From e49111edc7f1e6eb7869ac64ae6122076f24784f Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Tue, 30 Sep 2025 20:42:20 -0400
Subject: [PATCH 01/10] Make the check-container-tool target more robust

Signed-off-by: Hannah DeFazio <h2defazio@gmail.com>
---
 Makefile | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/Makefile b/Makefile
index 6c742ff5..d005172e 100644
--- a/Makefile
+++ b/Makefile
@@ -293,9 +293,16 @@ check-envsubst:
 
 .PHONY: check-container-tool
 check-container-tool:
-	@command -v $(CONTAINER_TOOL) >/dev/null 2>&1 || { \
-	  echo "❌ $(CONTAINER_TOOL) is not installed."; \
-	  echo "🔧 Try: sudo apt install $(CONTAINER_TOOL) OR brew install $(CONTAINER_TOOL)"; exit 1; }
+	@if [ -z "$(CONTAINER_TOOL)" ]; then \
+		echo "❌ Error: No container tool detected. Please install docker or podman."; \
+		exit 1; \
+	elif ! command -v $(CONTAINER_TOOL) >/dev/null 2>&1; then \
+		echo "❌ Error: '$(CONTAINER_TOOL)' is not installed or not in your PATH."; \
+		echo "🔧 Try: sudo apt install $(CONTAINER_TOOL) OR brew install $(CONTAINER_TOOL)"; \
+		exit 1; \
+	else \
+		echo "✅ Container tool '$(CONTAINER_TOOL)' found."; \
+	fi
 
 .PHONY: check-kubectl
 check-kubectl:

From a7bcad2bb18af581f3188532926e91f4a3ff0022 Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Tue, 30 Sep 2025 20:52:33 -0400
Subject: [PATCH 02/10] Use the CONTAINER_TOOL env var for running the e2e test
 script

Signed-off-by: Hannah DeFazio <h2defazio@gmail.com>
---
 Makefile                |  1 +
 test/scripts/run_e2e.sh | 16 ++++++++++------
 2 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/Makefile b/Makefile
index d005172e..41fd228f 100644
--- a/Makefile
+++ b/Makefile
@@ -37,6 +37,7 @@ TYPOS_ARCH = $(TYPOS_TARGET_ARCH)-unknown-linux-musl
 endif
 
 CONTAINER_TOOL := $(shell { command -v docker >/dev/null 2>&1 && echo docker; } || { command -v podman >/dev/null 2>&1 && echo podman; } || echo "")
+export CONTAINER_TOOL
 BUILDER := $(shell command -v buildah >/dev/null 2>&1 && echo buildah || echo $(CONTAINER_TOOL))
 PLATFORMS ?= linux/amd64 # linux/arm64 # linux/s390x,linux/ppc64le
 
diff --git a/test/scripts/run_e2e.sh b/test/scripts/run_e2e.sh
index add66dc3..fe13508a 100755
--- a/test/scripts/run_e2e.sh
+++ b/test/scripts/run_e2e.sh
@@ -1,5 +1,9 @@
 #!/bin/bash
 
+# Use the CONTAINER_TOOL from the environment, or default to docker if it's not set.
+CONTAINER_TOOL="${CONTAINER_TOOL:-docker}"
+echo "Using container tool: ${CONTAINER_TOOL}"
+
 # Set a default EPP_TAG if not provided
 export EPP_TAG="${EPP_TAG:-dev}"
 
@@ -9,27 +13,27 @@ export VLLM_SIMULATOR_TAG="${VLLM_SIMULATOR_TAG:-v0.4.0}"
 # Set the default routing side car image tag
 export ROUTING_SIDECAR_TAG="${ROUTING_SIDECAR_TAG:-v0.2.0}"
 
-SIMTAG=$(docker images | grep ghcr.io/llm-d/llm-d-inference-sim | awk '{print $2}' | grep ${VLLM_SIMULATOR_TAG})
+SIMTAG=$(${CONTAINER_TOOL} images | grep ghcr.io/llm-d/llm-d-inference-sim | awk '{print $2}' | grep ${VLLM_SIMULATOR_TAG})
 if [[ "${SIMTAG}" != "${VLLM_SIMULATOR_TAG}" ]]; then
-  docker pull ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}
+  ${CONTAINER_TOOL} pull ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}
   if [[ $? != 0 ]]; then
     echo "Failed to pull ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}"
     exit 1
   fi
 fi
 
-EPPTAG=$(docker images | grep ghcr.io/llm-d/llm-d-inference-scheduler | awk '{print $2}' | grep ${EPP_TAG})
+EPPTAG=$(${CONTAINER_TOOL} images | grep ghcr.io/llm-d/llm-d-inference-scheduler | awk '{print $2}' | grep ${EPP_TAG})
 if [[ "${EPPTAG}" != "${EPP_TAG}" ]]; then
-  docker pull ghcr.io/llm-d/llm-d-inference-scheduler:${EPP_TAG}
+  ${CONTAINER_TOOL} pull ghcr.io/llm-d/llm-d-inference-scheduler:${EPP_TAG}
   if [[ $? != 0 ]]; then
     echo "Failed to pull ghcr.io/llm-d/llm-d-inference-scheduler:${EPP_TAG}"
     exit 1
   fi
 fi
 
-SIDECARTAG=$(docker images | grep ghcr.io/llm-d/llm-d-routing-sidecar | awk '{print $2}' | grep ${ROUTING_SIDECAR_TAG})
+SIDECARTAG=$(${CONTAINER_TOOL} images | grep ghcr.io/llm-d/llm-d-routing-sidecar | awk '{print $2}' | grep ${ROUTING_SIDECAR_TAG})
 if [[ "${SIDECARTAG}" != "${ROUTING_SIDECAR_TAG}" ]]; then
-  docker pull ghcr.io/llm-d/llm-d-routing-sidecar:${ROUTING_SIDECAR_TAG}
+  ${CONTAINER_TOOL} pull ghcr.io/llm-d/llm-d-routing-sidecar:${ROUTING_SIDECAR_TAG}
   if [[ $? != 0 ]]; then
     echo "Failed to pull ghcr.io/llm-d/llm-d-routing-sidecar:${ROUTING_SIDECAR_TAG}"
     exit 1

From 8f620c5a332b6c8c89905708580cb3d270f24c17 Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Mon, 6 Oct 2025 14:41:54 -0400
Subject: [PATCH 03/10] Set the full docker image paths via env vars

---
 Makefile                                      | 47 +++++++++++--------
 .../inference-gateway/deployments.yaml        |  2 +-
 .../inference-gateway/kustomization.yaml      |  4 --
 .../components/vllm-sim-pd/deployments.yaml   |  6 +--
 .../components/vllm-sim-pd/kustomization.yaml |  6 ---
 deploy/components/vllm-sim/deployments.yaml   |  2 +-
 deploy/components/vllm-sim/kustomization.yaml |  4 --
 .../patch-deployments.yaml                    |  2 +-
 .../common/patch-statefulset.yaml             |  2 +-
 .../openshift-base/kustomization.yaml         |  6 ---
 scripts/kind-dev-env.sh                       | 27 +++++------
 scripts/kubernetes-dev-env.sh                 |  8 +---
 scripts/pull_images.sh                        | 36 ++++++++++++++
 test/e2e/e2e_suite_test.go                    | 15 +++---
 test/e2e/e2e_test.go                          | 20 ++++----
 test/e2e/yaml/deployments.yaml                |  2 +-
 test/e2e/yaml/vllm-sim-pd.yaml                |  6 +--
 test/e2e/yaml/vllm-sim.yaml                   |  2 +-
 test/scripts/run_e2e.sh                       | 40 ----------------
 19 files changed, 104 insertions(+), 133 deletions(-)
 create mode 100644 scripts/pull_images.sh

