author: dinhngtu

lib/windows/__init__.py

@@ -147,17 +147,24 @@ def vif_set_dns(vif: VIF, nameservers: List[str]):
     )
 
 
+def is_vm_offboarded_xenvif(vm: VM):
+    try:
+        val = vm.execute_powershell_script(
+            r'$null -eq (Get-ScheduledTask "Copy-XenVifSettings" -ErrorAction SilentlyContinue)', simple_output=True
+        )
+        # Sometimes, we may get an error like "invalid truth value 'mux_client_request_session: read from master failed:
+        # broken pipe'". If so, just retry.
+        if "broken pipe" in val.lower():
+            return False
+        return strtobool(val)
+    except SSHCommandFailed:
+        # This check itself can fail when Copy-XenVifSettings runs, so we have to tolerate its failure.
+        return False
+
+
 def wait_for_vm_xenvif_offboard(vm: VM):
     # Xenvif offboard will reset the NIC, so need to wait for it to disappear first
-    wait_for(
-        lambda: strtobool(
-            vm.execute_powershell_script(
-                r'$null -eq (Get-ScheduledTask "Copy-XenVifSettings" -ErrorAction SilentlyContinue)', simple_output=True
-            )
-        ),
-        timeout_secs=300,
-        retry_delay_secs=30,
-    )
+    wait_for(lambda: is_vm_offboarded_xenvif(vm), timeout_secs=180, retry_delay_secs=15)
 
 
 def set_vm_dns(vm: VM):

lib/windows/other_tools.py

@@ -1,4 +1,5 @@
 import logging
+import time
 from pathlib import PureWindowsPath
 
 from lib.common import strtobool, wait_for
@@ -50,3 +51,9 @@ def install_other_drivers(vm: VM, other_tools_iso_name: str, param: Dict[str, An
         vm.eject_cd()
         vm.start()
         wait_for_vm_running_and_ssh_up_without_tools(vm)
+        # HACK: For some reason, the XenServer 9.4.2 package causes a network blip after bootup. This could exhibit as
+        # random disconnections, SSH command failure, etc. The reason is not yet clear, but retry confirmation of the
+        # network link as a workaround.
+        # TODO: XCPNG-3038
+        time.sleep(30)
+        wait_for_vm_running_and_ssh_up_without_tools(vm)

tests/guest_tools/win/test_guest_tools_win.py

@@ -131,8 +131,11 @@ def test_xenvbd_ssd(self, vm_install_test_tools_per_test_class: VM):
         """Xenvbd must always advertise as SSD to avoid unnecessary defragging by Windows."""
         vm = vm_install_test_tools_per_test_class
         is_ssd = strtobool(
+            # We have to filter Get-PhysicalDisk instead of piping directly from Get-Disk since direct Get-PhysicalDisk
+            # by ID doesn't work on Server 2016.
             vm.execute_powershell_script(
-                r'''(Get-PhysicalDisk -DeviceNumber (Get-Partition -DriveLetter C).DiskNumber).MediaType -eq "SSD"'''
+                r'''$disk = Get-Partition -DriveLetter ($Env:SystemDrive[0]) | Get-Disk;
+(Get-PhysicalDisk | Where-Object UniqueId -eq $disk.UniqueId).MediaType -eq "SSD"'''
             )
         )
         assert is_ssd

tests/guest_tools/win/test_xenclean.py

@@ -24,7 +24,10 @@ def run_xenclean(vm: VM, guest_tools_iso: Dict[str, Any]):
 
     logging.info("Run XenClean")
     xenclean_path = PureWindowsPath("D:\\") / guest_tools_iso["xenclean_path"]
-    xenclean_cmd = f"Set-Location C:\\; {xenclean_path} -NoReboot -Confirm:$false; {WINDOWS_SHUTDOWN_COMMAND}"
+    if guest_tools_iso["xenclean_path"].lower().endswith(".ps1"):
+        xenclean_cmd = f"Set-Location C:\\; {xenclean_path} -NoReboot -Confirm:$false; {WINDOWS_SHUTDOWN_COMMAND}"
+    else:
+        xenclean_cmd = f"Set-Location C:\\; {xenclean_path} -noReboot -noConfirm; {WINDOWS_SHUTDOWN_COMMAND}"
     vm.start_background_powershell(xenclean_cmd)
 
     # XenClean sometimes takes a bit long due to all the calls to the uninstallers. We need an extended timeout.