polyaxon 2.0.6rc8__py3-none-any.whl → 2.1.0__py3-none-any.whl

polyaxon/_cli/config.py

@@ -105,7 +105,7 @@ def show():
     if _config:
         Printer.heading("User config:")
         config_user = dict_to_tabulate(
-            _config.to_dict(),
+            _config,
             humanize_values=True,
             exclude_attrs=["theme"],
         )

polyaxon/_cli/run.py

@@ -334,6 +334,12 @@ def _run(
     "If the name is not namespaced by the agent name the default agent is used: "
     "queue-name or agent-name/queue-name",
 )
+@click.option(
+    "--namespace",
+    "-ns",
+    type=str,
+    help="Namespace to use for this run. By default it will use the agent's namespace.",
+)
 @click.option(
     "--nocache",
     is_flag=True,
@@ -412,6 +418,7 @@ def run(
     matrix_num_runs,
     presets,
     queue,
+    namespace,
     nocache,
     cache,
     approved,
@@ -533,6 +540,7 @@ def run(
         matrix_num_runs=matrix_num_runs,
         presets=presets,
         queue=queue,
+        namespace=namespace,
         cache=cache,
         nocache=nocache,
         approved=approved,

polyaxon/_cli/services/clean_artifacts.py

@@ -5,7 +5,7 @@ import click
 from clipped.formatting import Printer
 from clipped.utils.lists import to_list
 
-from polyaxon._connections import V1Connection, V1ConnectionKind
+from polyaxon._connections import V1ConnectionKind
 
 
 @click.group()

polyaxon/_client/client.py

@@ -92,6 +92,7 @@ class PolyaxonClient:
         self._users_v1 = None
         self._versions_v1 = None
         self._agents_v1 = None
+        self._internal_agents_v1 = None
         self._queues_v1 = None
         self._service_accounts_v1 = None
         self._presets_v1 = None
@@ -109,6 +110,15 @@ class PolyaxonClient:
             )
         return ApiClient(self.config.sdk_config, **self.config.client_header)
 
+    def _get_internal_client(self):
+        if self.is_async:
+            return AsyncApiClient(
+                self.config.internal_sdk_config, **self.config.get_internal_header()
+            )
+        return ApiClient(
+            self.config.internal_sdk_config, **self.config.get_internal_header()
+        )
+
     def reset(self):
         self._projects_v1 = None
         self._runs_v1 = None
@@ -118,6 +128,7 @@ class PolyaxonClient:
         self._users_v1 = None
         self._versions_v1 = None
         self._agents_v1 = None
+        self._internal_agents_v1 = None
         self._queues_v1 = None
         self._service_accounts_v1 = None
         self._presets_v1 = None
@@ -169,6 +180,12 @@ class PolyaxonClient:
             self._agents_v1 = AgentsV1Api(self.api_client)
         return self._agents_v1
 
+    @property
+    def internal_agents_v1(self):
+        if not self._internal_agents_v1:
+            self._internal_agents_v1 = AgentsV1Api(self._get_internal_client())
+        return self._internal_agents_v1
+
     @property
     def queues_v1(self):
         if not self._queues_v1:

