ob-metaflow 2.13.0.1__py2.py3-none-any.whl → 2.13.1.1__py2.py3-none-any.whl

metaflow/plugins/kubernetes/kube_utils.py

@@ -1,7 +1,14 @@
-from metaflow.exception import CommandException
+import re
+from typing import Dict, List, Optional
+from metaflow.exception import CommandException, MetaflowException
 from metaflow.util import get_username, get_latest_run_id
 
 
+# avoid circular import by having the exception class contained here
+class KubernetesException(MetaflowException):
+    headline = "Kubernetes error"
+
+
 def parse_cli_options(flow_name, run_id, user, my_runs, echo):
     if user and my_runs:
         raise CommandException("--user and --my-runs are mutually exclusive.")
@@ -52,3 +59,50 @@ def qos_requests_and_limits(qos: str, cpu: int, memory: int, storage: int):
     # TODO: Add support for BestEffort once there is a use case for it.
     # BestEffort - no limit or requests for cpu/memory
     return qos_requests, qos_limits
+
+
+def validate_kube_labels(
+    labels: Optional[Dict[str, Optional[str]]],
+) -> bool:
+    """Validate label values.
+
+    This validates the kubernetes label values.  It does not validate the keys.
+    Ideally, keys should be static and also the validation rules for keys are
+    more complex than those for values.  For full validation rules, see:
+
+    https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#syntax-and-character-set
+    """
+
+    def validate_label(s: Optional[str]):
+        regex_match = r"^(([A-Za-z0-9][-A-Za-z0-9_.]{0,61})?[A-Za-z0-9])?$"
+        if not s:
+            # allow empty label
+            return True
+        if not re.search(regex_match, s):
+            raise KubernetesException(
+                'Invalid value: "%s"\n'
+                "A valid label must be an empty string or one that\n"
+                "  - Consist of alphanumeric, '-', '_' or '.' characters\n"
+                "  - Begins and ends with an alphanumeric character\n"
+                "  - Is at most 63 characters" % s
+            )
+        return True
+
+    return all([validate_label(v) for v in labels.values()]) if labels else True
+
+
+def parse_kube_keyvalue_list(items: List[str], requires_both: bool = True):
+    try:
+        ret = {}
+        for item_str in items:
+            item = item_str.split("=", 1)
+            if requires_both:
+                item[1]  # raise IndexError
+            if str(item[0]) in ret:
+                raise KubernetesException("Duplicate key found: %s" % str(item[0]))
+            ret[str(item[0])] = str(item[1]) if len(item) > 1 else None
+        return ret
+    except KubernetesException as e:
+        raise e
+    except (AttributeError, IndexError):
+        raise KubernetesException("Unable to parse kubernetes list: %s" % items)

metaflow/plugins/kubernetes/kubernetes.py

@@ -1,11 +1,8 @@
-import copy
 import json
 import math
 import os
-import re
 import shlex
 import time
-from typing import Dict, List, Optional
 from uuid import uuid4
 
 from metaflow import current, util
@@ -35,7 +32,6 @@ from metaflow.metaflow_config import (
     DEFAULT_SECRETS_BACKEND_TYPE,
     GCP_SECRET_MANAGER_PREFIX,
     KUBERNETES_FETCH_EC2_METADATA,
-    KUBERNETES_LABELS,
     KUBERNETES_SANDBOX_INIT_SCRIPT,
     OTEL_ENDPOINT,
     S3_ENDPOINT_URL,
@@ -193,6 +189,7 @@ class Kubernetes(object):
         persistent_volume_claims=None,
         tolerations=None,
         labels=None,
+        annotations=None,
         shared_memory=None,
         port=None,
         num_parallel=None,
@@ -304,10 +301,6 @@ class Kubernetes(object):
             # see get_datastore_root_from_config in datastore/local.py).
         )
 
