Split head memory and cpu requests/limits

Author: Bobbins228

src/codeflare_sdk/cluster/cluster.py

@@ -135,8 +135,10 @@ def create_app_wrapper(self):
 
         name = self.config.name
         namespace = self.config.namespace
-        head_cpus = self.config.head_cpus
-        head_memory = self.config.head_memory
+        head_cpu_requests = self.config.head_cpu_requests
+        head_cpu_limits = self.config.head_cpu_limits
+        head_memory_requests = self.config.head_memory_requests
+        head_memory_limits = self.config.head_memory_limits
         num_head_gpus = self.config.num_head_gpus
         worker_cpu_requests = self.config.worker_cpu_requests
         worker_cpu_limits = self.config.worker_cpu_limits
@@ -155,8 +157,10 @@ def create_app_wrapper(self):
         return generate_appwrapper(
             name=name,
             namespace=namespace,
-            head_cpus=head_cpus,
-            head_memory=head_memory,
+            head_cpu_requests=head_cpu_requests,
+            head_cpu_limits=head_cpu_limits,
+            head_memory_requests=head_memory_requests,
+            head_memory_limits=head_memory_limits,
             num_head_gpus=num_head_gpus,
             worker_cpu_requests=worker_cpu_requests,
             worker_cpu_limits=worker_cpu_limits,
@@ -887,12 +891,18 @@ def _map_to_ray_cluster(rc) -> Optional[RayCluster]:
         ]["resources"]["limits"]["cpu"],
         worker_gpu=0,  # hard to detect currently how many gpus, can override it with what the user asked for
         namespace=rc["metadata"]["namespace"],
-        head_cpus=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][0][
-            "resources"
-        ]["limits"]["cpu"],
-        head_mem=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][0][
-            "resources"
-        ]["limits"]["memory"],
+        head_cpu_requests=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][
+            0
+        ]["resources"]["requests"]["cpu"],
+        head_cpu_limits=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][
+            0
+        ]["resources"]["limits"]["cpu"],
+        head_mem_requests=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][
+            0
+        ]["resources"]["requests"]["memory"],
+        head_mem_limits=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][
+            0
+        ]["resources"]["limits"]["memory"],
         head_gpu=rc["spec"]["headGroupSpec"]["template"]["spec"]["containers"][0][
             "resources"
         ]["limits"]["nvidia.com/gpu"],
@@ -923,8 +933,10 @@ def _copy_to_ray(cluster: Cluster) -> RayCluster:
         worker_gpu=cluster.config.num_worker_gpus,
         namespace=cluster.config.namespace,
         dashboard=cluster.cluster_dashboard_uri(),
-        head_cpus=cluster.config.head_cpus,
-        head_mem=cluster.config.head_memory,
+        head_mem_requests=cluster.config.head_memory_requests,
+        head_mem_limits=cluster.config.head_memory_limits,
+        head_cpu_requests=cluster.config.head_cpu_requests,
+        head_cpu_limits=cluster.config.head_cpu_limits,
         head_gpu=cluster.config.num_head_gpus,
     )
     if ray.status == CodeFlareClusterStatus.READY:

src/codeflare_sdk/cluster/config.py

@@ -36,8 +36,12 @@ class ClusterConfiguration:
     name: str
     namespace: str = None
     head_info: list = field(default_factory=list)
-    head_cpus: typing.Union[int, str] = 2
-    head_memory: typing.Union[int, str] = 8
+    head_cpu_requests: typing.Union[int, str] = 2
+    head_cpu_limits: typing.Union[int, str] = 2
+    head_cpus: typing.Union[int, str] = None  # Deprecating
+    head_memory_requests: typing.Union[int, str] = 8
+    head_memory_limits: typing.Union[int, str] = 8
+    head_memory: typing.Union[int, str] = None  # Deprecating
     head_gpus: int = None  # Deprecating
     num_head_gpus: int = 0
     machine_types: list = field(default_factory=list)  # ["m4.xlarge", "g4dn.xlarge"]
@@ -74,8 +78,16 @@ def __post_init__(self):
         self._cpu_to_resource()
 
     def _str_mem_no_unit_add_GB(self):
-        if isinstance(self.head_memory, str) and self.head_memory.isdecimal():
-            self.head_memory = f"{self.head_memory}G"
+        if (
+            isinstance(self.head_memory_requests, str)
+            and self.head_memory_requests.isdecimal()
+        ):
+            self.head_memory_requests = f"{self.head_memory_requests}G"
+        if (
+            isinstance(self.head_memory_limits, str)
+            and self.head_memory_limits.isdecimal()
+        ):
+            self.head_memory_limits = f"{self.head_memory_limits}G"
         if (
             isinstance(self.worker_memory_requests, str)
             and self.worker_memory_requests.isdecimal()
@@ -88,8 +100,10 @@ def _str_mem_no_unit_add_GB(self):
             self.worker_memory_limits = f"{self.worker_memory_limits}G"
 
     def _memory_to_string(self):
-        if isinstance(self.head_memory, int):
-            self.head_memory = f"{self.head_memory}G"
+        if isinstance(self.head_memory_requests, int):
+            self.head_memory_requests = f"{self.head_memory_requests}G"
+        if isinstance(self.head_memory_limits, int):
+            self.head_memory_limits = f"{self.head_memory_limits}G"
         if isinstance(self.worker_memory_requests, int):
             self.worker_memory_requests = f"{self.worker_memory_requests}G"
         if isinstance(self.worker_memory_limits, int):
@@ -104,6 +118,11 @@ def _gpu_to_resource(self):
             self.num_worker_gpus = self.num_gpus
 