diff --git a/Makefile b/Makefile
index 41fd228f..e6c7832a 100644
--- a/Makefile
+++ b/Makefile
@@ -8,9 +8,10 @@ TARGETOS ?= $(shell go env GOOS)
 TARGETARCH ?= $(shell go env GOARCH)
 PROJECT_NAME ?= llm-d-inference-scheduler
 IMAGE_REGISTRY ?= ghcr.io/llm-d
-IMAGE_TAG_BASE ?= $(IMAGE_REGISTRY)/$(PROJECT_NAME)
+EPP_IMG_TAG_BASE ?= $(IMAGE_REGISTRY)/$(PROJECT_NAME)
 EPP_TAG ?= dev
-IMG = $(IMAGE_TAG_BASE):$(EPP_TAG)
+EPP_IMAGE = $(EPP_IMG_TAG_BASE):$(EPP_TAG)
+export EPP_IMAGE
 NAMESPACE ?= hc4ai-operator
 
 # Map go arch to typos arch
@@ -95,7 +96,7 @@ test-integration: download-tokenizer install-dependencies ## Run integration tes
 	go test -ldflags="$(LDFLAGS)" -v -tags=integration_tests ./test/integration/
 
 .PHONY: test-e2e
-test-e2e: image-build ## Run end-to-end tests against a new kind cluster
+test-e2e: image-build image-pull ## Run end-to-end tests against a new kind cluster
 	@printf "\033[33;1m==== Running End to End Tests ====\033[0m\n"
 	./test/scripts/run_e2e.sh
 
@@ -120,20 +121,26 @@ build: check-go install-dependencies download-tokenizer ## Build the project
 ##@ Container Build/Push
 
 .PHONY:	image-build
-image-build: check-container-tool ## Build Docker image ## Build Docker image using $(CONTAINER_TOOL)
-	@printf "\033[33;1m==== Building Docker image $(IMG) ====\033[0m\n"
+image-build: check-container-tool ## Build Docker image using $(CONTAINER_TOOL)
+	@printf "\033[33;1m==== Building Docker image $(EPP_IMAGE) ====\033[0m\n"
 	$(CONTAINER_TOOL) build \
 		--platform linux/$(TARGETARCH) \
  		--build-arg TARGETOS=linux \
 		--build-arg TARGETARCH=$(TARGETARCH) \
 		--build-arg COMMIT_SHA=${GIT_COMMIT_SHA} \
 		--build-arg BUILD_REF=${BUILD_REF} \
- 		-t $(IMG) .
+ 		-t $(EPP_IMAGE) .
 
 .PHONY: image-push
-image-push: check-container-tool ## Push Docker image $(IMG) to registry
-	@printf "\033[33;1m==== Pushing Docker image $(IMG) ====\033[0m\n"
-	$(CONTAINER_TOOL) push $(IMG)
+image-push: check-container-tool ## Push Docker image $(EPP_IMAGE) to registry
+	@printf "\033[33;1m==== Pushing Docker image $(EPP_IMAGE) ====\033[0m\n"
+	$(CONTAINER_TOOL) push $(EPP_IMAGE)
+
+.PHONY image-pull
+image-pull: check-container-tool ## Pull all related images using $(CONTAINER_TOOL)
+	@printf "\033[33;1m==== Pulling Docker images ====\033[0m\n"
+	./scripts/pull_images.sh
+
 
 ##@ Install/Uninstall Targets
 
@@ -149,7 +156,7 @@ uninstall: uninstall-docker ## Default uninstall using Docker
 .PHONY: install-docker
 install-docker: check-container-tool ## Install app using $(CONTAINER_TOOL)
 	@echo "Starting container with $(CONTAINER_TOOL)..."
-	$(CONTAINER_TOOL) run -d --name $(PROJECT_NAME)-container $(IMG)
+	$(CONTAINER_TOOL) run -d --name $(PROJECT_NAME)-container $(EPP_IMAGE)
 	@echo "$(CONTAINER_TOOL) installation complete."
 	@echo "To use $(PROJECT_NAME), run:"
 	@echo "alias $(PROJECT_NAME)='$(CONTAINER_TOOL) exec -it $(PROJECT_NAME)-container /app/$(PROJECT_NAME)'"
@@ -194,12 +201,12 @@ uninstall-k8s: check-kubectl check-kustomize check-envsubst ## Uninstall from Ku
 
 .PHONY: install-openshift
 install-openshift: check-kubectl check-kustomize check-envsubst ## Install on OpenShift
-	@echo $$PROJECT_NAME $$NAMESPACE $$IMAGE_TAG_BASE $$VERSION
+	@echo $$PROJECT_NAME $$NAMESPACE $$EPP_IMAGE
 	@echo "Creating namespace $(NAMESPACE)..."
 	kubectl create namespace $(NAMESPACE) 2>/dev/null || true
 	@echo "Deploying common resources from deploy/ ..."
 	# Build and substitute the base manifests from deploy, then apply them
-	kustomize build deploy/environments/openshift-base | envsubst '$$PROJECT_NAME $$NAMESPACE $$IMAGE_TAG_BASE $$VERSION' | kubectl apply -n $(NAMESPACE) -f -
+	kustomize build deploy/environments/openshift-base | envsubst '$$PROJECT_NAME $$NAMESPACE $$EPP_IMAGE' | kubectl apply -n $(NAMESPACE) -f -
 	@echo "Waiting for pod to become ready..."
 	sleep 5
 	@POD=$$(kubectl get pod -l app=$(PROJECT_NAME)-statefulset -n $(NAMESPACE) -o jsonpath='{.items[0].metadata.name}'); \
@@ -210,9 +217,9 @@ install-openshift: check-kubectl check-kustomize check-envsubst ## Install on Op
 .PHONY: uninstall-openshift
 uninstall-openshift: check-kubectl check-kustomize check-envsubst ## Uninstall from OpenShift
 	@echo "Removing resources from OpenShift..."
-	kustomize build deploy/environments/openshift-base | envsubst '$$PROJECT_NAME $$NAMESPACE $$IMAGE_TAG_BASE $$VERSION' | kubectl delete --force -f - || true
+	kustomize build deploy/environments/openshift-base | envsubst '$$PROJECT_NAME $$NAMESPACE $$EPP_IMAGE' | kubectl delete --force -f - || true
 	# @if kubectl api-resources --api-group=route.openshift.io | grep -q Route; then \
-	#   envsubst '$$PROJECT_NAME $$NAMESPACE $$IMAGE_TAG_BASE $$VERSION' < deploy/openshift/route.yaml | kubectl delete --force -f - || true; \
+	#   envsubst '$$PROJECT_NAME $$NAMESPACE $$EPP_IMAGE' < deploy/openshift/route.yaml | kubectl delete --force -f - || true; \
 	# fi
 	@POD=$$(kubectl get pod -l app=$(PROJECT_NAME)-statefulset -n $(NAMESPACE) -o jsonpath='{.items[0].metadata.name}'); \
 	echo "Deleting pod: $$POD"; \
@@ -224,19 +231,20 @@ uninstall-openshift: check-kubectl check-kustomize check-envsubst ## Uninstall f
 .PHONY: install-rbac
 install-rbac: check-kubectl check-kustomize check-envsubst ## Install RBAC
 	@echo "Applying RBAC configuration from deploy/rbac..."
-	kustomize build deploy/environments/openshift-base/rbac | envsubst '$$PROJECT_NAME $$NAMESPACE $$IMAGE_TAG_BASE $$VERSION' | kubectl apply -f -
+	kustomize build deploy/environments/openshift-base/rbac | envsubst '$$PROJECT_NAME $$NAMESPACE $$EPP_IMAGE' | kubectl apply -f -
 