-        _labels = self._get_labels(labels)
-        for k, v in _labels.items():
-            jobset.label(k, v)
-
         for k in list(
             [] if not secrets else [secrets] if isinstance(secrets, str) else secrets
         ) + KUBERNETES_SECRETS.split(","):
@@ -395,13 +388,16 @@ class Kubernetes(object):
         for name, value in env.items():
             jobset.environment_variable(name, value)
 
-        annotations = {
+        system_annotations = {
             "metaflow/user": user,
             "metaflow/flow_name": flow_name,
             "metaflow/control-task-id": task_id,
+            "metaflow/run_id": run_id,
+            "metaflow/step_name": step_name,
+            "metaflow/attempt": attempt,
         }
         if current.get("project_name"):
-            annotations.update(
+            system_annotations.update(
                 {
                     "metaflow/project_name": current.project_name,
                     "metaflow/branch_name": current.branch_name,
@@ -409,15 +405,15 @@ class Kubernetes(object):
                 }
             )
 
-        for name, value in annotations.items():
-            jobset.annotation(name, value)
+        system_labels = {
+            "app.kubernetes.io/name": "metaflow-task",
+            "app.kubernetes.io/part-of": "metaflow",
+        }
 
-        (
-            jobset.annotation("metaflow/run_id", run_id)
-            .annotation("metaflow/step_name", step_name)
-            .annotation("metaflow/attempt", attempt)
-            .label("app.kubernetes.io/name", "metaflow-task")
-            .label("app.kubernetes.io/part-of", "metaflow")
+        jobset.labels({**({} if not labels else labels), **system_labels})
+
+        jobset.annotations(
+            {**({} if not annotations else annotations), **system_annotations}
         )
         # We need this task-id set so that all the nodes are aware of the control
         # task's task-id. These "MF_" variables populate the `current.parallel` namedtuple
@@ -507,6 +503,7 @@ class Kubernetes(object):
         port=None,
         name_pattern=None,
         qos=None,
+        annotations=None,
     ):
         if env is None:
             env = {}
@@ -539,7 +536,8 @@ class Kubernetes(object):
                 retries=0,
                 step_name=step_name,
                 tolerations=tolerations,
-                labels=self._get_labels(labels),
+                labels=labels,
+                annotations=annotations,
                 use_tmpfs=use_tmpfs,
                 tmpfs_tempdir=tmpfs_tempdir,
                 tmpfs_size=tmpfs_size,
@@ -658,13 +656,25 @@ class Kubernetes(object):
 
         for name, value in env.items():
             job.environment_variable(name, value)
+        # Add job specific labels
+        system_labels = {
+            "app.kubernetes.io/name": "metaflow-task",
+            "app.kubernetes.io/part-of": "metaflow",
+        }
+        for name, value in system_labels.items():
+            job.label(name, value)
 
-        annotations = {
-            "metaflow/user": user,
+        # Add job specific annotations not set in the decorator.
+        system_annotations = {
             "metaflow/flow_name": flow_name,
+            "metaflow/run_id": run_id,
+            "metaflow/step_name": step_name,
+            "metaflow/task_id": task_id,
+            "metaflow/attempt": attempt,
+            "metaflow/user": user,
         }
         if current.get("project_name"):
-            annotations.update(
+            system_annotations.update(
                 {
                     "metaflow/project_name": current.project_name,
                     "metaflow/branch_name": current.branch_name,
@@ -672,7 +682,7 @@ class Kubernetes(object):
                 }
             )
 
-        for name, value in annotations.items():
+        for name, value in system_annotations.items():
             job.annotation(name, value)
 