     def _cpu_to_resource(self):
+        if self.head_cpus:
+            warnings.warn(
+                "head_cpus is being deprecated, use head_cpu_requests and head_cpu_limits"
+            )
+            self.head_cpu_requests = self.head_cpu_limits = self.head_cpus
         if self.min_cpus:
             warnings.warn("min_cpus is being deprecated, use worker_cpu_requests")
             self.worker_cpu_requests = self.min_cpus
@@ -112,6 +131,11 @@ def _cpu_to_resource(self):
             self.worker_cpu_limits = self.max_cpus
 
     def _memory_to_resource(self):
+        if self.head_memory:
+            warnings.warn(
+                "head_memory is being deprecated, use head_memory_requests and head_memory_limits"
+            )
+            self.head_memory_requests = self.head_memory_limits = self.head_memory
         if self.min_memory:
             warnings.warn("min_memory is being deprecated, use worker_memory_requests")
             self.worker_memory_requests = f"{self.min_memory}G"

src/codeflare_sdk/cluster/model.py

@@ -72,8 +72,10 @@ class RayCluster:
 
     name: str
     status: RayClusterStatus
-    head_cpus: int
-    head_mem: str
+    head_cpu_requests: int
+    head_cpu_limits: int
+    head_mem_requests: str
+    head_mem_limits: str
     head_gpu: int
     workers: int
     worker_mem_min: str

src/codeflare_sdk/utils/generate_yaml.py

@@ -108,24 +108,24 @@ def update_env(spec, env):
 
 def update_resources(
     spec,
-    worker_cpu_requests,
-    worker_cpu_limits,
-    worker_memory_requests,
-    worker_memory_limits,
-    num_worker_gpus,
+    cpu_requests,
+    cpu_limits,
+    memory_requests,
+    memory_limits,
+    num_gpus,
 ):
     container = spec.get("containers")
     for resource in container:
         requests = resource.get("resources").get("requests")
         if requests is not None:
-            requests["cpu"] = worker_cpu_requests
-            requests["memory"] = worker_memory_requests
-            requests["nvidia.com/gpu"] = num_worker_gpus
+            requests["cpu"] = cpu_requests
+            requests["memory"] = memory_requests
+            requests["nvidia.com/gpu"] = num_gpus
         limits = resource.get("resources").get("limits")
         if limits is not None:
-            limits["cpu"] = worker_cpu_limits
-            limits["memory"] = worker_memory_limits
-            limits["nvidia.com/gpu"] = num_worker_gpus
+            limits["cpu"] = cpu_limits
+            limits["memory"] = memory_limits
+            limits["nvidia.com/gpu"] = num_gpus
 
 
 def update_nodes(
@@ -140,8 +140,10 @@ def update_nodes(
     image,
     env,
     image_pull_secrets,
-    head_cpus,
-    head_memory,
+    head_cpu_requests,
+    head_cpu_limits,
+    head_memory_requests,
+    head_memory_limits,
     num_head_gpus,
 ):
     head = cluster_yaml.get("spec").get("headGroupSpec")
@@ -163,7 +165,12 @@ def update_nodes(
         if comp == head:
             # TODO: Eventually add head node configuration outside of template
             update_resources(
-                spec, head_cpus, head_cpus, head_memory, head_memory, num_head_gpus
+                spec,
+                head_cpu_requests,
+                head_cpu_limits,
+                head_memory_requests,
+                head_memory_limits,
+                num_head_gpus,
             )
         else:
             update_resources(
@@ -277,8 +284,10 @@ def write_user_yaml(user_yaml, output_file_name):
 def generate_appwrapper(
     name: str,
     namespace: str,
-    head_cpus: int,
-    head_memory: int,
+    head_cpu_requests: int,
+    head_cpu_limits: int,
+    head_memory_requests: int,
+    head_memory_limits: int,
     num_head_gpus: int,
     worker_cpu_requests: int,
     worker_cpu_limits: int,
@@ -310,8 +319,10 @@ def generate_appwrapper(
         image,
         env,
         image_pull_secrets,
-        head_cpus,
-        head_memory,
+        head_cpu_requests,
+        head_cpu_limits,
+        head_memory_requests,
+        head_memory_limits,
         num_head_gpus,
     )
     augment_labels(cluster_yaml, labels)

tests/e2e/local_interactive_sdk_kind_test.py

@@ -37,8 +37,10 @@ def run_local_interactives(self):
                 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

@@ -34,8 +34,10 @@ def run_mnist_raycluster_sdk_kind(self):
                 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,
                 min_cpus="500m",
                 max_cpus=1,
                 min_memory=1,

tests/e2e/mnist_raycluster_sdk_kind_test.py

@@ -35,8 +35,10 @@ def run_mnist_raycluster_sdk_kind(self):
                 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=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/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

@@ -877,8 +877,10 @@ def test_ray_details(mocker, capsys):
         worker_gpu=0,
         namespace="ns",
         dashboard="fake-uri",
-        head_cpus=2,
-        head_mem=8,
+        head_cpu_requests=2,
+        head_cpu_limits=2,
+        head_mem_limits=8,
+        head_mem_requests=8,
         head_gpu=0,
     )
     mocker.patch(
@@ -2304,8 +2306,10 @@ def test_cluster_status(mocker):
         worker_gpu=0,
         namespace="ns",
         dashboard="fake-uri",
-        head_cpus=2,
-        head_mem=8,
+        head_cpu_requests=2,
+        head_cpu_limits=2,
+        head_mem_limits=8,
+        head_mem_requests=8,
         head_gpu=0,
     )
     cf = Cluster(

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=2,
+                head_cpu_requests=1,
+                head_cpu_limits=1,
+                head_memory_requests=2,
+                head_memory_limits=2,
                 worker_cpu_requests=1,
                 worker_cpu_limits=1,
                 worker_memory_requests=1,