remove instascale

Author: dgrove-oss

docs/cluster-configuration.md

@@ -18,9 +18,8 @@ cluster = Cluster(ClusterConfiguration(
     min_memory=2, # Default 2
     max_memory=2, # Default 2
     num_gpus=0, # Default 0
-    appwrapper=True, # Default True
+    appwrapper=True, # Default False
     image="quay.io/project-codeflare/ray:latest-py39-cu118", # Mandatory Field
-    instascale=False, # Default False
     machine_types=["m5.xlarge", "g4dn.xlarge"],
 ))
 ```

src/codeflare_sdk/cluster/cluster.py

@@ -146,7 +146,6 @@ def create_app_wrapper(self):
         workers = self.config.num_workers
         template = self.config.template
         image = self.config.image
-        instascale = self.config.instascale
         appwrapper = self.config.appwrapper
         instance_types = self.config.machine_types
         env = self.config.envs
@@ -168,7 +167,6 @@ def create_app_wrapper(self):
             workers=workers,
             template=template,
             image=image,
-            instascale=instascale,
             appwrapper=appwrapper,
             instance_types=instance_types,
             env=env,
@@ -497,7 +495,6 @@ def from_k8_cluster_object(
                     "resources"
                 ]["limits"]["nvidia.com/gpu"]
             ),
-            instascale=True if machine_types else False,
             image=rc["spec"]["workerGroupSpecs"][0]["template"]["spec"]["containers"][
                 0
             ]["image"],

src/codeflare_sdk/cluster/config.py

@@ -46,7 +46,6 @@ class ClusterConfiguration:
     max_memory: typing.Union[int, str] = 2
     num_gpus: int = 0
     template: str = f"{dir}/templates/base-template.yaml"
-    instascale: bool = False
     appwrapper: bool = False
     envs: dict = field(default_factory=dict)
     image: str = ""

src/codeflare_sdk/utils/generate_yaml.py

@@ -87,20 +87,9 @@ def update_names(yaml, item, appwrapper_name, cluster_name, namespace):
     lower_meta["namespace"] = namespace
 
 
-def update_labels(yaml, instascale, instance_types):
+def update_labels(yaml, instance_types):
     metadata = yaml.get("metadata")
-    if instascale:
-        if not len(instance_types) > 0:
-            sys.exit(
-                "If instascale is set to true, must provide at least one instance type"
-            )
-        type_str = ""
-        for type in instance_types:
-            type_str += type + "_"
-        type_str = type_str[:-1]
-        metadata["labels"]["orderedinstance"] = type_str
-    else:
-        metadata.pop("labels")
+    metadata.pop("labels")
 
 
 def update_priority(yaml, item):
@@ -158,18 +147,8 @@ def update_custompodresources(
         sys.exit("Error: malformed template")
 
 
-def update_affinity(spec, appwrapper_name, instascale):
-    if instascale:
-        node_selector_terms = (
-            spec.get("affinity")
-            .get("nodeAffinity")
-            .get("requiredDuringSchedulingIgnoredDuringExecution")
-            .get("nodeSelectorTerms")
-        )
-        node_selector_terms[0]["matchExpressions"][0]["values"][0] = appwrapper_name
-        node_selector_terms[0]["matchExpressions"][0]["key"] = appwrapper_name
-    else:
-        spec.pop("affinity")
+def update_affinity(spec, appwrapper_name):
+    spec.pop("affinity")
 
 
 def update_image(spec, image):
@@ -220,7 +199,6 @@ def update_nodes(
     gpu,
     workers,
     image,
-    instascale,
     env,
     image_pull_secrets,
     head_cpus,
@@ -241,7 +219,7 @@ def update_nodes(
 
         for comp in [head, worker]:
             spec = comp.get("template").get("spec")
-            update_affinity(spec, appwrapper_name, instascale)
+            update_affinity(spec, appwrapper_name)
             update_image_pull_secrets(spec, image_pull_secrets)
             update_image(spec, image)
             update_env(spec, env)
@@ -373,7 +351,6 @@ def generate_appwrapper(
     workers: int,
     template: str,
     image: str,
-    instascale: bool,
     appwrapper: bool,
     instance_types: list,
     env,
@@ -393,7 +370,7 @@ def generate_appwrapper(
         cluster_name,
         namespace,
     )
-    update_labels(user_yaml, instascale, instance_types)
+    update_labels(user_yaml, instance_types)
     update_priority(user_yaml, item)
     update_custompodresources(
         item,
@@ -417,7 +394,6 @@ def generate_appwrapper(
         gpu,
         workers,
         image,