polyaxon/_client/run.py

@@ -501,6 +501,7 @@ class RunClient:
         matrix: Optional[Union[Dict, V1Matrix]] = None,
         presets: Optional[List[str]] = None,
         queue: Optional[str] = None,
+        namespace: Optional[str] = None,
         nocache: Optional[bool] = None,
         cache: Optional[Union[int, str, bool]] = None,
         approved: Optional[Union[int, str, bool]] = None,
@@ -530,6 +531,8 @@ class RunClient:
                  [presets](/docs/core/scheduling-presets/).
             queue: str, optional, the name of the
                  [queue](/docs/core/scheduling-strategies/queues/) to assign the run to.
+            namespace: str, optional, the namespace to create the run under, by default it will
+                 use the agent's namespace.
             nocache: bool, optional, DEPRECATED Please use `cache='f'`
                  simple flag to disable
                  [cache check](/docs/automation/helpers/cache/).
@@ -555,6 +558,7 @@ class RunClient:
             matrix=matrix,
             presets=presets,
             queue=queue,
+            namespace=namespace,
             nocache=nocache,
             cache=cache,
             approved=approved,
@@ -576,6 +580,7 @@ class RunClient:
         matrix: Optional[Union[Dict, V1Matrix]] = None,
         presets: Optional[List[str]] = None,
         queue: Optional[str] = None,
+        namespace: Optional[str] = None,
         nocache: Optional[bool] = None,
         cache: Optional[Union[int, str, bool]] = None,
         approved: Optional[Union[int, str, bool]] = None,
@@ -604,6 +609,8 @@ class RunClient:
                  [presets](/docs/core/scheduling-presets/).
             queue: str, optional, the name of the
                  [queue](/docs/core/scheduling-strategies/queues/) to assign the run to.
+            namespace: str, optional, the namespace to create the run under, by default it will
+                 use the agent's namespace.
             nocache: bool, optional, DEPRECATED Please use `cache='f'`
                  simple flag to disable
                  [cache check](/docs/automation/helpers/cache/).
@@ -629,6 +636,7 @@ class RunClient:
             matrix=matrix,
             presets=presets,
             queue=queue,
+            namespace=namespace,
             nocache=nocache,
             cache=cache,
             approved=approved,
@@ -650,6 +658,7 @@ class RunClient:
         matrix: Optional[Union[Dict, V1Matrix]] = None,
         presets: Optional[List[str]] = None,
         queue: Optional[str] = None,
+        namespace: Optional[str] = None,
         nocache: Optional[bool] = None,
         cache: Optional[Union[int, str, bool]] = None,
         approved: Optional[Union[int, str, bool]] = None,
@@ -677,6 +686,8 @@ class RunClient:
                  [presets](/docs/core/scheduling-presets/).
             queue: str, optional, the name of the
                  [queue](/docs/core/scheduling-strategies/queues/) to assign the run to.
+            namespace: str, optional, the namespace to create the run under, by default it will
+                 use the agent's namespace.
             nocache: bool, optional, DEPRECATED Please use `cache='f'`
                  simple flag to disable
                  [cache check](/docs/automation/helpers/cache/).
@@ -702,6 +713,7 @@ class RunClient:
             matrix=matrix,
             presets=presets,
             queue=queue,
+            namespace=namespace,
             nocache=nocache,
             cache=cache,
             approved=approved,

polyaxon/_compiler/resolver/agent.py

@@ -49,7 +49,7 @@ class AgentResolver(BaseSchemaModel):
 
         self.polyaxon_sidecar = agent_config.sidecar or get_default_sidecar_container()
         self.polyaxon_init = agent_config.init or get_default_init_container()
-        self.namespace = agent_config.namespace
+        self.namespace = compiled_operation.namespace or agent_config.namespace
 
     def _resolve_run_connections(
         self, compiled_operation: V1CompiledOperation, agent_config: AgentConfig

polyaxon/_compiler/resolver/runtime.py

@@ -181,7 +181,7 @@ class BaseResolver:
         )
         self.polyaxon_sidecar = agent_env.polyaxon_sidecar
         self.polyaxon_init = agent_env.polyaxon_init
-        self.namespace = agent_env.namespace
+        self.namespace = self.compiled_operation.namespace or agent_env.namespace
         self.secrets = agent_env.secrets
         self.config_maps = agent_env.config_maps
         self.connection_by_names = agent_env.connection_by_names

polyaxon/_deploy/schemas/service.py

@@ -12,6 +12,7 @@ class DeploymentService(BaseServiceConfig):
     enabled: Optional[bool]
     replicas: Optional[StrictInt]
     concurrency: Optional[StrictInt]
+    per_core: Optional[bool] = Field(alias="perCore")
     scheme: Optional[StrictStr]
 
 
@@ -22,6 +23,8 @@ class WorkerServiceConfig(DeploymentService):
 class AgentServiceConfig(DeploymentService):
     instance: Optional[StrictStr]
     token: Optional[StrictStr]
+    watch_cluster: Optional[bool] = Field(alias="watchCluster")
+    additional_namespaces: Optional[List[str]] = Field(alias="additionalNamespaces")
     is_replica: Optional[bool] = Field(alias="isReplica")
 
 
@@ -50,6 +53,7 @@ class HooksConfig(DeploymentService):
     tables: Optional[bool] = Field(alias="tables")
     sync_db: Optional[bool] = Field(alias="syncdb")
     admin_user: Optional[bool] = Field(alias="adminUser")
+    default_org: Optional[bool] = Field(alias="defaultOrg")
 
 
 class ThirdPartyService(DeploymentService):

polyaxon/_docker/executor.py

@@ -35,6 +35,7 @@ class Executor(BaseExecutor):
         run_uuid: str,
         run_kind: str,
         resource: List[docker_types.V1Container],
+        namespace: str = None,
     ) -> Dict:
         logger.info(f"[Executor] Starting operation {run_uuid} {run_kind}.")
         self._ops[run_uuid] = []
@@ -62,13 +63,15 @@ class Executor(BaseExecutor):
         self._clean_temp_execution_path(run_uuid)
         return {"status": V1Statuses.SUCCEEDED, "tasks": self._ops[run_uuid]}
 
-    def apply(self, run_uuid: str, run_kind: str, resource: Dict) -> Dict:
+    def apply(
+        self, run_uuid: str, run_kind: str, resource: Dict, namespace: str = None
+    ) -> Dict:
         raise PolyaxonAgentError(
             "Docker executor does not support apply method.\n"
             "Run: <kind: {}, uuid: {}>".format(run_kind, run_uuid)
         )
 
-    def stop(self, run_uuid: str, run_kind: str):
+    def stop(self, run_uuid: str, run_kind: str, namespace: str = None):
         proc = self._get_op_proc(run_uuid)
         if proc.poll() is None:
             # Kill the process tree rooted at the child if it's the leader of its own process
@@ -85,7 +88,7 @@ class Executor(BaseExecutor):
                 logger.debug(_msg)
             proc.wait()
 
-    def clean(self, run_uuid: str, run_kind: str):
+    def clean(self, run_uuid: str, run_kind: str, namespace: str = None):
         return self.apply(
             run_uuid=run_uuid,
             run_kind=run_kind,
@@ -100,6 +103,9 @@ class Executor(BaseExecutor):
             return V1Statuses.SUCCEEDED
         return V1Statuses.FAILED
 
-    def get(self, run_uuid: str, run_kind: str) -> V1Statuses:
+    def get(self, run_uuid: str, run_kind: str, namespace: str = None) -> V1Statuses:
         procs = self._get_op_proc(run_uuid)
         return self._get_task_status(procs[-1])
+
+    def list_ops(self, namespace: str = None):
+        return []

polyaxon/_env_vars/getters/run.py

@@ -74,7 +74,10 @@ def get_run_info(run_instance: Optional[str] = None):
             "Could not get run info, "
             "please make sure this is run is correctly started by Polyaxon."
         )
+    return get_run_info_from_instance(run_instance)
 
+
+def get_run_info_from_instance(run_instance: str):
     parts = run_instance.split(".")
     if not len(parts) == 4:
         raise PolyaxonClientException(

polyaxon/_env_vars/keys.py

@@ -78,6 +78,9 @@ ENV_KEYS_AGENT_ARTIFACTS_STORE = "POLYAXON_AGENT_ARTIFACTS_STORE"
 ENV_KEYS_AGENT_CONNECTIONS = "POLYAXON_AGENT_CONNECTIONS"
 ENV_KEYS_SET_AGENT = "POLYAXON_SET_AGENT"
 ENV_KEYS_K8S_APP_SECRET_NAME = "POLYAXON_K8S_APP_SECRET_NAME"  # noqa
+ENV_KEYS_WATCH_CLUSTER = "POLYAXON_WATCH_CLUSTER"
+ENV_KEYS_SINGLE_NAMESPACE = "POLYAXON_SINGLE_NAMESPACE"
+ENV_KEYS_ADDITIONAL_NAMESPACES = "POLYAXON_ADDITIONAL_NAMESPACES"
 ENV_KEYS_AGENT_SECRET_NAME = "POLYAXON_AGENT_SECRET_NAME"  # noqa
 ENV_KEYS_AGENT_RUNS_SA = "POLYAXON_AGENT_RUNS_SA"
 ENV_KEYS_AGENT_ENABLE_HEALTH_CHECKS = "POLYAXON_AGENT_ENABLE_HEALTH_CHECKS"
@@ -120,6 +123,8 @@ ENV_KEYS_PROXY_NAMESPACES = "POLYAXON_PROXY_NAMESPACES"
 ENV_KEYS_PROXY_GATEWAY_PORT = "POLYAXON_PROXY_GATEWAY_PORT"
 ENV_KEYS_PROXY_GATEWAY_TARGET_PORT = "POLYAXON_PROXY_GATEWAY_TARGET_PORT"
 ENV_KEYS_PROXY_GATEWAY_HOST = "POLYAXON_PROXY_GATEWAY_HOST"
+ENV_KEYS_PROXY_GATEWAY_CONCURRENCY = "POLYAXON_PROXY_GATEWAY_CONCURRENCY"
+ENV_KEYS_PROXY_GATEWAY_PER_CORE = "POLYAXON_PROXY_GATEWAY_PER_CORE"
 ENV_KEYS_PROXY_STREAMS_PORT = "POLYAXON_PROXY_STREAMS_PORT"
 ENV_KEYS_PROXY_STREAMS_TARGET_PORT = "POLYAXON_PROXY_STREAMS_TARGET_PORT"
 ENV_KEYS_PROXY_STREAMS_HOST = "POLYAXON_PROXY_STREAMS_HOST"

polyaxon/_flow/__init__.py

@@ -75,8 +75,10 @@ from polyaxon._flow.run import (
     V1MPIJob,
     V1MXJob,
     V1NotifierJob,
+    V1PaddleElasticPolicy,
     V1PaddleJob,
     V1PipelineKind,
+    V1PytorchElasticPolicy,
     V1PytorchJob,
     V1RayJob,
     V1RayReplica,

polyaxon/_flow/builds/__init__.py

@@ -28,6 +28,7 @@ class V1Build(BaseSchemaModel):
         hub_ref: str
         connection: str
         queue: str, optional
+        namespace: str, optional
         presets: List[str], optional
         cache: [V1Cache](/docs/automation/helpers/cache/), optional
         params: Dict[str, [V1Param](/docs/core/specification/params/)], optional
@@ -94,6 +95,17 @@ class V1Build(BaseSchemaModel):
     ...
     ```
 
+    ### presets
+
+    The [presets](/docs/management/organizations/presets/) to use for the hook operation,
+    if provided, it will override the component's presets otherwise
+    the presets of the component will be used if available.
+
+    ```yaml
+    >>> build:
+    >>>   presets: [test]
+    ```
+
     ### queue
 
     The [queue](/docs/core/scheduling-strategies/queues/) to use.
@@ -112,15 +124,15 @@ class V1Build(BaseSchemaModel):
     >>>   queue: queue-name
     ```
 
-    ### presets
+    ### namespace
 
-    The [presets](/docs/management/organizations/presets/) to use for the hook operation,
-    if provided, it will override the component's presets otherwise
-    the presets of the component will be used if available.
+    > **Note**: Please note that this field is only available in some commercial editions.
+
+    The namespace to use, if not provided, it will default to the agent's namespace.
 
     ```yaml
     >>> build:
-    >>>   presets: [test]
+    >>>   namespace: polyaxon
     ```
 
     ### cache
@@ -211,8 +223,9 @@ class V1Build(BaseSchemaModel):
     _IDENTIFIER = "build"
     hub_ref: StrictStr = Field(alias="hubRef")
     connection: Optional[StrictStr]
-    queue: Optional[StrictStr]
     presets: Optional[Union[List[StrictStr], RefField]]
+    queue: Optional[StrictStr]
+    namespace: Optional[StrictStr]
     cache: Optional[Union[V1Cache, RefField]]
     params: Optional[Dict[str, Union[V1Param, RefField]]]
     run_patch: Optional[Dict[str, Any]] = Field(alias="runPatch")

polyaxon/_flow/component/base.py

@@ -20,6 +20,7 @@ class BaseComponent(BaseSchemaModel):
     tags: Optional[List[StrictStr]]
     presets: Optional[List[StrictStr]]
     queue: Optional[StrictStr]
+    namespace: Optional[StrictStr]
     cache: Optional[Union[V1Cache, RefField]]
     termination: Optional[Union[V1Termination, RefField]]
     plugins: Optional[Union[V1Plugins, RefField]]

polyaxon/_flow/component/component.py

@@ -39,6 +39,7 @@ class V1Component(
         tags: List[str], optional
         presets: List[str], optional
         queue: str, optional
+        namespace: str, optional
         cache: [V1Cache](/docs/automation/helpers/cache/), optional
         termination: [V1Termination](/docs/core/specification/termination/), optional
         plugins: [V1Plugins](/docs/core/specification/plugins/), optional
@@ -60,6 +61,7 @@ class V1Component(
     >>>   tags:
     >>>   presets:
     >>>   queue:
+    >>>   namespace:
     >>>   cache:
     >>>   termination:
     >>>   plugins:
@@ -85,6 +87,7 @@ class V1Component(
     >>>     tags=["test"],
     >>>     presets=["test"],
     >>>     queue="test",
+    >>>     namespace="test",
     >>>     cache=V1Cache(...),
     >>>     termination=V1Termination(...),
     >>>     plugins=V1Plugins(...),
@@ -197,6 +200,17 @@ class V1Component(
     >>>   queue: queue-name
     ```
 
+    ### namespace
+
+    > **Note**: Please note that this field is only available in some commercial editions.
+
+    The namespace to use, if not provided, it will default to the agent's namespace.
+
+    ```yaml
+    >>> component:
+    >>>   namespace: polyaxon
+    ```
+
     ### cache
 
     The default component [cache](/docs/automation/helpers/cache/).

polyaxon/_flow/environment/__init__.py

@@ -207,14 +207,14 @@ class V1Environment(BaseSchemaModel):
     ```yaml
     >>> rules:
     >>>   - apiGroups: [""]
-    >>>     resources: ["pods"]
+    >>>     resources: ["pods", "services", "events", "pods/status", "pods/log"]
     >>>     verbs: ["get", "watch", "list"]
     >>>   - apiGroups: ["metrics.k8s.io"]
-    >>>     resources: ["pods", "nodes", "apis"]
+    >>>     resources: ["pods", "nodes"]
     >>>     verbs: ["get", "list", "watch"]
-    >>>   - apiGroups: ["", "*"]
-    >>>     resources: ["events", "pods/status", "pods/log"]
-    >>>     verbs: ["watch", "get", "list"]
+    >>>   - apiGroups: ["core.polyaxon.com"]
+    >>>     resources: ["operations"]
+    >>>     verbs: ["get", "watch", "list"]
     ```
 
     ### hostAliases

polyaxon/_flow/hooks/__init__.py

@@ -49,6 +49,7 @@ class V1Hook(BaseSchemaModel):
         hub_ref: str
         conditions: str, optional
         queue: str, optional
+        namespace: str, optional
         presets: List[str], optional
         disableDefaults: bool, optional
         params: Dict[str, [V1Param](/docs/core/specification/params/)], optional
@@ -131,6 +132,17 @@ class V1Hook(BaseSchemaModel):
     In the example above, the hook will only run if a param is passed, or an output is logged and
     is equal to "some-value".
 
+    ### presets
+
+    The [presets](/docs/management/organizations/presets/) to use for the hook operation,
+    if provided, it will override the component's presets otherwise
+    the presets of the component will be used if available.
+
+    ```yaml
+    >>> hook:
+    >>>   presets: [test]
+    ```
+
     ### queue
 
     The [queue](/docs/core/scheduling-strategies/queues/) to use.
@@ -149,15 +161,15 @@ class V1Hook(BaseSchemaModel):
     >>>   queue: queue-name
     ```
 
-    ### presets
+    ### namespace
 
-    The [presets](/docs/management/organizations/presets/) to use for the hook operation,
-    if provided, it will override the component's presets otherwise
-    the presets of the component will be used if available.
+    > **Note**: Please note that this field is only available in some commercial editions.
+
+    The namespace to use, if not provided, it will default to the agent's namespace.
 