         (
@@ -791,60 +801,3 @@ class Kubernetes(object):
             "stderr",
             job_id=self._job.id,
         )
-
-    @staticmethod
-    def _get_labels(extra_labels=None):
-        if extra_labels is None:
-            extra_labels = {}
-        env_labels = KUBERNETES_LABELS.split(",") if KUBERNETES_LABELS else []
-        env_labels = parse_kube_keyvalue_list(env_labels, False)
-        labels = {**env_labels, **extra_labels}
-        validate_kube_labels(labels)
-        return labels
-
-
-def validate_kube_labels(
-    labels: Optional[Dict[str, Optional[str]]],
-) -> bool:
-    """Validate label values.
-
-    This validates the kubernetes label values.  It does not validate the keys.
-    Ideally, keys should be static and also the validation rules for keys are
-    more complex than those for values.  For full validation rules, see:
-
-    https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#syntax-and-character-set
-    """
-
-    def validate_label(s: Optional[str]):
-        regex_match = r"^(([A-Za-z0-9][-A-Za-z0-9_.]{0,61})?[A-Za-z0-9])?$"
-        if not s:
-            # allow empty label
-            return True
-        if not re.search(regex_match, s):
-            raise KubernetesException(
-                'Invalid value: "%s"\n'
-                "A valid label must be an empty string or one that\n"
-                "  - Consist of alphanumeric, '-', '_' or '.' characters\n"
-                "  - Begins and ends with an alphanumeric character\n"
-                "  - Is at most 63 characters" % s
-            )
-        return True
-
-    return all([validate_label(v) for v in labels.values()]) if labels else True
-
-
-def parse_kube_keyvalue_list(items: List[str], requires_both: bool = True):
-    try:
-        ret = {}
-        for item_str in items:
-            item = item_str.split("=", 1)
-            if requires_both:
-                item[1]  # raise IndexError
-            if str(item[0]) in ret:
-                raise KubernetesException("Duplicate key found: %s" % str(item[0]))
-            ret[str(item[0])] = str(item[1]) if len(item) > 1 else None
-        return ret
-    except KubernetesException as e:
-        raise e
-    except (AttributeError, IndexError):
-        raise KubernetesException("Unable to parse kubernetes list: %s" % items)

metaflow/plugins/kubernetes/kubernetes_cli.py

@@ -3,14 +3,17 @@ import sys
 import time
 import traceback
 
-from metaflow.plugins.kubernetes.kube_utils import parse_cli_options
+from metaflow.plugins.kubernetes.kube_utils import (
+    parse_cli_options,
+    parse_kube_keyvalue_list,
+)
 from metaflow.plugins.kubernetes.kubernetes_client import KubernetesClient
 import metaflow.tracing as tracing
 from metaflow import JSONTypeClass, util
 from metaflow._vendor import click
 from metaflow.exception import METAFLOW_EXIT_DISALLOW_RETRY, MetaflowException
 from metaflow.metadata_provider.util import sync_local_metadata_from_datastore
-from metaflow.metaflow_config import DATASTORE_LOCAL_DIR, KUBERNETES_LABELS
+from metaflow.metaflow_config import DATASTORE_LOCAL_DIR
 from metaflow.mflog import TASK_LOG_SOURCE
 from metaflow.unbounded_foreach import UBF_CONTROL, UBF_TASK
 
@@ -18,9 +21,7 @@ from .kubernetes import (
     Kubernetes,
     KubernetesException,
     KubernetesKilledException,
-    parse_kube_keyvalue_list,
 )
-from .kubernetes_decorator import KubernetesDecorator
 
 
 @click.group()
@@ -132,6 +133,18 @@ def kubernetes():
     type=str,
     help="Quality of Service class for the Kubernetes pod",
 )
+@click.option(
+    "--labels",
+    default=None,
+    type=JSONTypeClass(),
+    multiple=False,
+)
+@click.option(
+    "--annotations",
+    default=None,
+    type=JSONTypeClass(),
+    multiple=False,
+)
 @click.pass_context
 def step(
     ctx,
@@ -161,6 +174,8 @@ def step(
     port=None,
     num_parallel=None,
     qos=None,
+    labels=None,
+    annotations=None,
     **kwargs
 ):
     def echo(msg, stream="stderr", job_id=None, **kwargs):
@@ -302,8 +317,10 @@ def step(
                 port=port,
                 num_parallel=num_parallel,
                 qos=qos,
+                labels=labels,
+                annotations=annotations,
             )
-    except Exception as e:
+    except Exception:
         traceback.print_exc(chain=False)
         _sync_metadata()
         sys.exit(METAFLOW_EXIT_DISALLOW_RETRY)

