revert: drop AWS/Azure workflow comment-only edits from #1644

Those files do not change serve behavior. #1644 is the GKE inference
cell; AWS/Azure steps stay disabled as on main.

Signed-off-by: Rohit Rajani <rorajani@nvidia.com>

Author: rorajani

.github/workflows/uat-aws.yaml

@@ -813,15 +813,18 @@ jobs:
           RUN_ID: ${{ github.run_id }}
         run: ./tests/uat/aws/run train "${TEST_CONFIG}"
 
-      # STILL DISABLED (step commented out below). phase_serve already selects the
-      # GPU pool for both graph components (#1644), so the Frontend no longer
-      # cold-pulls the ~12GB vllm-runtime image on a small CPU-pool node. Only
-      # uat-gcp.yaml re-enables the serve *step* for now (GKE x inference is what
-      # #1644 tracks); re-enable this one after that lane has a served completion,
-      # and restore the Serve summary row below to `${{ steps.serve.outcome }}`.
-      # Until then the inference STACK (dynamo + KAI scheduler + DRA driver) is
-      # still stood up and validated by prep/install/conformance/verify; only the
-      # served-workload run is skipped (Serve summary row is a static `disabled`).
+      # TEMPORARILY DISABLED (step commented out below). The vLLM workload run is
+      # excluded from UAT for now because the Frontend component pulls the full
+      # ~12GB vllm-runtime image on a fresh (non-GPU) node and never finishes
+      # within the 30m readiness budget: the decode worker reuses the GPU node's
+      # cached image (`already present on machine`), but the Frontend lands on a
+      # different node and pulls from scratch, wedging in ContainerCreating until
+      # the phase times out. The inference STACK (dynamo + KAI scheduler + DRA
+      # driver) is still stood up and validated by prep/install/conformance/verify;
+      # only the served-workload run is skipped (the Serve summary row is a static
+      # `disabled`). Re-enable by uncommenting the step once the pull is addressed
+      # (pre-pull/image cache, or pin the Frontend to the GPU node so it reuses
+      # the cached image). Kept in lockstep with uat-gcp.yaml. Tracked by #1644.
       #
       # - name: UAT - serve (DynamoGraphDeployment + endpoint)
       #   id: serve

.github/workflows/uat-azure.yaml

@@ -797,15 +797,18 @@ jobs:
           RUN_ID: ${{ github.run_id }}
         run: ./tests/uat/azure/run train "${TEST_CONFIG}"
 
-      # STILL DISABLED (step commented out below). phase_serve already selects the
-      # GPU pool for both graph components (#1644), so the Frontend no longer
-      # cold-pulls the ~12GB vllm-runtime image on a small CPU-pool node. Only
-      # uat-gcp.yaml re-enables the serve *step* for now (GKE x inference is what
-      # #1644 tracks); re-enable this one after that lane has a served completion,
-      # and restore the Serve summary row below to `${{ steps.serve.outcome }}`.
-      # Until then the inference STACK (dynamo + KAI scheduler + DRA driver) is
-      # still stood up and validated by prep/install/conformance/verify; only the
-      # served-workload run is skipped (Serve summary row is a static `disabled`).
+      # TEMPORARILY DISABLED (step commented out below). The vLLM workload run is
+      # excluded from UAT for now because the Frontend component pulls the full
+      # ~12GB vllm-runtime image on a fresh (non-GPU) node and never finishes
+      # within the 30m readiness budget: the decode worker reuses the GPU node's
+      # cached image (`already present on machine`), but the Frontend lands on a
+      # different node and pulls from scratch, wedging in ContainerCreating until
+      # the phase times out. The inference STACK (dynamo + KAI scheduler + DRA
+      # driver) is still stood up and validated by prep/install/conformance/verify;
+      # only the served-workload run is skipped (the Serve summary row is a static
+      # `disabled`). Re-enable by uncommenting the step once the pull is addressed
+      # (pre-pull/image cache, or pin the Frontend to the GPU node so it reuses
+      # the cached image).
       #
       # - name: UAT - serve (DynamoGraphDeployment + endpoint)
       #   id: serve