ob-metaflow 2.16.8.2rc1__py2.py3-none-any.whl → 2.17.0.1__py2.py3-none-any.whl

metaflow/plugins/aws/step_functions/step_functions_deployer_objects.py

@@ -56,6 +56,20 @@ class StepFunctionsDeployedFlow(DeployedFlow):
 
     TYPE: ClassVar[Optional[str]] = "step-functions"
 
+    @classmethod
+    def list_deployed_flows(cls, flow_name: Optional[str] = None):
+        """
+        This method is not currently implemented for Step Functions.
+
+        Raises
+        ------
+        NotImplementedError
+            This method is not implemented for Step Functions.
+        """
+        raise NotImplementedError(
+            "list_deployed_flows is not implemented for StepFunctions"
+        )
+
     @classmethod
     def from_deployment(cls, identifier: str, metadata: Optional[str] = None):
         """
@@ -70,6 +84,22 @@ class StepFunctionsDeployedFlow(DeployedFlow):
             "from_deployment is not implemented for StepFunctions"
         )
 
+    @classmethod
+    def get_triggered_run(
+        cls, identifier: str, run_id: str, metadata: Optional[str] = None
+    ):
+        """
+        This method is not currently implemented for Step Functions.
+
+        Raises
+        ------
+        NotImplementedError
+            This method is not implemented for Step Functions.
+        """
+        raise NotImplementedError(
+            "get_triggered_run is not implemented for StepFunctions"
+        )
+
     @property
     def production_token(self: DeployedFlow) -> Optional[str]:
         """

metaflow/plugins/cards/card_modules/basic.py

@@ -20,15 +20,12 @@ def transform_flow_graph(step_info):
             return "split"
         elif node_type == "split-parallel" or node_type == "split-foreach":
             return "foreach"
-        elif node_type == "split-switch":
-            return "switch"
         return "unknown"  # Should never happen
 
     graph_dict = {}
     for stepname in step_info:
-        node_type = node_to_type(step_info[stepname]["type"])
-        node_info = {
-            "type": node_type,
+        graph_dict[stepname] = {
+            "type": node_to_type(step_info[stepname]["type"]),
             "box_next": step_info[stepname]["type"] not in ("linear", "join"),
             "box_ends": (
                 None
@@ -38,15 +35,6 @@ def transform_flow_graph(step_info):
             "next": step_info[stepname]["next"],
             "doc": step_info[stepname]["doc"],
         }
-
-        if node_type == "switch":
-            if "condition" in step_info[stepname]:
-                node_info["condition"] = step_info[stepname]["condition"]
-            if "switch_cases" in step_info[stepname]:
-                node_info["switch_cases"] = step_info[stepname]["switch_cases"]
-
-        graph_dict[stepname] = node_info
-
     return graph_dict
 
 
@@ -430,6 +418,7 @@ class TaskInfoComponent(MetaflowCardComponent):
             "Task Finished On": task_data_dict["finished_at"],
             # Remove Microseconds from timedelta
             "Tags": ", ".join(tags),
+            "Attempt": self._task.current_attempt,
         }
         if not self.runtime:
             task_metadata_dict["Task Duration"] = str(

metaflow/plugins/cards/card_modules/convert_to_native_type.py

@@ -146,7 +146,13 @@ class TaskToDict:
             # Python 3.13 removes the standard ``imghdr`` module. Metaflow
             # vendors a copy so we can keep using ``what`` to detect image
             # formats irrespective of the Python version.
-            from metaflow._vendor import imghdr
+            import warnings
+
+            with warnings.catch_warnings():
+                warnings.filterwarnings(
+                    "ignore", category=DeprecationWarning, module="imghdr"
+                )
+                from metaflow._vendor import imghdr
 
             resp = imghdr.what(None, h=data_object)
             # Only accept types supported on the web

metaflow/plugins/kubernetes/kubernetes_decorator.py

@@ -98,7 +98,7 @@ class KubernetesDecorator(StepDecorator):
         the scheduled node should not have GPUs.
     gpu_vendor : str, default KUBERNETES_GPU_VENDOR
         The vendor of the GPUs to be used for this step.
-    tolerations : List[str], default []
+    tolerations : List[Dict[str,str]], default []
         The default is extracted from METAFLOW_KUBERNETES_TOLERATIONS.
         Kubernetes tolerations to use when launching pod in Kubernetes.
     labels: Dict[str, str], default: METAFLOW_KUBERNETES_LABELS