     ```yaml
     >>> hook:
-    >>>   presets: [test]
+    >>>   namespace: polyaxon
     ```
 
     ### disableDefaults
@@ -199,8 +211,9 @@ class V1Hook(BaseSchemaModel):
     connection: Optional[StrictStr]
     trigger: Optional[V1Statuses]
     conditions: Optional[StrictStr]
-    queue: Optional[StrictStr]
     presets: Optional[List[StrictStr]]
+    queue: Optional[StrictStr]
+    namespace: Optional[StrictStr]
     params: Optional[Union[Dict[str, V1Param], RefField]]
     disable_defaults: Optional[BoolOrRef] = Field(alias="disableDefaults")
 

polyaxon/_flow/matrix/tuner.py

@@ -60,6 +60,17 @@ class V1Tuner(BaseSchemaModel):
     ...
     ```
 
+    ### presets
+
+    The [presets](/docs/management/organizations/presets/) to use for the tuner operation,
+    if provided, it will override the component's presets otherwise
+    the presets of the component will be used if available.
+
+    ```yaml
+    >>> tuner:
+    >>>   presets: [test]
+    ```
+
     ### queue
 
     The [queue](/docs/core/scheduling-strategies/queues/) to use.
@@ -78,15 +89,15 @@ class V1Tuner(BaseSchemaModel):
     >>>   queue: queue-name
     ```
 
-    ### presets
+    ### namespace
 
-    The [presets](/docs/management/organizations/presets/) to use for the tuner operation,
-    if provided, it will override the component's presets otherwise
-    the presets of the component will be used if available.
+    > **Note**: Please note that this field is only available in some commercial editions.
+
+    The namespace to use, if not provided, it will default to the agent's namespace.
 
     ```yaml
     >>> tuner:
-    >>>   presets: [test]
+    >>>   namespace: polyaxon
     ```
 
     ### params
@@ -108,6 +119,7 @@ class V1Tuner(BaseSchemaModel):
     _IDENTIFIER = "tuner"
 
     hub_ref: StrictStr = Field(alias="hubRef")
-    queue: Optional[StrictStr]
     presets: Optional[Union[List[StrictStr], RefField]]
+    queue: Optional[StrictStr]
+    namespace: Optional[StrictStr]
     params: Optional[Union[Dict[str, V1Param], RefField]]

polyaxon/_flow/operations/operation.py

@@ -45,6 +45,7 @@ class V1Operation(BaseOp, TemplateMixinConfig):
         tags: List[str], optional
         presets: str, optional
         queue: str, optional
+        namespace: str, optional
         cache: [V1Cache](/docs/automation/helpers/cache/), optional
         termination: [V1Termination](/docs/core/specification/termination/), optional
         plugins: [V1Plugins](/docs/core/specification/plugins/), optional
@@ -81,6 +82,7 @@ class V1Operation(BaseOp, TemplateMixinConfig):
     >>>   tags:
     >>>   presets:
     >>>   queue:
+    >>>   namespace:
     >>>   cache:
     >>>   termination:
     >>>   plugins:
@@ -111,6 +113,7 @@ class V1Operation(BaseOp, TemplateMixinConfig):
     >>>     tags=["test"],
     >>>     presets=["test"],
     >>>     queue="test",
+    >>>     namespace="test",
     >>>     cache=V1Cache(...),
     >>>     termination=V1Termination(...),
     >>>     plugins=V1Plugins(...),
@@ -248,6 +251,20 @@ class V1Operation(BaseOp, TemplateMixinConfig):
     >>>   queue: queue-name
     ```
 
+    ### namespace
+
+    > **Note**: Please note that this field is only available in some commercial editions.
+
+    The namespace to use for this operation run,
+    if provided, it will override the component's namespace otherwise
+    the namesace of the component will be used if it exists or
+    it will default to the agent's namespace.
+
+    ```yaml
+    >>> operation:
+    >>>   namespace: polyaxon
+    ```
+
     ### cache
 
     The [cache](/docs/automation/helpers/cache/) to use for this operation run,
@@ -650,6 +667,7 @@ class V1Operation(BaseOp, TemplateMixinConfig):
             hub_ref=hook.hub_ref,
             presets=hook.presets,
             queue=hook.queue,
+            namespace=hook.namespace,
             params=params,
         )
 