 .PHONY: uninstall-rbac
 uninstall-rbac: check-kubectl check-kustomize check-envsubst ## Uninstall RBAC
 	@echo "Removing RBAC configuration from deploy/rbac..."
-	kustomize build deploy/environments/openshift-base/rbac | envsubst '$$PROJECT_NAME $$NAMESPACE $$IMAGE_TAG_BASE $$VERSION' | kubectl delete -f - || true
+	kustomize build deploy/environments/openshift-base/rbac | envsubst '$$PROJECT_NAME $$NAMESPACE $$EPP_IMAGE' | kubectl delete -f - || true
 
 ##@ Environment
 .PHONY: env
 env: ## Print environment variables
-	@echo "IMAGE_TAG_BASE=$(IMAGE_TAG_BASE)"
-	@echo "IMG=$(IMG)"
+	@echo "EPP_IMAGE=$(EPP_IMAGE)"
 	@echo "CONTAINER_TOOL=$(CONTAINER_TOOL)"
+	@echo "NAMESPACE=${NAMESPACE}"
+	@echo "GIT_COMMIT_SHA=${GIT_COMMIT_SHA}"
 
 .PHONY: check-typos
 check-typos: $(TYPOS) ## Check for spelling errors using typos (exits with error if found)
@@ -356,8 +364,7 @@ env-dev-kind: ## Run under kind ($(KIND_CLUSTER_NAME))
 		$(MAKE) image-build && \
 		CLUSTER_NAME=$(KIND_CLUSTER_NAME) \
 		GATEWAY_HOST_PORT=$(KIND_GATEWAY_HOST_PORT) \
-		IMAGE_REGISTRY=$(IMAGE_REGISTRY) \
-		EPP_TAG=$(EPP_TAG) \
+		EPP_IMAGE=$(EPP_IMAGE) \
 		./scripts/kind-dev-env.sh; \
 	fi
 
diff --git a/deploy/components/inference-gateway/deployments.yaml b/deploy/components/inference-gateway/deployments.yaml
index 1b86e8ed..57ab61c9 100644
--- a/deploy/components/inference-gateway/deployments.yaml
+++ b/deploy/components/inference-gateway/deployments.yaml
@@ -18,7 +18,7 @@ spec:
       terminationGracePeriodSeconds: 130
       containers:
       - name: epp
-        image: ghcr.io/llm-d/llm-d-inference-scheduler:latest
+        image: ${EPP_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - --pool-name
diff --git a/deploy/components/inference-gateway/kustomization.yaml b/deploy/components/inference-gateway/kustomization.yaml
index 68fd981b..7f6dec8f 100644
--- a/deploy/components/inference-gateway/kustomization.yaml
+++ b/deploy/components/inference-gateway/kustomization.yaml
@@ -18,7 +18,3 @@ resources:
 - deployments.yaml
 - gateways.yaml
 - httproutes.yaml
-
-images:
-- name: ghcr.io/llm-d/llm-d-inference-scheduler
-  newTag: ${EPP_TAG}
diff --git a/deploy/components/vllm-sim-pd/deployments.yaml b/deploy/components/vllm-sim-pd/deployments.yaml
index 5a644b85..21368df3 100644
--- a/deploy/components/vllm-sim-pd/deployments.yaml
+++ b/deploy/components/vllm-sim-pd/deployments.yaml
@@ -17,7 +17,7 @@ spec:
     spec:
       containers:
       - name: vllm
-        image: ghcr.io/llm-d/llm-d-inference-sim:latest
+        image: ${VLLM_SIMULATOR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8000"
@@ -49,7 +49,7 @@ spec:
     spec:
       initContainers:
       - name: routing-sidecar
-        image: ghcr.io/llm-d/llm-d-routing-sidecar:latest
+        image: ${ROUTING_SIDECAR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8000"
@@ -61,7 +61,7 @@ spec:
         restartPolicy: Always
       containers:
       - name: vllm
-        image: ghcr.io/llm-d/llm-d-inference-sim:latest
+        image: ${VLLM_SIMULATOR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8200"
diff --git a/deploy/components/vllm-sim-pd/kustomization.yaml b/deploy/components/vllm-sim-pd/kustomization.yaml
index 14d7f1b3..40ac17d3 100644
--- a/deploy/components/vllm-sim-pd/kustomization.yaml
+++ b/deploy/components/vllm-sim-pd/kustomization.yaml
@@ -10,9 +10,3 @@ kind: Kustomization
 
 resources:
 - deployments.yaml
-
-images:
-- name: ghcr.io/llm-d/llm-d-inference-sim
-  newTag: ${VLLM_SIMULATOR_TAG}
-- name: ghcr.io/llm-d/llm-d-routing-sidecar
-  newTag: ${ROUTING_SIDECAR_TAG}
diff --git a/deploy/components/vllm-sim/deployments.yaml b/deploy/components/vllm-sim/deployments.yaml
index e6d8cae1..d116d53b 100644
--- a/deploy/components/vllm-sim/deployments.yaml
+++ b/deploy/components/vllm-sim/deployments.yaml
@@ -16,7 +16,7 @@ spec:
     spec:
       containers:
       - name: vllm
-        image: ghcr.io/llm-d/llm-d-inference-sim:latest
+        image: ${VLLM_SIMULATOR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8000"
diff --git a/deploy/components/vllm-sim/kustomization.yaml b/deploy/components/vllm-sim/kustomization.yaml
index 9748f239..40ac17d3 100644
--- a/deploy/components/vllm-sim/kustomization.yaml
+++ b/deploy/components/vllm-sim/kustomization.yaml
@@ -10,7 +10,3 @@ kind: Kustomization
 
 resources:
 - deployments.yaml
-
-images:
-- name: ghcr.io/llm-d/llm-d-inference-sim
-  newTag: ${VLLM_SIMULATOR_TAG}
diff --git a/deploy/environments/dev/kubernetes-kgateway/patch-deployments.yaml b/deploy/environments/dev/kubernetes-kgateway/patch-deployments.yaml
index 6b7c4430..4bddd5b7 100644
--- a/deploy/environments/dev/kubernetes-kgateway/patch-deployments.yaml
+++ b/deploy/environments/dev/kubernetes-kgateway/patch-deployments.yaml
@@ -7,7 +7,7 @@ spec:
     spec:
       containers:
       - name: epp
-        image: ${EPP_IMAGE}:${EPP_TAG}
+        image: ${EPP_IMAGE}
         imagePullPolicy: Always
         args:
         - --pool-name
diff --git a/deploy/environments/openshift-base/common/patch-statefulset.yaml b/deploy/environments/openshift-base/common/patch-statefulset.yaml
index 5b7676ff..c3a8a39c 100644
--- a/deploy/environments/openshift-base/common/patch-statefulset.yaml
+++ b/deploy/environments/openshift-base/common/patch-statefulset.yaml
@@ -16,5 +16,5 @@ spec:
       serviceAccountName: operator-controller-manager
       containers:
         - name: cmd
-          image: ${IMAGE_TAG_BASE}:${VERSION}
+          image: ${EPP_IMAGE}
           imagePullPolicy: Always
diff --git a/deploy/environments/openshift-base/kustomization.yaml b/deploy/environments/openshift-base/kustomization.yaml
index c690de16..9fff1639 100644
--- a/deploy/environments/openshift-base/kustomization.yaml
+++ b/deploy/environments/openshift-base/kustomization.yaml
@@ -22,12 +22,6 @@ configMapGenerator:
     disableNameSuffixHash: true
 