metaflow/plugins/kubernetes/kubernetes_job.py

@@ -522,12 +522,10 @@ class RunningJob(object):
         # 3. If the pod object hasn't shown up yet, we set the parallelism to 0
         #    to preempt it.
         client = self._client.get()
-
         if not self.is_done:
             if self.is_running:
                 # Case 1.
                 from kubernetes.stream import stream
-
                 api_instance = client.CoreV1Api
                 try:
                     # TODO: stream opens a web-socket connection. It may
@@ -593,6 +591,10 @@ class RunningJob(object):
             return self.id
         return "job %s" % self._name
 
+    @property
+    def is_unschedulable(self):
+        return self._job["metadata"]["annotations"].get("metaflow/job_status", "") == "Unsatisfiable_Resource_Request"
+
     @property
     def is_done(self):
         # Check if the container is done. As a side effect, also refreshes self._job and
@@ -606,6 +608,7 @@ class RunningJob(object):
                 or bool(self._job["status"].get("failed"))
                 or self._are_pod_containers_done
                 or (self._job["spec"]["parallelism"] == 0)
+                or self.is_unschedulable
             )
 
         if not done():
@@ -663,6 +666,7 @@ class RunningJob(object):
             bool(self._job["status"].get("failed"))
             or self._has_any_container_failed
             or (self._job["spec"]["parallelism"] == 0)
+            or self.is_unschedulable
         )
         return retval
 
@@ -760,6 +764,8 @@ class RunningJob(object):
                 return 0, None
             # Best effort since Pod object can disappear on us at anytime
             else:
+                if self.is_unschedulable:
+                    return 1, self._job["metadata"]["annotations"].get("metaflow/job_status_reason", "")
                 if self._pod.get("status", {}).get("phase") not in (
                     "Succeeded",
                     "Failed",

metaflow/plugins/kubernetes/kubernetes_jobsets.py

@@ -6,6 +6,7 @@ from collections import namedtuple
 from metaflow.exception import MetaflowException
 from metaflow.metaflow_config import KUBERNETES_JOBSET_GROUP, KUBERNETES_JOBSET_VERSION
 from metaflow.tracing import inject_tracing_vars
+from metaflow._vendor import yaml
 
 from .kube_utils import qos_requests_and_limits
 
@@ -1025,34 +1026,32 @@ class KubernetesArgoJobSet(object):
 
     def dump(self):
         client = self._kubernetes_sdk
-
-        data = json.dumps(
-            client.ApiClient().sanitize_for_serialization(
-                dict(
-                    apiVersion=self._group + "/" + self._version,
-                    kind="JobSet",
-                    metadata=client.api_client.ApiClient().sanitize_for_serialization(
-                        client.V1ObjectMeta(
-                            name=self.name,
-                            labels=self._labels,
-                            annotations=self._annotations,
-                        )
-                    ),
-                    spec=dict(
-                        replicatedJobs=[self.control.dump(), self.worker.dump()],
-                        suspend=False,
-                        startupPolicy=None,
-                        successPolicy=None,
-                        # The Failure Policy helps setting the number of retries for the jobset.
-                        # but we don't rely on it and instead rely on either the local scheduler
-                        # or the Argo Workflows to handle retries.
-                        failurePolicy=None,
-                        network=None,
-                    ),
-                    status=None,
-                )
+        js_dict = client.ApiClient().sanitize_for_serialization(
+            dict(
+                apiVersion=self._group + "/" + self._version,
+                kind="JobSet",
+                metadata=client.api_client.ApiClient().sanitize_for_serialization(
+                    client.V1ObjectMeta(
+                        name=self.name,
+                        labels=self._labels,
+                        annotations=self._annotations,
+                    )
+                ),
+                spec=dict(
+                    replicatedJobs=[self.control.dump(), self.worker.dump()],
+                    suspend=False,
+                    startupPolicy=None,
+                    successPolicy=None,
+                    # The Failure Policy helps setting the number of retries for the jobset.
+                    # but we don't rely on it and instead rely on either the local scheduler
+                    # or the Argo Workflows to handle retries.
+                    failurePolicy=None,
+                    network=None,
+                ),
+                status=None,
             )
         )
+        data = yaml.dump(js_dict, default_flow_style=False, indent=2)
         # The values we populate in the Jobset manifest (for Argo Workflows) piggybacks on the Argo Workflow's templating engine.
         # Even though Argo Workflows's templating helps us constructing all the necessary IDs and populating the fields
         # required by Metaflow, we run into one glitch. When we construct JSON/YAML serializable objects,
@@ -1067,7 +1066,6 @@ class KubernetesArgoJobSet(object):
         # Since the value of `num_parallel` can be dynamic and can change from run to run, we need to ensure that the
         # value can be passed-down dynamically and is **explicitly set as a integer** in the Jobset Manifest submitted as a
         # part of the Argo Workflow
-
-        quoted_substring = '"{{=asInt(inputs.parameters.workerCount)}}"'
+        quoted_substring = "'{{=asInt(inputs.parameters.workerCount)}}'"
         unquoted_substring = "{{=asInt(inputs.parameters.workerCount)}}"
         return data.replace(quoted_substring, unquoted_substring)

