[cluster launcher] [vSphere] Do not fetch runtime-info of vm from cac…

…hed_nodes (ray-project#40655) Power-on-off status is runtime info of VM, should not fetch it from cached-nodes, which is probably dirty data. It should query by pyvmomi_sdk every time. Signed-off-by: Chen Hui <[email protected]>
miqdigital · Oct 25, 2023 · 7f98a20 · 7f98a20
1 parent 45c7410
commit 7f98a20
Showing 1 changed file with 8 additions and 4 deletions.
diff --git a/python/ray/autoscaler/_private/vsphere/node_provider.py b/python/ray/autoscaler/_private/vsphere/node_provider.py
@@ -176,12 +176,16 @@ def non_terminated_nodes(self, tag_filters):
             return nodes
 
     def is_running(self, node_id):
-        node = self._get_cached_node(node_id)
-        return node.power_state in {HardPower.State.POWERED_ON}
+        vm = self.pyvmomi_sdk_provider.get_pyvmomi_obj(
+            [vim.VirtualMachine], obj_id=node_id
+        )
+        return vm.runtime.powerState != vim.VirtualMachinePowerState.poweredOn
 
     def is_terminated(self, node_id):
-        node = self._get_cached_node(node_id)
-        return node.power_state not in {HardPower.State.POWERED_ON}
+        vm = self.pyvmomi_sdk_provider.get_pyvmomi_obj(
+            [vim.VirtualMachine], obj_id=node_id
+        )
+        return vm.runtime.powerState != vim.VirtualMachinePowerState.poweredOn
 
     def node_tags(self, node_id):
         with self.tag_cache_lock: