ob-metaflow-extensions 1.3.1__tar.gz → 1.3.2__tar.gz

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ob_metaflow_extensions
-Version: 1.3.1
+Version: 1.3.2
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/metaflow_extensions/outerbounds/plugins/apps/core/_state_machine.py

@@ -180,6 +180,7 @@ class WorkerInfoDict(TypedDict):
     pending: Dict[str, List[WorkerStatus]]
     running: Dict[str, List[WorkerStatus]]
     crashlooping: Dict[str, List[WorkerStatus]]
+    failed: Dict[str, List[WorkerStatus]]
 
 
 class CurrentWorkerInfo(TypedDict):
@@ -442,9 +443,11 @@ def _capsule_worker_semantic_status(
                 xx[worker_version].append(w)
         return xx
 
+    # phases can be Pending, Running, Succeeded, Failed, Unknown, CrashLoopBackOff
     pending_workers = _make_version_dict(workers, "Pending")
     running_workers = _make_version_dict(workers, "Running")
     crashlooping_workers = _make_version_dict(workers, "CrashLoopBackOff")
+    failed_workers = _make_version_dict(workers, "Failed")
 
     # current_status (formulated basis):
     # - at least one pods are pending for `_end_state_capsule_version`
@@ -464,7 +467,8 @@ def _capsule_worker_semantic_status(
         "at_least_one_running": (
             count_for_version(running_workers) >= min(min_replicas, 1)
         ),
-        "at_least_one_crashlooping": count_for_version(crashlooping_workers) > 0,
+        "at_least_one_crashlooping": count_for_version(crashlooping_workers) > 0
+        or count_for_version(failed_workers) > 0,
         "none_present": (
             count_for_version(running_workers) == 0
             and count_for_version(pending_workers) == 0
@@ -484,6 +488,7 @@ def _capsule_worker_semantic_status(
             "pending": count_for_version(pending_workers),
             "running": count_for_version(running_workers),
             "crashlooping": count_for_version(crashlooping_workers),
+            "failed": count_for_version(failed_workers),
         },
     }
 
@@ -491,6 +496,7 @@ def _capsule_worker_semantic_status(
         "pending": pending_workers,
         "running": running_workers,
         "crashlooping": crashlooping_workers,
+        "failed": failed_workers,
     }
 
     return {

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/metaflow_extensions/outerbounds/plugins/apps/core/capsule.py

@@ -255,6 +255,12 @@ class CapsuleInput:
             replicas.get("min"),
             replicas.get("max"),
         )
+        rpm = replicas.get("scaling_policy", {}).get("rpm", None)
+        autoscaling_config = {}
+        if rpm:
+            autoscaling_config = {
+                "requestRateBasedAutoscalingConfig": {"targetRequestsPerMinute": rpm}
+            }
         if fixed is not None:
             _min, _max = fixed, fixed
         gpu_resource = app_config.get_state("resources").get("gpu")
@@ -296,6 +302,7 @@ class CapsuleInput:
             "autoscalingConfig": {
                 "minReplicas": _min,
                 "maxReplicas": _max,
+                **autoscaling_config,
             },
             **_scheduling_config,
             "containerStartupConfig": {
@@ -713,7 +720,7 @@ class CapsuleDeployer:
             workers_status: List[WorkerStatus],
         ):
             for worker in workers_status:
-                if worker["phase"] == "CrashLoopBackOff":
+                if worker["phase"] == "CrashLoopBackOff" or worker["phase"] == "Failed":
                     return worker["workerId"]
             return None
 
@@ -851,7 +858,7 @@ class CapsuleDeployer:
                 workers_state_machine.save_debug_info(self._debug_dir)
                 if i % 3 == 0:  # Every 3 seconds report the status
                     logger(
-                        f"[debug] 💊 {self.capsule_type} {self.identifier} deployment status: {state_machine.current_status} | worker states: {workers_state_machine.current_status}"
+                        f"[debug] 💊 {self.capsule_type} {self.identifier} deployment status: {state_machine.current_status} | worker states: {workers_state_machine.current_status} | capsule_ready : {capsule_ready} | further_check_worker_readiness {further_check_worker_readiness}"
                     )
 
         # We will only check ready_to_serve_traffic under the following conditions:

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/metaflow_extensions/outerbounds/plugins/apps/core/config/typed_configs.py

@@ -51,6 +51,11 @@ class ReplicaConfigDict(TypedDict, total=False):
     fixed: Optional[int]
     min: Optional[int]
     max: Optional[int]
+    scaling_policy: Optional["ScalingPolicyConfigDict"]
+
+
+class ScalingPolicyConfigDict(TypedDict, total=False):
+    rpm: Optional[int]
 
 
 class DependencyConfigDict(TypedDict, total=False):

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/metaflow_extensions/outerbounds/plugins/apps/core/config/unified_config.py

@@ -301,6 +301,29 @@ class AuthConfig(metaclass=ConfigMeta):
         )
 
 
+class ScalingPolicyConfig(metaclass=ConfigMeta):
+    """
+    Policies for autoscaling replicas. Available policies:
+    - Request based Autoscaling (rpm)
+    """
+
+    # TODO Change the defaulting if we have more autoscaling policies.
+    rpm = ConfigField(
+        field_type=int,
+        # TODO: Add a little more to the docstring where we explain the behavior.
+        cli_meta=CLIOption(
+            name="scaling_rpm",
+            cli_option_str="--scaling-rpm",
+            help=(
+                "Scale up replicas when the requests per minute crosses this threshold. "
+                "If nothing is provided and the replicas.max and replicas.min is set then  "
+                "the default rpm would be 60."
+            ),
+        ),
+        default=60,
+    )
+
+
 class ReplicaConfig(metaclass=ConfigMeta):
     """Replica configuration."""
 