metaflow/plugins/pypi/conda_decorator.py

@@ -243,9 +243,11 @@ class CondaStepDecorator(StepDecorator):
         # Ensure local installation of Metaflow is visible to user code
         python_path = self.__class__._metaflow_home.name
         addl_env_vars = {}
-        if self.__class__._addl_env_vars is not None:
+        if self.__class__._addl_env_vars:
             for key, value in self.__class__._addl_env_vars.items():
-                if key == "PYTHONPATH":
+                if key.endswith(":"):
+                    addl_env_vars[key[:-1]] = value
+                elif key == "PYTHONPATH":
                     addl_env_vars[key] = os.pathsep.join([value, python_path])
                 else:
                     addl_env_vars[key] = value

metaflow/runner/click_api.py

@@ -43,6 +43,7 @@ from metaflow._vendor.click.types import (
 )
 from metaflow.decorators import add_decorator_options
 from metaflow.exception import MetaflowException
+from metaflow.flowspec import _FlowState
 from metaflow.includefile import FilePathClass
 from metaflow.metaflow_config import CLICK_API_PROCESS_CONFIG
 from metaflow.parameters import JSONTypeClass, flow_context
@@ -171,7 +172,6 @@ def _lazy_load_command(
     _self,
     name: str,
 ):
-
     # Context is not used in get_command so we can pass None. Since we pin click,
     # this won't change from under us.
 
@@ -516,6 +516,11 @@ class MetaflowAPI(object):
         # Note that if CLICK_API_PROCESS_CONFIG is False, we still do this because
         # it will init all parameters (config_options will be None)
         # We ignore any errors if we don't check the configs in the click API.
+
+        # Init all values in the flow mutators and then process them
+        for decorator in self._flow_cls._flow_state.get(_FlowState.FLOW_MUTATORS, []):
+            decorator.external_init()
+
         new_cls = self._flow_cls._process_config_decorators(
             config_options, process_configs=CLICK_API_PROCESS_CONFIG
         )
@@ -541,14 +546,16 @@ def extract_all_params(cmd_obj: Union[click.Command, click.Group]):
 
     for each_param in cmd_obj.params:
         if isinstance(each_param, click.Argument):
-            arg_params_sigs[each_param.name], annotations[each_param.name] = (
-                get_inspect_param_obj(each_param, inspect.Parameter.POSITIONAL_ONLY)
-            )
+            (
+                arg_params_sigs[each_param.name],
+                annotations[each_param.name],
+            ) = get_inspect_param_obj(each_param, inspect.Parameter.POSITIONAL_ONLY)
             arg_parameters[each_param.name] = each_param
         elif isinstance(each_param, click.Option):
-            opt_params_sigs[each_param.name], annotations[each_param.name] = (
-                get_inspect_param_obj(each_param, inspect.Parameter.KEYWORD_ONLY)
-            )
+            (
+                opt_params_sigs[each_param.name],
+                annotations[each_param.name],
+            ) = get_inspect_param_obj(each_param, inspect.Parameter.KEYWORD_ONLY)
             opt_parameters[each_param.name] = each_param
 