@@ -674,6 +692,7 @@ class V1Operation(BaseOp, TemplateMixinConfig):
             hub_ref=build.hub_ref,
             presets=build.presets,
             queue=build.queue,
+            namespace=build.namespace,
             cache=build.cache,
             params=params,
         )

polyaxon/_flow/run/__init__.py

@@ -17,8 +17,8 @@ from polyaxon._flow.run.job import V1Job
 from polyaxon._flow.run.kubeflow.clean_pod_policy import V1CleanPodPolicy
 from polyaxon._flow.run.kubeflow.mpi_job import V1MPIJob
 from polyaxon._flow.run.kubeflow.mx_job import MXJobMode, V1MXJob
-from polyaxon._flow.run.kubeflow.paddle_job import V1PaddleJob
-from polyaxon._flow.run.kubeflow.pytorch_job import V1PytorchJob
+from polyaxon._flow.run.kubeflow.paddle_job import V1PaddleElasticPolicy, V1PaddleJob
+from polyaxon._flow.run.kubeflow.pytorch_job import V1PytorchElasticPolicy, V1PytorchJob
 from polyaxon._flow.run.kubeflow.replica import V1KFReplica
 from polyaxon._flow.run.kubeflow.scheduling_policy import V1SchedulingPolicy
 from polyaxon._flow.run.kubeflow.tf_job import V1TFJob

polyaxon/_flow/run/kubeflow/paddle_job.py

@@ -1,8 +1,8 @@
-from typing import Optional, Union
+from typing import Dict, List, Optional, Union
 from typing_extensions import Literal
 
 from clipped.compact.pydantic import Field
-from clipped.types.ref_or_obj import RefField
+from clipped.types.ref_or_obj import IntOrRef, RefField
 
 from polyaxon._flow.run.base import BaseRun
 from polyaxon._flow.run.enums import V1RunKind
@@ -12,6 +12,25 @@ from polyaxon._flow.run.kubeflow.scheduling_policy import V1SchedulingPolicy
 from polyaxon._flow.run.resources import V1RunResources
 from polyaxon._flow.run.utils import DestinationImageMixin
 from polyaxon._k8s.k8s_schemas import V1Container
+from polyaxon._schemas.base import BaseSchemaModel
+
+
+class V1PaddleElasticPolicy(BaseSchemaModel):
+    """Elastic policy for Paddle distributed runs.
+
+    Args:
+        minReplicas: int, optional
+        maxReplicas: int, optional
+        maxRestarts: int, optional
+        metrics: List[Dict], optional
+    """
+
+    _IDENTIFIER = "elasticPolicy"
+
+    min_replicas: Optional[IntOrRef] = Field(alias="minReplicas")
+    max_replicas: Optional[IntOrRef] = Field(alias="maxReplicas")
+    max_restarts: Optional[IntOrRef] = Field(alias="maxRestarts")
+    metrics: Optional[List[Dict]] = Field(alias="Metrics")
 
 
 class V1PaddleJob(BaseRun, DestinationImageMixin):
@@ -88,6 +107,18 @@ class V1PaddleJob(BaseRun, DestinationImageMixin):
     >>>  ...
     ```
 
+    ### elasticPolicy
+
+    Elastic policy for Paddle distributed runs.
+
+    ```yaml
+    >>> run:
+    >>>   kind: paddlejob
+    >>>   elasticPolicy:
+    >>>     ...
+    >>>  ...
+    ```
+
     ### master
 
     The ,aster is responsible for orchestrating training and performing
@@ -124,6 +155,7 @@ class V1PaddleJob(BaseRun, DestinationImageMixin):
     kind: Literal[_IDENTIFIER] = _IDENTIFIER
     clean_pod_policy: Optional[V1CleanPodPolicy] = Field(alias="cleanPodPolicy")
     scheduling_policy: Optional[V1SchedulingPolicy] = Field(alias="schedulingPolicy")
+    elastic_policy: Optional[V1PaddleElasticPolicy] = Field(alias="elasticPolicy")
     master: Optional[Union[V1KFReplica, RefField]]
     worker: Optional[Union[V1KFReplica, RefField]]