test: update unit and e2e tests with split head resources

Signed-off-by: Bobbins228 <[email protected]>

Author: Bobbins228

.github/workflows/guided_notebook_tests.yaml

@@ -84,7 +84,7 @@ jobs:
           jq -r 'del(.cells[] | select(.source[] | contains("Create authentication object for user permissions")))' 0_basic_ray.ipynb > 0_basic_ray.ipynb.tmp && mv 0_basic_ray.ipynb.tmp 0_basic_ray.ipynb
           jq -r 'del(.cells[] | select(.source[] | contains("auth.logout()")))' 0_basic_ray.ipynb > 0_basic_ray.ipynb.tmp && mv 0_basic_ray.ipynb.tmp 0_basic_ray.ipynb
           # Set explicit namespace as SDK need it (currently) to resolve local queues
-          sed -i "s/head_memory=2,/head_memory=2, namespace='default',/" 0_basic_ray.ipynb
+          sed -i "s/head_memory_limits=2,/head_memory_limits=2, namespace='default',/" 0_basic_ray.ipynb
           # Run notebook
           poetry run papermill 0_basic_ray.ipynb 0_basic_ray_out.ipynb --log-output --execution-timeout 600
         working-directory: demo-notebooks/guided-demos
@@ -206,7 +206,7 @@ jobs:
           JOB_WAIT=$(jq -r '.' ${GITHUB_WORKSPACE}/.github/resources/wait_for_job_cell.json)
           jq --argjson job_wait "$JOB_WAIT" -r '(.cells[] | select(.source[] | contains("async for lines in client.tail_job_logs"))) |= $job_wait' 1_cluster_job_client.ipynb > 1_cluster_job_client.ipynb.tmp && mv 1_cluster_job_client.ipynb.tmp 1_cluster_job_client.ipynb
           # Set explicit namespace as SDK need it (currently) to resolve local queues
-          sed -i "s/head_cpus=1,/head_cpus=1, namespace='default',/" 1_cluster_job_client.ipynb
+          sed -i "s/head_cpu_limits=1,/head_cpu_limits=1, namespace='default',/" 1_cluster_job_client.ipynb
           # Run notebook
           poetry run papermill 1_cluster_job_client.ipynb 1_cluster_job_client_out.ipynb --log-output --execution-timeout 1200
         working-directory: demo-notebooks/guided-demos
@@ -332,7 +332,7 @@ jobs:
           # Rewrite cluster_uri() to local_client_url() to retrieve client URL available out of cluster, as the test is executed outside of cluster
           sed -i "s/cluster_uri()/local_client_url()/" 2_basic_interactive.ipynb
           # Set explicit namespace as SDK need it (currently) to resolve local queues
-          sed -i "s/head_cpus=1,/head_cpus=1, namespace='default',/" 2_basic_interactive.ipynb
+          sed -i "s/head_cpu_limits=1,/head_cpu_limits=1, namespace='default',/" 2_basic_interactive.ipynb
           # Add MINIO related modules to runtime environment
           sed -i "s/\\\\\"transformers/\\\\\"s3fs\\\\\", \\\\\"pyarrow\\\\\", \\\\\"transformers/" 2_basic_interactive.ipynb
           # Replace markdown cell with remote configuration for MINIO

tests/e2e/local_interactive_sdk_kind_test.py

@@ -44,8 +44,10 @@ def run_local_interactives(
                 name=cluster_name,
                 namespace=self.namespace,
                 num_workers=1,
-                head_cpus="500m",
-                head_memory=2,
+                head_cpu_requests="500m",
+                head_cpu_limits="500m",
+                head_memory_requests=2,
+                head_memory_limits=2,
                 worker_cpu_requests="500m",
                 worker_cpu_limits=1,
                 worker_memory_requests=1,

tests/e2e/mnist_raycluster_sdk_aw_kind_test.py

@@ -42,8 +42,10 @@ def run_mnist_raycluster_sdk_kind(
                 name="mnist",
                 namespace=self.namespace,
                 num_workers=1,
-                head_cpus="500m",
-                head_memory=2,
+                head_cpu_requests="500m",
+                head_cpu_limits="500m",
+                head_memory_requests=2,
+                head_memory_limits=2,
                 worker_cpu_requests="500m",
                 worker_cpu_limits=1,
                 worker_memory_requests=1,