         defaults[each_param.name] = each_param.default

metaflow/runner/deployer.py

@@ -13,7 +13,9 @@ def generate_fake_flow_file_contents(
 ):
     params_code = ""
     for _, param_details in param_info.items():
-        param_python_var_name = param_details["python_var_name"]
+        param_python_var_name = param_details.get(
+            "python_var_name", param_details["name"]
+        )
         param_name = param_details["name"]
         param_type = param_details["type"]
         param_help = param_details["description"]
@@ -229,7 +231,68 @@ class DeployedFlowMeta(type):
                 }
             )
 
-            def _default_injected_method():
+            def _get_triggered_run_injected_method():
+                def f(
+                    cls,
+                    identifier: str,
+                    run_id: str,
+                    metadata: Optional[str] = None,
+                    impl: str = DEFAULT_FROM_DEPLOYMENT_IMPL.replace("-", "_"),
+                ) -> "TriggeredRun":
+                    """
+                    Retrieves a `TriggeredRun` object from an identifier, a run id and optional
+                    metadata. The `impl` parameter specifies the deployer implementation
+                    to use (like `argo-workflows`).
+
+                    Parameters
+                    ----------
+                    identifier : str
+                        Deployer specific identifier for the workflow to retrieve
+                    run_id : str
+                        Run ID for the which to fetch the triggered run object
+                    metadata : str, optional, default None
+                        Optional deployer specific metadata.
+                    impl : str, optional, default given by METAFLOW_DEFAULT_FROM_DEPLOYMENT_IMPL
+                        The default implementation to use if not specified
+
+                    Returns
+                    -------
+                    TriggeredRun
+                        A `TriggeredRun` object representing the triggered run corresponding
+                        to the identifier and the run id.
+                    """
+                    if impl in allowed_providers:
+                        return (
+                            allowed_providers[impl]
+                            .deployed_flow_type()
+                            .get_triggered_run(identifier, run_id, metadata)
+                        )
+                    else:
+                        raise ValueError(
+                            f"No deployer '{impl}' exists; valid deployers are: "
+                            f"{list(allowed_providers.keys())}"
+                        )
+
+                f.__name__ = "get_triggered_run"
+                return f
+
+            def _per_type_get_triggered_run_injected_method(method_name, impl):
+                def f(
+                    cls,
+                    identifier: str,
+                    run_id: str,
+                    metadata: Optional[str] = None,
+                ):
+                    return (
+                        allowed_providers[impl]
+                        .deployed_flow_type()
+                        .get_triggered_run(identifier, run_id, metadata)
+                    )
+
+                f.__name__ = method_name
+                return f
+
+            def _from_deployment_injected_method():
                 def f(
                     cls,
                     identifier: str,
@@ -271,7 +334,7 @@ class DeployedFlowMeta(type):
                 f.__name__ = "from_deployment"
                 return f
 
-            def _per_type_injected_method(method_name, impl):
+            def _per_type_from_deployment_injected_method(method_name, impl):
                 def f(
                     cls,
                     identifier: str,
@@ -286,14 +349,104 @@ class DeployedFlowMeta(type):
                 f.__name__ = method_name
                 return f
 
-            setattr(cls, "from_deployment", classmethod(_default_injected_method()))
+            def _list_deployed_flows_injected_method():
+                def f(
+                    cls,
+                    flow_name: Optional[str] = None,
+                    impl: str = DEFAULT_FROM_DEPLOYMENT_IMPL.replace("-", "_"),
+                ):
+                    """
+                    List all deployed flows for the specified implementation.
+
+                    Parameters
+                    ----------
+                    flow_name : str, optional, default None
+                        If specified, only list deployed flows for this specific flow name.
+                        If None, list all deployed flows.
+                    impl : str, optional, default given by METAFLOW_DEFAULT_FROM_DEPLOYMENT_IMPL
+                        The default implementation to use if not specified
+
+                    Yields
+                    ------
+                    DeployedFlow
+                        `DeployedFlow` objects representing deployed flows.
+                    """
+                    if impl in allowed_providers:
+                        return (
+                            allowed_providers[impl]
+                            .deployed_flow_type()
+                            .list_deployed_flows(flow_name)
+                        )
+                    else:
+                        raise ValueError(
+                            f"No deployer '{impl}' exists; valid deployers are: "
+                            f"{list(allowed_providers.keys())}"
+                        )
+
+                f.__name__ = "list_deployed_flows"
+                return f
+
+            def _per_type_list_deployed_flows_injected_method(method_name, impl):
+                def f(
+                    cls,
+                    flow_name: Optional[str] = None,
+                ):
+                    return (
+                        allowed_providers[impl]
+                        .deployed_flow_type()
+                        .list_deployed_flows(flow_name)
+                    )
+
+                f.__name__ = method_name
+                return f
+
+            setattr(
+                cls, "from_deployment", classmethod(_from_deployment_injected_method())
+            )
+            setattr(
+                cls,
+                "list_deployed_flows",
+                classmethod(_list_deployed_flows_injected_method()),
+            )
+            setattr(
+                cls,
+                "get_triggered_run",
+                classmethod(_get_triggered_run_injected_method()),
+            )
 
             for impl in allowed_providers:
-                method_name = f"from_{impl}"
+                from_deployment_method_name = f"from_{impl}"
+                list_deployed_flows_method_name = f"list_{impl}"
+                get_triggered_run_method_name = f"get_triggered_{impl}_run"
+
+                setattr(
+                    cls,
+                    from_deployment_method_name,
+                    classmethod(
+                        _per_type_from_deployment_injected_method(
+                            from_deployment_method_name, impl
+                        )
+                    ),
+                )
+
                 setattr(
                     cls,
-                    method_name,
-                    classmethod(_per_type_injected_method(method_name, impl)),
+                    list_deployed_flows_method_name,
+                    classmethod(
+                        _per_type_list_deployed_flows_injected_method(
+                            list_deployed_flows_method_name, impl
+                        )
+                    ),
+                )
+
+                setattr(
+                    cls,
+                    get_triggered_run_method_name,
+                    classmethod(
+                        _per_type_get_triggered_run_injected_method(
+                            get_triggered_run_method_name, impl
+                        )
+                    ),
                 )
 
         return cls

metaflow/runner/subprocess_manager.py

@@ -152,12 +152,20 @@ class SubprocessManager(object):
         int
             The process ID of the subprocess.
         """
-        updated_env = MetaflowCodeContent.get_env_vars_for_packaged_metaflow(
-            get_metaflow_root()
-        )
-        if updated_env:
-            env = env or {}
-            env.update(updated_env)
+        env = env or {}
+        installed_root = os.environ.get("METAFLOW_EXTRACTED_ROOT", get_metaflow_root())
+
+        for k, v in MetaflowCodeContent.get_env_vars_for_packaged_metaflow(
+            installed_root
+        ).items():
+            if k.endswith(":"):
+                # Override
+                env[k[:-1]] = v
+            elif k in env:
+                env[k] = "%s:%s" % (v, env[k])
+            else:
+                env[k] = v
+
         command_obj = CommandManager(command, env, cwd)
         pid = command_obj.run(show_output=show_output)
         self.commands[pid] = command_obj
@@ -188,12 +196,12 @@ class SubprocessManager(object):
         int
             The process ID of the subprocess.
         """
-        updated_env = MetaflowCodeContent.get_env_vars_for_packaged_metaflow(
-            get_metaflow_root()
-        )
-        if updated_env:
-            env = env or {}
-            env.update(updated_env)
+        env = env or {}
+        if "PYTHONPATH" in env:
+            env["PYTHONPATH"] = "%s:%s" % (get_metaflow_root(), env["PYTHONPATH"])
+        else:
+            env["PYTHONPATH"] = get_metaflow_root()
+
         command_obj = CommandManager(command, env, cwd)
         pid = await command_obj.async_run()
         self.commands[pid] = command_obj