metaflow/plugins/kubernetes/kubernetes_decorator.py

@@ -19,6 +19,8 @@ from metaflow.metaflow_config import (
     KUBERNETES_GPU_VENDOR,
     KUBERNETES_IMAGE_PULL_POLICY,
     KUBERNETES_MEMORY,
+    KUBERNETES_LABELS,
+    KUBERNETES_ANNOTATIONS,
     KUBERNETES_NAMESPACE,
     KUBERNETES_NODE_SELECTOR,
     KUBERNETES_PERSISTENT_VOLUME_CLAIMS,
@@ -34,7 +36,8 @@ from metaflow.sidecar import Sidecar
 from metaflow.unbounded_foreach import UBF_CONTROL
 
 from ..aws.aws_utils import get_docker_registry, get_ec2_instance_metadata
-from .kubernetes import KubernetesException, parse_kube_keyvalue_list
+from .kubernetes import KubernetesException
+from .kube_utils import validate_kube_labels, parse_kube_keyvalue_list
 
 from metaflow.metaflow_config import MAX_MEMORY_PER_TASK, MAX_CPU_PER_TASK
 
@@ -91,6 +94,10 @@ class KubernetesDecorator(StepDecorator):
     tolerations : List[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
+        Kubernetes labels to use when launching pod in Kubernetes.
+    annotations: Dict[str, str], default: METAFLOW_KUBERNETES_ANNOTATIONS
+        Kubernetes annotations to use when launching pod in Kubernetes.
     use_tmpfs : bool, default False
         This enables an explicit tmpfs mount for this step.
     tmpfs_tempdir : bool, default True
@@ -133,6 +140,8 @@ class KubernetesDecorator(StepDecorator):
         "gpu_vendor": None,
         "tolerations": None,  # e.g., [{"key": "arch", "operator": "Equal", "value": "amd"},
         #                              {"key": "foo", "operator": "Equal", "value": "bar"}]
+        "labels": None,  # e.g. {"test-label": "value", "another-label":"value2"}
+        "annotations": None,  # e.g. {"note": "value", "another-note": "value2"}
         "use_tmpfs": None,
         "tmpfs_tempdir": True,
         "tmpfs_size": None,
@@ -219,6 +228,36 @@ class KubernetesDecorator(StepDecorator):
             self.attributes["memory"] = KUBERNETES_MEMORY
         if self.attributes["disk"] == self.defaults["disk"] and KUBERNETES_DISK:
             self.attributes["disk"] = KUBERNETES_DISK
+        # Label source precedence (decreasing):
+        # - System labels (set outside of decorator)
+        # - Decorator labels: @kubernetes(labels={})
+        # - Environment variable labels: METAFLOW_KUBERNETES_LABELS=
+        deco_labels = {}
+        if self.attributes["labels"] is not None:
+            deco_labels = self.attributes["labels"]
+
+        env_labels = {}
+        if KUBERNETES_LABELS:
+            env_labels = parse_kube_keyvalue_list(KUBERNETES_LABELS.split(","), False)
+
+        self.attributes["labels"] = {**env_labels, **deco_labels}
+
+        # Annotations
+        # annotation precedence (decreasing):
+        # - System annotations (set outside of decorator)
+        # - Decorator annotations: @kubernetes(annotations={})
+        # - Environment annotations: METAFLOW_KUBERNETES_ANNOTATIONS=
+        deco_annotations = {}
+        if self.attributes["annotations"] is not None:
+            deco_annotations = self.attributes["annotations"]
+
+        env_annotations = {}
+        if KUBERNETES_ANNOTATIONS:
+            env_annotations = parse_kube_keyvalue_list(
+                KUBERNETES_ANNOTATIONS.split(","), False
+            )
+
+        self.attributes["annotations"] = {**env_annotations, **deco_annotations}
 