@@ -333,6 +356,16 @@ class ReplicaConfig(metaclass=ConfigMeta):
         example=10,
     )
 
+    scaling_policy = ConfigField(
+        cli_meta=None,
+        field_type=ScalingPolicyConfig,
+        help=(
+            "Scaling policy defines the the metric based on which the replicas will horizontally scale. "
+            "If min and max replicas are set and are not the same, then a scaling policy will be applied. "
+            "Default scaling policies can be 60 rpm (ie 1 rps). "
+        ),
+    )
+
     @staticmethod
     def defaults(replica_config: "ReplicaConfig"):
         if all(
@@ -346,6 +379,7 @@ class ReplicaConfig(metaclass=ConfigMeta):
             replica_config.fixed = 1
         elif replica_config.min is not None and replica_config.max is None:
             replica_config.max = replica_config.min
+
         return
 
     @staticmethod

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/metaflow_extensions/outerbounds/plugins/apps/core/config_schema.yaml

@@ -161,6 +161,21 @@ properties:
         type: integer
         example: 10
         mutation_behavior: union
+      scaling_policy:
+        title: ScalingPolicyConfig
+        description: |-
+          Policies for autoscaling replicas. Available policies:
+          - Request based Autoscaling (rpm)
+        type: object
+        required: []
+        properties:
+          rpm:
+            description: |-
+              Scale up replicas when the requests per minute crosses this threshold. If nothing is provided and the replicas.max and replicas.min is set then  the default rpm would be 60.
+            type: integer
+            default: 60
+            mutation_behavior: union
+        mutation_behavior: union
     mutation_behavior: union
   dependencies:
     title: DependencyConfig

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/metaflow_extensions/outerbounds/plugins/apps/core/deployer.py

@@ -151,7 +151,6 @@ class AppDeployer(TypedCoreConfig):
             final_status["id"],
             final_status["auth_type"],
             final_status["public_url"],
-            final_status["available_replicas"],
             final_status["name"],
             final_status["deployed_version"],
             final_status["deployed_at"],
@@ -164,7 +163,6 @@ class DeployedApp:
         _id: str,
         capsule_type: str,
         public_url: str,
-        available_replicas: int,
         name: str,
         deployed_version: str,
         deployed_at: str,
@@ -172,7 +170,6 @@ class DeployedApp:
         self._id = _id
         self._capsule_type = capsule_type
         self._public_url = public_url
-        self._available_replicas = available_replicas
         self._name = name
         self._deployed_version = deployed_version
         self._deployed_at = deployed_at
@@ -208,6 +205,10 @@ class DeployedApp:
         capsule = capsule_api.get(self._id)
         return capsule
 
+    def replicas(self):
+        capsule_api = self._get_capsule_api()
+        return capsule_api.get_workers(self._id)
+
     def scale_to_zero(self):
         """
         Scales the DeployedApp to 0 replicas.
@@ -243,10 +244,6 @@ class DeployedApp:
     def public_url(self) -> str:
         return self._public_url
 
-    @property
-    def available_replicas(self) -> int:
-        return self._available_replicas
-
     @property
     def name(self) -> str:
         return self._name
@@ -260,7 +257,6 @@ class DeployedApp:
             "id": self._id,
             "auth_style": self.auth_style,  # TODO : Fix naming here.
             "public_url": self._public_url,
-            "available_replicas": self._available_replicas,
             "name": self._name,
             "deployed_version": self._deployed_version,
             "deployed_at": self._deployed_at,
@@ -272,7 +268,6 @@ class DeployedApp:
             _id=data["id"],
             capsule_type=data["capsule_type"],
             public_url=data["public_url"],
-            available_replicas=data["available_replicas"],
             name=data["name"],
             deployed_version=data["deployed_version"],
             deployed_at=data["deployed_at"],
@@ -287,7 +282,6 @@ class DeployedApp:
             f"DeployedApp(id='{self._id}', "
             f"name='{self._name}', "
             f"public_url='{self._public_url}', "
-            f"available_replicas={self._available_replicas}, "
             f"deployed_version='{self._deployed_version}')"
         )
 

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/ob_metaflow_extensions.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ob-metaflow-extensions
-Version: 1.3.1
+Version: 1.3.2
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial

{ob_metaflow_extensions-1.3.1 → ob_metaflow_extensions-1.3.2}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_namespace_packages
 from pathlib import Path
 
 
-version = "1.3.1"
+version = "1.3.2"
 this_directory = Path(__file__).parent
 long_description = (this_directory / "README.md").read_text()