 # Include patches to update the Service, StatefulSet, Route, and RBAC resources.
-
-# Define the image to be updated.
-# images:
-# - name: ghcr.io/llm-d/placeholder
-#   newName: ghcr.io/llm-d/${IMAGE_TAG_BASE}
-#   newTag: ${VERSION}
 patches:
 - path: common/patch-service.yaml
 - path: common/patch-statefulset.yaml
diff --git a/scripts/kind-dev-env.sh b/scripts/kind-dev-env.sh
index 64770615..f8baf041 100755
--- a/scripts/kind-dev-env.sh
+++ b/scripts/kind-dev-env.sh
@@ -25,14 +25,11 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 # Set a default VLLM_SIMULATOR_IMAGE if not provided
 : "${VLLM_SIMULATOR_IMAGE:=llm-d-inference-sim}"
 
-# Set a default VLLM_SIMULATOR_TAG if not provided
-export VLLM_SIMULATOR_TAG="${VLLM_SIMULATOR_TAG:-latest}"
+# Set a default VLLM_SIMULATOR_IMAGE if not provided
+export VLLM_SIMULATOR_IMAGE="${VLLM_SIMULATOR_IMAGE:-ghcr.io/llm-d/llm-d-inference-sim:v0.4.0}"
 
 # Set a default EPP_IMAGE if not provided
-: "${EPP_IMAGE:=llm-d-inference-scheduler}"
-
-# Set a default EPP_TAG if not provided
-export EPP_TAG="${EPP_TAG:-dev}"
+export EPP_IMAGE="${EPP_IMAGE:-ghcr.io/llm-d/llm-d-inference-scheduler:dev}"
 
 # Set the model name to deploy
 export MODEL_NAME="${MODEL_NAME:-food-review}"
@@ -46,8 +43,8 @@ export MODEL_NAME_SAFE=$(echo "${MODEL_ID}" | tr '[:upper:]' '[:lower:]' | tr '
 # Set the endpoint-picker to deploy
 export EPP_NAME="${EPP_NAME:-${MODEL_NAME_SAFE}-endpoint-picker}"
 
-# Set the default routing side car image tag
-export ROUTING_SIDECAR_TAG="${ROUTING_SIDECAR_TAG:-0.0.6}"
+# Set the default routing side car image
+export ROUTING_SIDECAR_IMAGE="${ROUTING_SIDECAR_IMAGE:-ghcr.io/llm-d/llm-d-routing-sidecar:v0.2.0}"
 
 # Set the inference pool name for the deployment
 export POOL_NAME="${POOL_NAME:-${MODEL_NAME_SAFE}-inference-pool}"
@@ -152,19 +149,19 @@ kubectl --context ${KUBE_CONTEXT} -n local-path-storage wait --for=condition=Rea
 
 # Load the vllm simulator image into the cluster
 if [ "${CONTAINER_RUNTIME}" == "podman" ]; then
-	podman save ${IMAGE_REGISTRY}/${VLLM_SIMULATOR_IMAGE}:${VLLM_SIMULATOR_TAG} -o /dev/stdout | kind --name ${CLUSTER_NAME} load image-archive /dev/stdin
+	podman save ${VLLM_SIMULATOR_IMAGE} -o /dev/stdout | kind --name ${CLUSTER_NAME} load image-archive /dev/stdin
 else
-	if docker image inspect "${IMAGE_REGISTRY}/${VLLM_SIMULATOR_IMAGE}:${VLLM_SIMULATOR_TAG}" > /dev/null 2>&1; then
+	if docker image inspect ${VLLM_SIMULATOR_IMAGE} > /dev/null 2>&1; then
 		echo "INFO: Loading image into KIND cluster..."
-		kind --name ${CLUSTER_NAME} load docker-image ${IMAGE_REGISTRY}/${VLLM_SIMULATOR_IMAGE}:${VLLM_SIMULATOR_TAG}
+		kind --name ${CLUSTER_NAME} load docker-image ${VLLM_SIMULATOR_IMAGE}
 	fi
 fi
 
 # Load the ext_proc endpoint-picker image into the cluster
 if [ "${CONTAINER_RUNTIME}" == "podman" ]; then
-	podman save ${IMAGE_REGISTRY}/${EPP_IMAGE}:${EPP_TAG} -o /dev/stdout | kind --name ${CLUSTER_NAME} load image-archive /dev/stdin
+	podman save ${EPP_IMAGE} -o /dev/stdout | kind --name ${CLUSTER_NAME} load image-archive /dev/stdin
 else
-	kind --name ${CLUSTER_NAME} load docker-image ${IMAGE_REGISTRY}/${EPP_IMAGE}:${EPP_TAG}
+	kind --name ${CLUSTER_NAME} load docker-image ${EPP_IMAGE}
 fi
 # ------------------------------------------------------------------------------
 # CRD Deployment (Gateway API + GIE)
@@ -194,8 +191,8 @@ kubectl --context ${KUBE_CONTEXT} delete configmap epp-config --ignore-not-found
 kubectl --context ${KUBE_CONTEXT} create configmap epp-config --from-file=epp-config.yaml=${EPP_CONFIG}
 
 kustomize build --enable-helm  ${KUSTOMIZE_DIR} \
-	| envsubst '${POOL_NAME} ${MODEL_NAME} ${MODEL_NAME_SAFE} ${EPP_NAME} ${EPP_TAG} ${VLLM_SIMULATOR_TAG} \
-  ${PD_ENABLED} ${KV_CACHE_ENABLED} ${ROUTING_SIDECAR_TAG} \
+	| envsubst '${POOL_NAME} ${MODEL_NAME} ${MODEL_NAME_SAFE} ${EPP_NAME} ${EPP_IMAGE} ${VLLM_SIMULATOR_IMAGE} \
+  ${PD_ENABLED} ${KV_CACHE_ENABLED} ${ROUTING_SIDECAR_IMAGE} \
   ${VLLM_REPLICA_COUNT} ${VLLM_REPLICA_COUNT_P} ${VLLM_REPLICA_COUNT_D}' \
   | kubectl --context ${KUBE_CONTEXT} apply -f -
 
diff --git a/scripts/kubernetes-dev-env.sh b/scripts/kubernetes-dev-env.sh
index 71f355d2..27cd1bc7 100755
--- a/scripts/kubernetes-dev-env.sh
+++ b/scripts/kubernetes-dev-env.sh
@@ -31,9 +31,6 @@ if [[ ! -f "$VLLM_CHART_DIR/Chart.yaml" ]]; then
   exit 1
 fi
 
-# Default image registry for pulling deployment images
-export IMAGE_REGISTRY="${IMAGE_REGISTRY:-ghcr.io/llm-d}"
-
 # -----------------------------------------------------------------------------
 # Model Configuration
 # -----------------------------------------------------------------------------
@@ -72,10 +69,7 @@ export POOL_NAME="${POOL_NAME:-${MODEL_NAME_SAFE}-inference-pool}"
 export EPP_NAME="${EPP_NAME:-${MODEL_NAME_SAFE}-endpoint-picker}"
 