         # If no docker image is explicitly specified, impute a default image.
         if not self.attributes["image"]:
@@ -386,6 +425,9 @@ class KubernetesDecorator(StepDecorator):
                     )
                 )
 
+        validate_kube_labels(self.attributes["labels"])
+        # TODO: add validation to annotations as well?
+
     def package_init(self, flow, step_name, environment):
         try:
             # Kubernetes is a soft dependency.
@@ -441,7 +483,12 @@ class KubernetesDecorator(StepDecorator):
                         "=".join([key, str(val)]) if val else key
                         for key, val in v.items()
                     ]
-                elif k in ["tolerations", "persistent_volume_claims"]:
+                elif k in [
+                    "tolerations",
+                    "persistent_volume_claims",
+                    "labels",
+                    "annotations",
+                ]:
                     cli_args.command_options[k] = json.dumps(v)
                 else:
                     cli_args.command_options[k] = v

metaflow/plugins/kubernetes/kubernetes_job.py

@@ -1,23 +1,20 @@
-import copy
 import json
 import math
 import random
-import sys
 import time
 
 from metaflow.exception import MetaflowException
 from metaflow.metaflow_config import KUBERNETES_SECRETS
 from metaflow.tracing import inject_tracing_vars
-from metaflow.unbounded_foreach import UBF_CONTROL, UBF_TASK
 from metaflow.metaflow_config_funcs import init_config
 
 CLIENT_REFRESH_INTERVAL_SECONDS = 300
+
+from .kube_utils import qos_requests_and_limits
 from .kubernetes_jobsets import (
     KubernetesJobSet,
 )  # We need this import for Kubernetes Client.
 
-from .kube_utils import qos_requests_and_limits
-
 
 class KubernetesJobException(MetaflowException):
     headline = "Kubernetes job error"
@@ -451,7 +448,7 @@ class RunningJob(object):
         def best_effort_kill():
             try:
                 self.kill()
-            except Exception as ex:
+            except Exception:
                 pass
 
         atexit.register(best_effort_kill)

metaflow/plugins/kubernetes/kubernetes_jobsets.py

@@ -1,4 +1,3 @@
-import copy
 import json
 import math
 import random
@@ -7,7 +6,6 @@ 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.metaflow_config import KUBERNETES_SECRETS
 
 from .kube_utils import qos_requests_and_limits
 
@@ -257,7 +255,7 @@ class RunningJobSet(object):
         def best_effort_kill():
             try:
                 self.kill()
-            except Exception as ex:
+            except Exception:
                 pass
 
         atexit.register(best_effort_kill)
@@ -342,7 +340,7 @@ class RunningJobSet(object):
                 stdout=True,
                 tty=False,
             )
-        except Exception as e:
+        except Exception:
             with client.ApiClient() as api_client:
                 # If we are unable to kill the control pod then
                 # Delete the jobset to kill the subsequent pods.
@@ -862,6 +860,16 @@ class KubernetesJobSet(object):
         self._annotations = dict(self._annotations, **{name: value})
         return self
 
+    def labels(self, labels):
+        for k, v in labels.items():
+            self.label(k, v)
+        return self
+
+    def annotations(self, annotations):
+        for k, v in annotations.items():
+            self.annotation(k, v)
+        return self
+
     def secret(self, name):
         self.worker.secret(name)
         self.control.secret(name)
@@ -987,15 +995,24 @@ class KubernetesArgoJobSet(object):
         self._labels = dict(self._labels, **{name: value})
         return self
 
+    def labels(self, labels):
+        for k, v in labels.items():
+            self.label(k, v)
+        return self
+
     def annotation(self, name, value):
         self.worker.annotation(name, value)
         self.control.annotation(name, value)
         self._annotations = dict(self._annotations, **{name: value})
         return self
 
+    def annotations(self, annotations):
+        for k, v in annotations.items():
+            self.annotation(k, v)
+        return self
+
     def dump(self):
         client = self._kubernetes_sdk
-        import json
 
         data = json.dumps(
             client.ApiClient().sanitize_for_serialization(