tests/e2e/mnist_raycluster_sdk_kind_test.py

@@ -42,8 +42,10 @@ def run_mnist_raycluster_sdk_kind(
                 name="mnist",
                 namespace=self.namespace,
                 num_workers=1,
-                head_cpus="500m",
-                head_memory=2,
+                head_cpu_requests="500m",
+                head_cpu_limits="500m",
+                head_memory_requests=2,
+                head_memory_limits=2,
                 worker_cpu_requests="500m",
                 worker_cpu_limits=1,
                 worker_memory_requests=1,

tests/e2e/mnist_raycluster_sdk_oauth_test.py

@@ -42,8 +42,10 @@ def run_mnist_raycluster_sdk_oauth(self):
                 name="mnist",
                 namespace=self.namespace,
                 num_workers=1,
-                head_cpus="500m",
-                head_memory=4,
+                head_cpu_requests="500m",
+                head_cpu_limits="500m",
+                head_memory_requests=4,
+                head_memory_limits=4,
                 worker_cpu_requests=1,
                 worker_cpu_limits=1,
                 worker_memory_requests=1,

tests/e2e/start_ray_cluster.py

@@ -13,8 +13,10 @@
         name="mnist",
         namespace=namespace,
         num_workers=1,
-        head_cpus="500m",
-        head_memory=2,
+        head_cpu_requests="500m",
+        head_cpu_limits="500m",
+        head_memory_requests=2,
+        head_memory_limits=2,
         worker_cpu_requests="500m",
         worker_cpu_limits=1,
         worker_memory_requests=1,

tests/unit_test.py

@@ -942,13 +942,15 @@ def test_ray_details(mocker, capsys):
         name="raytest1",
         status=RayClusterStatus.READY,
         workers=1,
-        worker_mem_min="2G",
-        worker_mem_max="2G",
+        worker_mem_requests="2G",
+        worker_mem_limits="2G",
         worker_cpu=1,
         namespace="ns",
         dashboard="fake-uri",
-        head_cpus=2,
-        head_mem=8,
+        head_cpu_requests=2,
+        head_cpu_limits=2,
+        head_mem_requests=8,
+        head_mem_limits=8,
     )
     mocker.patch(
         "codeflare_sdk.cluster.cluster.Cluster.status",
@@ -978,8 +980,8 @@ def test_ray_details(mocker, capsys):
     assert ray2.name == "raytest2"
     assert ray1.namespace == ray2.namespace
     assert ray1.workers == ray2.workers
-    assert ray1.worker_mem_min == ray2.worker_mem_min
-    assert ray1.worker_mem_max == ray2.worker_mem_max
+    assert ray1.worker_mem_requests == ray2.worker_mem_requests
+    assert ray1.worker_mem_limits == ray2.worker_mem_limits
     assert ray1.worker_cpu == ray2.worker_cpu
     assert ray1.worker_extended_resources == ray2.worker_extended_resources
     try:
@@ -2356,13 +2358,15 @@ def test_cluster_status(mocker):
         name="test",
         status=RayClusterStatus.UNKNOWN,
         workers=1,
-        worker_mem_min=2,
-        worker_mem_max=2,
+        worker_mem_requests=2,
+        worker_mem_limits=2,
         worker_cpu=1,
         namespace="ns",
         dashboard="fake-uri",
-        head_cpus=2,
-        head_mem=8,
+        head_cpu_requests=2,
+        head_cpu_limits=2,
+        head_mem_requests=8,
+        head_mem_limits=8,
     )
     cf = Cluster(
         ClusterConfiguration(

tests/upgrade/raycluster_sdk_upgrade_sleep_test.py

@@ -53,8 +53,10 @@ def run_mnist_raycluster_sdk_oauth(self):
                 name="mnist",
                 namespace=self.namespace,
                 num_workers=1,
-                head_cpus=1,
-                head_memory=4,
+                head_cpu_requests=1,
+                head_cpu_limits=1,
+                head_memory_requests=4,
+                head_memory_limits=4,
                 worker_cpu_requests=1,
                 worker_cpu_limits=1,
                 worker_memory_requests=4,

tests/upgrade/raycluster_sdk_upgrade_test.py

@@ -48,8 +48,10 @@ def run_mnist_raycluster_sdk_oauth(self):
                 name="mnist",
                 namespace=self.namespace,
                 num_workers=1,
-                head_cpus=1,
-                head_memory=4,
+                head_cpu_requests=1,
+                head_cpu_limits=1,
+                head_memory_requests=4,
+                head_memory_limits=4,
                 worker_cpu_requests=1,
                 worker_cpu_limits=1,
                 worker_memory_requests=4,