 # EPP container image name
-export EPP_IMAGE="${EPP_IMAGE:-${IMAGE_REGISTRY}/llm-d-inference-scheduler}"
-
-# EPP image tag
-export EPP_TAG="${EPP_TAG:-v0.1.0}"
+export EPP_IMAGE="${EPP_IMAGE:-ghcr.io/llm-d/llm-d-inference-scheduler:dev}"
 
 # Whether P/D mode is enabled for this deployment
 export PD_ENABLED="\"${PD_ENABLED:-false}\""
diff --git a/scripts/pull_images.sh b/scripts/pull_images.sh
new file mode 100644
index 00000000..61e8fbee
--- /dev/null
+++ b/scripts/pull_images.sh
@@ -0,0 +1,36 @@
+#!/bin/bash
+
+# Use the CONTAINER_TOOL from the environment, or default to docker if it's not set.
+CONTAINER_TOOL="${CONTAINER_TOOL:-docker}"
+echo "Using container tool: ${CONTAINER_TOOL}"
+
+export EPP_IMAGE="${EPP_IMAGE:-ghcr.io/llm-d/llm-d-inference-scheduler:dev}"
+export VLLM_SIMULATOR_IMAGE="${VLLM_SIMULATOR_IMAGE:-ghcr.io/llm-d/llm-d-inference-sim:v0.4.0}"
+export ROUTING_SIDECAR_IMAGE="${ROUTING_SIDECAR_IMAGE:-ghcr.io/llm-d/llm-d-routing-sidecar:v0.2.0}"
+
+# --- Print Final Images and Pull Dependencies ---
+echo "--- Using the following images for the E2E test ---"
+echo "Scheduler Image:     ${EPP_IMAGE}"
+echo "Simulator Image:     ${VLLM_SIMULATOR_IMAGE}"
+echo "Sidecar Image:       ${ROUTING_SIDECAR_IMAGE}"
+echo "----------------------------------------------------"
+
+echo "Pulling dependencies..."
+${CONTAINER_TOOL} pull ${EPP_IMAGE}
+if [[ $? != 0 ]]; then
+  echo "Failed to pull ${EPP_IMAGE}"
+  exit 1
+fi
+
+${CONTAINER_TOOL} pull ${VLLM_SIMULATOR_IMAGE}
+if [[ $? != 0 ]]; then
+  echo "Failed to pull ${VLLM_SIMULATOR_IMAGE}"
+  exit 1
+fi
+
+${CONTAINER_TOOL} pull ${ROUTING_SIDECAR_IMAGE}
+if [[ $? != 0 ]]; then
+  echo "Failed to pull ${ROUTING_SIDECAR_IMAGE}"
+  exit 1
+fi
+echo "Successfully pulled dependencies"
diff --git a/test/e2e/e2e_suite_test.go b/test/e2e/e2e_suite_test.go
index a69246b2..8ccff2d9 100644
--- a/test/e2e/e2e_suite_test.go
+++ b/test/e2e/e2e_suite_test.go
@@ -60,9 +60,9 @@ var (
 	port      string
 	scheme    = runtime.NewScheme()
 
-	eppTag            = env.GetEnvString("EPP_TAG", "dev", ginkgo.GinkgoLogr)
-	vllmSimTag        = env.GetEnvString("VLLM_SIMULATOR_TAG", "dev", ginkgo.GinkgoLogr)
-	routingSideCarTag = env.GetEnvString("ROUTING_SIDECAR_TAG", "v0.2.0", ginkgo.GinkgoLogr)
+	eppImage            = env.GetEnvString("EPP_IMAGE", "ghcr.io/llm-d/llm-d-inference-scheduler:dev", ginkgo.GinkgoLogr)
+	vllmSimImage        = env.GetEnvString("VLLM_SIMULATOR_IMAGE", "ghcr.io/llm-d/llm-d-inference-sim:dev", ginkgo.GinkgoLogr)
+	routingSideCarImage = env.GetEnvString("ROUTING_SIDECAR_IMAGE", "ghcr.io/llm-d/llm-d-routing-sidecar:v0.2.0", ginkgo.GinkgoLogr)
 
 	existsTimeout     = env.GetEnvDuration("EXISTS_TIMEOUT", defaultExistsTimeout, ginkgo.GinkgoLogr)
 	readyTimeout      = env.GetEnvDuration("READY_TIMEOUT", defaultReadyTimeout, ginkgo.GinkgoLogr)
@@ -118,20 +118,17 @@ func setupK8sCluster() {
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 
-	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image",
-		"ghcr.io/llm-d/llm-d-inference-sim:"+vllmSimTag)
+	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image", vllmSimImage)
 	session, err = gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 
-	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image",
-		"ghcr.io/llm-d/llm-d-inference-scheduler:"+eppTag)
+	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image", eppImage)
 	session, err = gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 
-	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image",
-		"ghcr.io/llm-d/llm-d-routing-sidecar:"+routingSideCarTag)
+	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image", routingSideCarImage)
 	session, err = gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
diff --git a/test/e2e/e2e_test.go b/test/e2e/e2e_test.go
index 54bf48e8..7b49954f 100644
--- a/test/e2e/e2e_test.go
+++ b/test/e2e/e2e_test.go
@@ -146,15 +146,15 @@ func createModelServers(withPD, withKV bool, vllmReplicas, prefillReplicas, deco
 	manifests := readYaml(yaml)
 	manifests = substituteMany(manifests,
 		map[string]string{
-			"${MODEL_NAME}":           theModelName,
-			"${MODEL_NAME_SAFE}":      theSafeModelName,
-			"${POOL_NAME}":            poolName,
-			"${KV_CACHE_ENABLED}":     strconv.FormatBool(withKV),
-			"${ROUTING_SIDECAR_TAG}":  routingSideCarTag,
-			"${VLLM_REPLICA_COUNT}":   strconv.Itoa(vllmReplicas),
-			"${VLLM_REPLICA_COUNT_D}": strconv.Itoa(decodeReplicas),
-			"${VLLM_REPLICA_COUNT_P}": strconv.Itoa(prefillReplicas),
-			"${VLLM_SIMULATOR_TAG}":   vllmSimTag,
+			"${MODEL_NAME}":            theModelName,
+			"${MODEL_NAME_SAFE}":       theSafeModelName,
+			"${POOL_NAME}":             poolName,
+			"${KV_CACHE_ENABLED}":      strconv.FormatBool(withKV),
+			"${ROUTING_SIDECAR_IMAGE}": routingSideCarImage,
+			"${VLLM_REPLICA_COUNT}":    strconv.Itoa(vllmReplicas),
+			"${VLLM_REPLICA_COUNT_D}":  strconv.Itoa(decodeReplicas),
+			"${VLLM_REPLICA_COUNT_P}":  strconv.Itoa(prefillReplicas),
+			"${VLLM_SIMULATOR_IMAGE}":  vllmSimImage,
 		})
 
 	objects := createObjsFromYaml(manifests)
@@ -183,7 +183,7 @@ func createEndPointPicker(eppConfig string) []string {
 	eppYamls := readYaml(eppManifest)
 	eppYamls = substituteMany(eppYamls,
 		map[string]string{
-			"${EPP_TAG}":   eppTag,
+			"${EPP_IMAGE}": eppImage,
 			"${POOL_NAME}": modelName + "-inference-pool",
 		})
 
