[k8s] Update k8s docker image to use ray 2.9.3 (#3350)

* Update to ray 2.9.3

* remove serve tests for k8s

* remove tests with autostop for k8s

Dockerfile_k8s

@@ -28,7 +28,7 @@ USER sky
 # Install SkyPilot pip dependencies preemptively to speed up provisioning time
 RUN pip install wheel Click colorama cryptography jinja2 jsonschema && \
     pip install networkx oauth2client pandas pendulum PrettyTable && \
-    pip install ray[default]==2.4.0 rich tabulate filelock && \
+    pip install ray[default]==2.9.3 rich tabulate filelock && \
     pip install packaging 'protobuf<4.0.0' pulp && \
     pip install pycryptodome==3.12.0 && \
     pip install docker kubernetes==28.1.0

Dockerfile_k8s_gpu

@@ -1,7 +1,7 @@
 # TODO(romilb) - The base image used here (ray) is very large (11.4GB).
 #  as a result, this built image is about 13.5GB. We need to pick a lighter base
 #  image.
-FROM rayproject/ray:2.4.0-py310-gpu
+FROM rayproject/ray:2.9.3-py310-gpu
 
 # Initialize conda for root user, install ssh and other local dependencies
 # We remove cuda lists to avoid conflicts with the cuda version installed by ray
@@ -39,8 +39,7 @@ RUN pip install wheel Click colorama cryptography jinja2 jsonschema && \
     pip install rich tabulate filelock && \
     pip install packaging 'protobuf<4.0.0' pulp && \
     pip install pycryptodome==3.12.0 && \
-    pip install docker kubernetes==28.1.0 && \
-    pip install -U boto3==1.34.46 # Ray base image has a stale boto3 version - https://github.com/skypilot-org/skypilot/issues/3158
+    pip install docker kubernetes==28.1.0
 
 # Add /home/sky/.local/bin/ to PATH
 RUN echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc

tests/test_smoke.py

@@ -683,6 +683,7 @@ def test_image_no_conda():
     run_one_test(test)
 
 
+@pytest.mark.no_kubernetes  # Kubernetes does not support stopping instances
 def test_custom_default_conda_env(generic_cloud: str):
     name = _get_cluster_name()
     test = Test('custom_default_conda_env', [
@@ -3027,6 +3028,7 @@ def test_skyserve_azure_http():
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_llm(generic_cloud: str):
     """Test skyserve with real LLM usecase"""
     name = _get_service_name()
@@ -3084,6 +3086,7 @@ def test_skyserve_spot_recovery():
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_base_ondemand_fallback(generic_cloud: str):
     name = _get_service_name()
     test = Test(
@@ -3147,6 +3150,7 @@ def test_skyserve_dynamic_ondemand_fallback():
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_user_bug_restart(generic_cloud: str):
     """Tests that we restart the service after user bug."""
     # TODO(zhwu): this behavior needs some rethinking.
@@ -3180,6 +3184,7 @@ def test_skyserve_user_bug_restart(generic_cloud: str):
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_load_balancer(generic_cloud: str):
     """Test skyserve load balancer round-robin policy"""
     name = _get_service_name()
@@ -3203,6 +3208,7 @@ def test_skyserve_load_balancer(generic_cloud: str):
 
 @pytest.mark.gcp
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_auto_restart():
     """Test skyserve with auto restart"""
     name = _get_service_name()
@@ -3244,6 +3250,7 @@ def test_skyserve_auto_restart():
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_cancel(generic_cloud: str):
     """Test skyserve with cancel"""
     name = _get_service_name()
@@ -3269,6 +3276,7 @@ def test_skyserve_cancel(generic_cloud: str):
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_update(generic_cloud: str):
     """Test skyserve with update"""
     name = _get_service_name()
@@ -3297,6 +3305,7 @@ def test_skyserve_update(generic_cloud: str):
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_rolling_update(generic_cloud: str):
     """Test skyserve with rolling update"""
     name = _get_service_name()
@@ -3333,6 +3342,7 @@ def test_skyserve_rolling_update(generic_cloud: str):
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_fast_update(generic_cloud: str):
     """Test skyserve with fast update (Increment version of old replicas)"""
     name = _get_service_name()
@@ -3374,6 +3384,7 @@ def test_skyserve_fast_update(generic_cloud: str):
 
 
 @pytest.mark.serve
+@pytest.mark.no_kubernetes
 def test_skyserve_update_autoscale(generic_cloud: str):
     """Test skyserve update with autoscale"""
     name = _get_service_name()
@@ -3411,6 +3422,7 @@ def test_skyserve_update_autoscale(generic_cloud: str):
 
 @pytest.mark.serve
 @pytest.mark.parametrize('mode', ['rolling', 'blue_green'])
+@pytest.mark.no_kubernetes
 def test_skyserve_new_autoscaler_update(mode: str, generic_cloud: str):
     """Test skyserve with update that changes autoscaler"""
     name = _get_service_name() + mode