make populator incremental handling

make populator handling incremental

fix - remove unused func

refine

refine

refine

fixes llm-d-incubation#450

fix

remove duplicated call to setStatus

add work queue

add lock

Author: osswangxining

pkg/controller/common/interface.go

@@ -32,6 +32,9 @@ const (
 	// corresponding LauncherConfig object's PodTemplate that the server-providing Pod uses.
 	LauncherConfigHashAnnotationKey = "dual-pods.llm-d.ai/launcher-config-hash"
 
+	// LauncherTemplateHashAnnotationKey is the node-independent template hash on a launcher Pod, used for spec-drift detection.
+	LauncherTemplateHashAnnotationKey = "dual-pods.llm-d.ai/launcher-populator-template-hash"
+
 	// LauncherServicePort is the port number on which the launcher exposes its HTTP service
 	// for the management of vLLM instances.
 	// This is a contract between the controllers and the launcher implementation.

pkg/controller/dual-pods/inference-server.go

@@ -589,7 +589,11 @@ func (item infSvrItem) process(urCtx context.Context, ctl *controller, nodeDat *
 		)
 	}
 
-	desiredLauncherPod, err := utils.BuildLauncherPodFromTemplate(lc.Spec.PodTemplate, ctl.namespace, requestingPod.Spec.NodeName, lcName)
+	desiredLauncherTemplateHash, err := utils.ComputeLauncherTemplateHash(lc.Spec.PodTemplate)
+	if err != nil {
+		return fmt.Errorf("failed to compute template hash for LauncherConfig %q: %w", lcName, err), true
+	}
+	desiredLauncherPod, err := utils.BuildLauncherPodFromTemplate(lc.Spec.PodTemplate, ctl.namespace, requestingPod.Spec.NodeName, lcName, desiredLauncherTemplateHash)
 	if err != nil {
 		return fmt.Errorf("failed to build launcher Pod from LauncherConfig %q: %w", lcName, err), true
 	}