-        instascale,
         env,
         image_pull_secrets,
         head_cpus,

tests/e2e/start_ray_cluster.py

@@ -20,7 +20,6 @@
         min_memory=1,
         max_memory=2,
         num_gpus=0,
-        instascale=False,
         image=ray_image,
         appwrapper=True,
     )

tests/test-case-no-mcad.yamls

@@ -29,15 +29,6 @@ spec:
     serviceType: ClusterIP
     template:
       spec:
-        affinity:
-          nodeAffinity:
-            requiredDuringSchedulingIgnoredDuringExecution:
-              nodeSelectorTerms:
-              - matchExpressions:
-                - key: unit-test-cluster-ray
-                  operator: In
-                  values:
-                  - unit-test-cluster-ray
         containers:
         - image: quay.io/project-codeflare/ray:latest-py39-cu118
           imagePullPolicy: Always
@@ -111,15 +102,6 @@ spec:
         labels:
           key: value
       spec:
-        affinity:
-          nodeAffinity:
-            requiredDuringSchedulingIgnoredDuringExecution:
-              nodeSelectorTerms:
-              - matchExpressions:
-                - key: unit-test-cluster-ray
-                  operator: In
-                  values:
-                  - unit-test-cluster-ray
         containers:
         - image: quay.io/project-codeflare/ray:latest-py39-cu118
           lifecycle:

tests/test-case.yaml

@@ -1,8 +1,6 @@
 apiVersion: workload.codeflare.dev/v1beta1
 kind: AppWrapper
 metadata:
-  labels:
-    orderedinstance: cpu.small_gpu.large
   name: unit-test-cluster
   namespace: ns
 spec:
@@ -58,15 +56,6 @@ spec:
             serviceType: ClusterIP
             template:
               spec:
-                affinity:
-                  nodeAffinity:
-                    requiredDuringSchedulingIgnoredDuringExecution:
-                      nodeSelectorTerms:
-                      - matchExpressions:
-                        - key: unit-test-cluster
-                          operator: In
-                          values:
-                          - unit-test-cluster
                 containers:
                 - image: quay.io/project-codeflare/ray:latest-py39-cu118
                   imagePullPolicy: Always
@@ -140,15 +129,6 @@ spec:
                 labels:
                   key: value
               spec:
-                affinity:
-                  nodeAffinity:
-                    requiredDuringSchedulingIgnoredDuringExecution:
-                      nodeSelectorTerms:
-                      - matchExpressions:
-                        - key: unit-test-cluster
-                          operator: In
-                          values:
-                          - unit-test-cluster
                 containers:
                 - image: quay.io/project-codeflare/ray:latest-py39-cu118
                   lifecycle:

tests/unit_test.py

@@ -252,7 +252,6 @@ def test_config_creation():
     assert config.num_gpus == 7
     assert config.image == "quay.io/project-codeflare/ray:latest-py39-cu118"
     assert config.template == f"{parent}/src/codeflare_sdk/templates/base-template.yaml"
-    assert config.instascale
     assert config.machine_types == ["cpu.small", "gpu.large"]
     assert config.image_pull_secrets == ["unit-test-pull-secret"]
     assert config.appwrapper == True
@@ -365,7 +364,6 @@ def test_cluster_creation_no_mcad_local_queue(mocker):
         min_memory=5,
         max_memory=6,
         num_gpus=7,
-        instascale=True,
         machine_types=["cpu.small", "gpu.large"],
         image_pull_secrets=["unit-test-pull-secret"],
         image="quay.io/project-codeflare/ray:latest-py39-cu118",
@@ -2447,7 +2445,6 @@ def test_get_cluster(mocker):
     assert cluster_config.min_cpus == 1 and cluster_config.max_cpus == 1
     assert cluster_config.min_memory == "2G" and cluster_config.max_memory == "2G"
     assert cluster_config.num_gpus == 0
-    assert cluster_config.instascale
     assert (
         cluster_config.image
         == "ghcr.io/foundation-model-stack/base:ray2.1.0-py38-gpu-pytorch1.12.0cu116-20221213-193103"

tests/unit_test_support.py

@@ -15,7 +15,6 @@ def createClusterConfig():
         max_memory=6,
         num_gpus=7,
         appwrapper=True,
-        instascale=True,
         machine_types=["cpu.small", "gpu.large"],
         image_pull_secrets=["unit-test-pull-secret"],
         image="quay.io/project-codeflare/ray:latest-py39-cu118",