diff --git a/test/e2e/yaml/deployments.yaml b/test/e2e/yaml/deployments.yaml
index 041ce44c..2f47722b 100644
--- a/test/e2e/yaml/deployments.yaml
+++ b/test/e2e/yaml/deployments.yaml
@@ -18,7 +18,7 @@ spec:
       terminationGracePeriodSeconds: 130
       containers:
       - name: epp
-        image: ghcr.io/llm-d/llm-d-inference-scheduler:${EPP_TAG}
+        image: ${EPP_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - --pool-name
diff --git a/test/e2e/yaml/vllm-sim-pd.yaml b/test/e2e/yaml/vllm-sim-pd.yaml
index c3f38f7b..779f397f 100644
--- a/test/e2e/yaml/vllm-sim-pd.yaml
+++ b/test/e2e/yaml/vllm-sim-pd.yaml
@@ -17,7 +17,7 @@ spec:
     spec:
       containers:
       - name: vllm
-        image: ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}
+        image: ${VLLM_SIMULATOR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8000"
@@ -60,7 +60,7 @@ spec:
     spec:
       initContainers:
       - name: routing-sidecar
-        image: ghcr.io/llm-d/llm-d-routing-sidecar:${ROUTING_SIDECAR_TAG}
+        image: ${ROUTING_SIDECAR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8000"
@@ -74,7 +74,7 @@ spec:
         restartPolicy: Always
       containers:
       - name: vllm
-        image: ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}
+        image: ${VLLM_SIMULATOR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--port=8200"
diff --git a/test/e2e/yaml/vllm-sim.yaml b/test/e2e/yaml/vllm-sim.yaml
index ce5a71af..36036996 100644
--- a/test/e2e/yaml/vllm-sim.yaml
+++ b/test/e2e/yaml/vllm-sim.yaml
@@ -16,7 +16,7 @@ spec:
     spec:
       containers:
       - name: vllm
-        image: ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}
+        image: ${VLLM_SIMULATOR_IMAGE}
         imagePullPolicy: IfNotPresent
         args:
         - "--mode=echo"
diff --git a/test/scripts/run_e2e.sh b/test/scripts/run_e2e.sh
index fe13508a..5d81e4ce 100755
--- a/test/scripts/run_e2e.sh
+++ b/test/scripts/run_e2e.sh
@@ -1,45 +1,5 @@
 #!/bin/bash
 
-# Use the CONTAINER_TOOL from the environment, or default to docker if it's not set.
-CONTAINER_TOOL="${CONTAINER_TOOL:-docker}"
-echo "Using container tool: ${CONTAINER_TOOL}"
-
-# Set a default EPP_TAG if not provided
-export EPP_TAG="${EPP_TAG:-dev}"
-
-# Set a default VLLM_SIMULATOR_TAG if not provided
-export VLLM_SIMULATOR_TAG="${VLLM_SIMULATOR_TAG:-v0.4.0}"
-
-# Set the default routing side car image tag
-export ROUTING_SIDECAR_TAG="${ROUTING_SIDECAR_TAG:-v0.2.0}"
-
-SIMTAG=$(${CONTAINER_TOOL} images | grep ghcr.io/llm-d/llm-d-inference-sim | awk '{print $2}' | grep ${VLLM_SIMULATOR_TAG})
-if [[ "${SIMTAG}" != "${VLLM_SIMULATOR_TAG}" ]]; then
-  ${CONTAINER_TOOL} pull ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}
-  if [[ $? != 0 ]]; then
-    echo "Failed to pull ghcr.io/llm-d/llm-d-inference-sim:${VLLM_SIMULATOR_TAG}"
-    exit 1
-  fi
-fi
-
-EPPTAG=$(${CONTAINER_TOOL} images | grep ghcr.io/llm-d/llm-d-inference-scheduler | awk '{print $2}' | grep ${EPP_TAG})
-if [[ "${EPPTAG}" != "${EPP_TAG}" ]]; then
-  ${CONTAINER_TOOL} pull ghcr.io/llm-d/llm-d-inference-scheduler:${EPP_TAG}
-  if [[ $? != 0 ]]; then
-    echo "Failed to pull ghcr.io/llm-d/llm-d-inference-scheduler:${EPP_TAG}"
-    exit 1
-  fi
-fi
-
-SIDECARTAG=$(${CONTAINER_TOOL} images | grep ghcr.io/llm-d/llm-d-routing-sidecar | awk '{print $2}' | grep ${ROUTING_SIDECAR_TAG})
-if [[ "${SIDECARTAG}" != "${ROUTING_SIDECAR_TAG}" ]]; then
-  ${CONTAINER_TOOL} pull ghcr.io/llm-d/llm-d-routing-sidecar:${ROUTING_SIDECAR_TAG}
-  if [[ $? != 0 ]]; then
-    echo "Failed to pull ghcr.io/llm-d/llm-d-routing-sidecar:${ROUTING_SIDECAR_TAG}"
-    exit 1
-  fi
-fi
-
 echo "Running end to end tests"
 
 DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"

From 40a7a9080ef8e512ff5e47333b89baff7c18c7d6 Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Mon, 6 Oct 2025 15:02:59 -0400
Subject: [PATCH 04/10] Cleanup

---
 scripts/kind-dev-env.sh | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/scripts/kind-dev-env.sh b/scripts/kind-dev-env.sh
index f8baf041..2b0f261c 100755
--- a/scripts/kind-dev-env.sh
+++ b/scripts/kind-dev-env.sh
@@ -23,10 +23,7 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 : "${IMAGE_REGISTRY:=ghcr.io/llm-d}"
 
 # Set a default VLLM_SIMULATOR_IMAGE if not provided
-: "${VLLM_SIMULATOR_IMAGE:=llm-d-inference-sim}"
-
-# Set a default VLLM_SIMULATOR_IMAGE if not provided
-export VLLM_SIMULATOR_IMAGE="${VLLM_SIMULATOR_IMAGE:-ghcr.io/llm-d/llm-d-inference-sim:v0.4.0}"
+export VLLM_SIMULATOR_IMAGE="${VLLM_SIMULATOR_IMAGE:-ghcr.io/llm-d/llm-d-inference-sim:latest}"
 
 # Set a default EPP_IMAGE if not provided
 export EPP_IMAGE="${EPP_IMAGE:-ghcr.io/llm-d/llm-d-inference-scheduler:dev}"

From 40e875efc5cdcda0f5e8b9301a22256dd3c845a6 Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Mon, 6 Oct 2025 15:06:48 -0400
Subject: [PATCH 05/10] Set VLLM_SIMULATOR_IMAGE and ROUTING_SIDECAR_IMAGE from
 the makefile

---
 Makefile | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/Makefile b/Makefile
index e6c7832a..bac1fe03 100644
--- a/Makefile
+++ b/Makefile
@@ -6,13 +6,23 @@ SHELL := /usr/bin/env bash
 # Defaults
 TARGETOS ?= $(shell go env GOOS)
 TARGETARCH ?= $(shell go env GOARCH)
+NAMESPACE ?= hc4ai-operator
+# Image default
 PROJECT_NAME ?= llm-d-inference-scheduler
 IMAGE_REGISTRY ?= ghcr.io/llm-d
 EPP_IMG_TAG_BASE ?= $(IMAGE_REGISTRY)/$(PROJECT_NAME)
 EPP_TAG ?= dev
 EPP_IMAGE = $(EPP_IMG_TAG_BASE):$(EPP_TAG)
 export EPP_IMAGE
-NAMESPACE ?= hc4ai-operator
+# Image dependency defaults
+VLLM_SIM_IMG_TAG_BASE ?= $(IMAGE_REGISTRY)/llm-d-inference-sim
+VLLM_SIMULATOR_TAG ?= latest
+VLLM_SIMULATOR_IMAGE = $(VLLM_SIM_IMG_TAG_BASE):${VLLM_SIMULATOR_TAG}
+export VLLM_SIMULATOR_IMAGE
+ROUTING_SIDECAR_IMG_TAG_BASE ?= $(IMAGE_REGISTRY)/llm-d-routing-sidecar
+ROUTING_SIDECAR_TAG ?= v0.2.0
+ROUTING_SIDECAR_IMAGE = ${ROUTING_SIDECAR_IMG_TAG_BASE}:${ROUTING_SIDECAR_TAG}
+export ROUTING_SIDECAR_IMAGE
 
 # Map go arch to typos arch
 ifeq ($(TARGETARCH),amd64)

From 47949982ceeedd69062082afbe260a72041b22e6 Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Mon, 13 Oct 2025 12:07:22 -0400
Subject: [PATCH 06/10] Make the image pull script more robust to local-only
 image tags

Signed-off-by: Hannah DeFazio <h2defazio@gmail.com>
---
 Makefile               |  2 +-
 scripts/pull_images.sh | 50 ++++++++++++++++++++++++++----------------
 2 files changed, 32 insertions(+), 20 deletions(-)
 mode change 100644 => 100755 scripts/pull_images.sh

diff --git a/Makefile b/Makefile
index bac1fe03..1313c4e1 100644
--- a/Makefile
+++ b/Makefile
@@ -146,7 +146,7 @@ image-push: check-container-tool ## Push Docker image $(EPP_IMAGE) to registry
 	@printf "\033[33;1m==== Pushing Docker image $(EPP_IMAGE) ====\033[0m\n"
 	$(CONTAINER_TOOL) push $(EPP_IMAGE)
 
-.PHONY image-pull
+.PHONY: image-pull
 image-pull: check-container-tool ## Pull all related images using $(CONTAINER_TOOL)
 	@printf "\033[33;1m==== Pulling Docker images ====\033[0m\n"
 	./scripts/pull_images.sh
diff --git a/scripts/pull_images.sh b/scripts/pull_images.sh
old mode 100644
new mode 100755
index 61e8fbee..3cfb32be
--- a/scripts/pull_images.sh
+++ b/scripts/pull_images.sh
@@ -4,33 +4,45 @@
 CONTAINER_TOOL="${CONTAINER_TOOL:-docker}"
 echo "Using container tool: ${CONTAINER_TOOL}"
 
-export EPP_IMAGE="${EPP_IMAGE:-ghcr.io/llm-d/llm-d-inference-scheduler:dev}"
+export EPP_IMAGE="${EPP_IMAGE:-ghcr.io/llm-d/llm-d-inference-scheduler:latest}"
 export VLLM_SIMULATOR_IMAGE="${VLLM_SIMULATOR_IMAGE:-ghcr.io/llm-d/llm-d-inference-sim:v0.4.0}"
 export ROUTING_SIDECAR_IMAGE="${ROUTING_SIDECAR_IMAGE:-ghcr.io/llm-d/llm-d-routing-sidecar:v0.2.0}"
 
+# --- Helper Function to Ensure Image Availability ---
+# This function checks the registry first, then falls back to a local-only check.
+ensure_image() {
+  local image_name="$1"
+  echo "Checking for image: ${image_name}"
+
+  # Attempt to inspect the image manifest on the remote registry.
+  if ${CONTAINER_TOOL} manifest inspect "${image_name}" > /dev/null 2>&1; then
+    echo " -> Image found on registry. Pulling..."
+    if ! ${CONTAINER_TOOL} pull "${image_name}"; then
+        echo "    ❌ ERROR: Failed to pull image '${image_name}'."
+        exit 1
+    fi
+    echo "    ✅ Successfully pulled image."
+  else
+    # If the image is not on the registry, check if it's already available locally.
+    echo " -> Image not found on registry. Checking for a local version..."
+    if [ -z "$(${CONTAINER_TOOL} images -q "${image_name}")" ]; then
+      # If it's not on the registry AND not local, it's an error.
+      echo "    ❌ ERROR: Image '${image_name}' is not available locally and could not be found on the registry."
+      exit 1
+    fi
+    echo " -> Found local-only image. Proceeding."
+  fi
+}
+
 # --- Print Final Images and Pull Dependencies ---
-echo "--- Using the following images for the E2E test ---"
+echo "--- Using the following images ---"
 echo "Scheduler Image:     ${EPP_IMAGE}"
 echo "Simulator Image:     ${VLLM_SIMULATOR_IMAGE}"
 echo "Sidecar Image:       ${ROUTING_SIDECAR_IMAGE}"
 echo "----------------------------------------------------"
 
 echo "Pulling dependencies..."
-${CONTAINER_TOOL} pull ${EPP_IMAGE}
-if [[ $? != 0 ]]; then
-  echo "Failed to pull ${EPP_IMAGE}"
-  exit 1
-fi
-
-${CONTAINER_TOOL} pull ${VLLM_SIMULATOR_IMAGE}
-if [[ $? != 0 ]]; then
-  echo "Failed to pull ${VLLM_SIMULATOR_IMAGE}"
-  exit 1
-fi
-
-${CONTAINER_TOOL} pull ${ROUTING_SIDECAR_IMAGE}
-if [[ $? != 0 ]]; then
-  echo "Failed to pull ${ROUTING_SIDECAR_IMAGE}"
-  exit 1
-fi
+ensure_image "${EPP_IMAGE}"
+ensure_image "${VLLM_SIMULATOR_IMAGE}"
+ensure_image "${ROUTING_SIDECAR_IMAGE}"
 echo "Successfully pulled dependencies"

From 2d376a20686e45b15dd4b3d60022eee8e260f56b Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Thu, 16 Oct 2025 20:06:42 -0400
Subject: [PATCH 07/10] Allow the e2e test setup to use podman

---
 test/e2e/e2e_suite_test.go | 63 +++++++++++++++++++++++++++++---------
 1 file changed, 49 insertions(+), 14 deletions(-)

diff --git a/test/e2e/e2e_suite_test.go b/test/e2e/e2e_suite_test.go
index 8ccff2d9..6a4de257 100644
--- a/test/e2e/e2e_suite_test.go
+++ b/test/e2e/e2e_suite_test.go
@@ -3,6 +3,7 @@ package e2e
 import (
 	"context"
 	"io"
+	"os"
 	"os/exec"
 	"strings"
 	"testing"
@@ -100,6 +101,51 @@ var _ = ginkgo.AfterSuite(func() {
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 })
 
+// loadImageIntoKind loads the specified image
+// into the Kind cluster using the most appropriate method based on the container runtime.
+func loadImageIntoKind(imageName string) {
+	container_runtime := env.GetEnvString("CONTAINER_TOOL", "docker", ginkgo.GinkgoLogr)
+	ginkgo.By("Loading image into Kind cluster: " + imageName)
+
+	switch container_runtime {
+	case "podman":
+		// Detect if podman is available
+		podmanPath, podmanErr := exec.LookPath("podman")
+		gomega.Expect(podmanErr).ShouldNot(gomega.HaveOccurred(), "Could not find podman in PATH")
+		ginkgo.GinkgoLogr.Info("Podman detected, using image-archive method.", "path", podmanPath)
+
+		// Create a temporary file to hold the image archive.
+		tmpFile, err := os.CreateTemp("", "image-archive-*.tar")
+		gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
+		// Ensure the temporary file is cleaned up when the function exits.
+		defer os.Remove(tmpFile.Name())
+
+		// Save the image to the temp file
+		cmdPodmanSave := exec.Command("podman", "save", "-o", tmpFile.Name(), imageName)
+		saveSession, err := gexec.Start(cmdPodmanSave, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
+		gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
+		gomega.Eventually(saveSession).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
+
+		// Load the temp file image into kind
+		cmdKindLoad := exec.Command("kind", "load", "image-archive", "--name", "e2e-tests", tmpFile.Name())
+		loadSession, err := gexec.Start(cmdKindLoad, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
+		gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
+		gomega.Eventually(loadSession).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
+	case "docker":
+		// Detect if podman is available
+		dockerPath, dockerErr := exec.LookPath("docker")
+		gomega.Expect(dockerErr).ShouldNot(gomega.HaveOccurred(), "Could not find docker in PATH")
+
+		ginkgo.GinkgoLogr.Info("Docker detected, using docker-image method.", "path", dockerPath)
+		command := exec.Command("kind", "load", "docker-image", "--name", "e2e-tests", imageName)
+		session, err := gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
+		gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
+		gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
+	default:
+		ginkgo.Fail("ERROR: Could not find 'podman' or 'docker' in the system's PATH. Please install one to continue.")
+	}
+}
+
 // Create the Kubernetes cluster for the E2E tests and load the local images
 func setupK8sCluster() {
 	command := exec.Command("kind", "create", "cluster", "--name", "e2e-tests", "--config", "-")
@@ -118,20 +164,9 @@ func setupK8sCluster() {
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 
-	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image", vllmSimImage)
-	session, err = gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
-	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
-	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
-
-	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image", eppImage)
-	session, err = gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
-	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
-	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
-
-	command = exec.Command("kind", "--name", "e2e-tests", "load", "docker-image", routingSideCarImage)
-	session, err = gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
-	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
-	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
+	loadImageIntoKind(vllmSimImage)
+	loadImageIntoKind(eppImage)
+	loadImageIntoKind(routingSideCarImage)
 }
 
 func setupK8sClient() {

From 5a23fdc42fea5cc8a5ff6f095a91c6ae3e3457ae Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Thu, 16 Oct 2025 20:32:52 -0400
Subject: [PATCH 08/10] Load CONTAINER_TOOL only once

---
 test/e2e/e2e_suite_test.go | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/test/e2e/e2e_suite_test.go b/test/e2e/e2e_suite_test.go
index 6a4de257..294aa944 100644
--- a/test/e2e/e2e_suite_test.go
+++ b/test/e2e/e2e_suite_test.go
@@ -61,6 +61,7 @@ var (
 	port      string
 	scheme    = runtime.NewScheme()
 
+	container_runtime   = env.GetEnvString("CONTAINER_TOOL", "docker", ginkgo.GinkgoLogr)
 	eppImage            = env.GetEnvString("EPP_IMAGE", "ghcr.io/llm-d/llm-d-inference-scheduler:dev", ginkgo.GinkgoLogr)
 	vllmSimImage        = env.GetEnvString("VLLM_SIMULATOR_IMAGE", "ghcr.io/llm-d/llm-d-inference-sim:dev", ginkgo.GinkgoLogr)
 	routingSideCarImage = env.GetEnvString("ROUTING_SIDECAR_IMAGE", "ghcr.io/llm-d/llm-d-routing-sidecar:v0.2.0", ginkgo.GinkgoLogr)
@@ -103,8 +104,7 @@ var _ = ginkgo.AfterSuite(func() {
 
 // loadImageIntoKind loads the specified image
 // into the Kind cluster using the most appropriate method based on the container runtime.
-func loadImageIntoKind(imageName string) {
-	container_runtime := env.GetEnvString("CONTAINER_TOOL", "docker", ginkgo.GinkgoLogr)
+func loadImageIntoKind(container_runtime string, imageName string) {
 	ginkgo.By("Loading image into Kind cluster: " + imageName)
 
 	switch container_runtime {
@@ -164,9 +164,9 @@ func setupK8sCluster() {
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 
-	loadImageIntoKind(vllmSimImage)
-	loadImageIntoKind(eppImage)
-	loadImageIntoKind(routingSideCarImage)
+	loadImageIntoKind(container_runtime, vllmSimImage)
+	loadImageIntoKind(container_runtime, eppImage)
+	loadImageIntoKind(container_runtime, routingSideCarImage)
 }
 
 func setupK8sClient() {

From f3332443003d40363b52f84f1ded15060c58f1ad Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Thu, 16 Oct 2025 20:36:58 -0400
Subject: [PATCH 09/10] Don't check if docker is installed as it is not
 directly run here

---
 test/e2e/e2e_suite_test.go | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/test/e2e/e2e_suite_test.go b/test/e2e/e2e_suite_test.go
index 294aa944..57370c9e 100644
--- a/test/e2e/e2e_suite_test.go
+++ b/test/e2e/e2e_suite_test.go
@@ -132,17 +132,13 @@ func loadImageIntoKind(container_runtime string, imageName string) {
 		gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 		gomega.Eventually(loadSession).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 	case "docker":
-		// Detect if podman is available
-		dockerPath, dockerErr := exec.LookPath("docker")
-		gomega.Expect(dockerErr).ShouldNot(gomega.HaveOccurred(), "Could not find docker in PATH")
-
-		ginkgo.GinkgoLogr.Info("Docker detected, using docker-image method.", "path", dockerPath)
+		ginkgo.GinkgoLogr.Info("Docker detected, using docker-image method.")
 		command := exec.Command("kind", "load", "docker-image", "--name", "e2e-tests", imageName)
 		session, err := gexec.Start(command, ginkgo.GinkgoWriter, ginkgo.GinkgoWriter)
 		gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 		gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 	default:
-		ginkgo.Fail("ERROR: Could not find 'podman' or 'docker' in the system's PATH. Please install one to continue.")
+		ginkgo.Fail("ERROR: The CONTAINER_TOOL value must be 'docker' or 'podman'")
 	}
 }
 

From 399d919465af3ea0c5a336a030837e82264df13a Mon Sep 17 00:00:00 2001
From: Hannah DeFazio <h2defazio@gmail.com>
Date: Thu, 16 Oct 2025 20:38:52 -0400
Subject: [PATCH 10/10] container_runtime does not need to be passed to the
 function

---
 test/e2e/e2e_suite_test.go | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/test/e2e/e2e_suite_test.go b/test/e2e/e2e_suite_test.go
index 57370c9e..d733ceac 100644
--- a/test/e2e/e2e_suite_test.go
+++ b/test/e2e/e2e_suite_test.go
@@ -104,7 +104,7 @@ var _ = ginkgo.AfterSuite(func() {
 
 // loadImageIntoKind loads the specified image
 // into the Kind cluster using the most appropriate method based on the container runtime.
-func loadImageIntoKind(container_runtime string, imageName string) {
+func loadImageIntoKind(imageName string) {
 	ginkgo.By("Loading image into Kind cluster: " + imageName)
 
 	switch container_runtime {
@@ -160,9 +160,9 @@ func setupK8sCluster() {
 	gomega.Expect(err).ShouldNot(gomega.HaveOccurred())
 	gomega.Eventually(session).WithTimeout(600 * time.Second).Should(gexec.Exit(0))
 
-	loadImageIntoKind(container_runtime, vllmSimImage)
-	loadImageIntoKind(container_runtime, eppImage)
-	loadImageIntoKind(container_runtime, routingSideCarImage)
+	loadImageIntoKind(vllmSimImage)
+	loadImageIntoKind(eppImage)
+	loadImageIntoKind(routingSideCarImage)
 }
 
 func setupK8sClient() {