metaflow 2.13.4__tar.gz → 2.13.6__tar.gz

{metaflow-2.13.4/metaflow.egg-info → metaflow-2.13.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: metaflow
-Version: 2.13.4
+Version: 2.13.6
 Summary: Metaflow: More Data Science, Less Engineering
 Author: Metaflow Developers
 Author-email: help@metaflow.org
@@ -26,7 +26,7 @@ License-File: LICENSE
 Requires-Dist: requests
 Requires-Dist: boto3
 Provides-Extra: stubs
-Requires-Dist: metaflow-stubs==2.13.4; extra == "stubs"
+Requires-Dist: metaflow-stubs==2.13.6; extra == "stubs"
 Dynamic: author
 Dynamic: author-email
 Dynamic: classifier

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow/plugins/__init__.py

@@ -16,6 +16,7 @@ CLIS_DESC = [
     ("argo-workflows", ".argo.argo_workflows_cli.cli"),
     ("card", ".cards.card_cli.cli"),
     ("tag", ".tag_cli.cli"),
+    ("spot-metadata", ".kubernetes.spot_metadata_cli.cli"),
     ("logs", ".logs_cli.cli"),
 ]
 
@@ -104,6 +105,10 @@ SIDECARS_DESC = [
         "save_logs_periodically",
         "..mflog.save_logs_periodically.SaveLogsPeriodicallySidecar",
     ),
+    (
+        "spot_termination_monitor",
+        ".kubernetes.spot_monitor_sidecar.SpotTerminationMonitorSidecar",
+    ),
     ("heartbeat", "metaflow.metadata_provider.heartbeat.MetadataHeartBeat"),
 ]
 

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow/plugins/argo/argo_workflows.py

@@ -624,6 +624,16 @@ class ArgoWorkflows(object):
             for event in trigger_on_finish_deco.triggers:
                 # Actual filters are deduced here since we don't have access to
                 # the current object in the @trigger_on_finish decorator.
+                project_name = event.get("project") or current.get("project_name")
+                branch_name = event.get("branch") or current.get("branch_name")
+                # validate that we have complete project info for an event name
+                if project_name or branch_name:
+                    if not (project_name and branch_name):
+                        # if one of the two is missing, we would end up listening to an event that will never be broadcast.
+                        raise ArgoWorkflowsException(
+                            "Incomplete project info. Please specify both 'project' and 'project_branch' or use the @project decorator"
+                        )
+
                 triggers.append(
                     {
                         # Make sure this remains consistent with the event name format
@@ -632,18 +642,16 @@ class ArgoWorkflows(object):
                         % ".".join(
                             v
                             for v in [
-                                event.get("project") or current.get("project_name"),
-                                event.get("branch") or current.get("branch_name"),
+                                project_name,
+                                branch_name,
                                 event["flow"],
                             ]
                             if v
                         ),
                         "filters": {
                             "auto-generated-by-metaflow": True,
-                            "project_name": event.get("project")
-                            or current.get("project_name"),
-                            "branch_name": event.get("branch")
-                            or current.get("branch_name"),
+                            "project_name": project_name,
+                            "branch_name": branch_name,
                             # TODO: Add a time filters to guard against cached events
                         },
                         "type": "run",
@@ -841,8 +849,15 @@ class ArgoWorkflows(object):
                     Metadata()
                     .labels(self._base_labels)
                     .label("app.kubernetes.io/name", "metaflow-task")
-                    .annotations(annotations)
-                    .annotations(self._base_annotations)
+                    .annotations(
+                        {
+                            **annotations,
+                            **self._base_annotations,
+                            **{
+                                "metaflow/run_id": "argo-{{workflow.name}}"
+                            },  # we want pods of the workflow to have the run_id as an annotation as well
+                        }
+                    )
                 )
                 # Set the entrypoint to flow name
                 .entrypoint(self.flow.name)
@@ -1705,6 +1720,7 @@ class ArgoWorkflows(object):
                     },
                     **{
                         # Some optional values for bookkeeping
+                        "METAFLOW_FLOW_FILENAME": os.path.basename(sys.argv[0]),
                         "METAFLOW_FLOW_NAME": self.flow.name,
                         "METAFLOW_STEP_NAME": node.name,
                         "METAFLOW_RUN_ID": run_id,

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow/plugins/events_decorator.py

@@ -398,111 +398,23 @@ class TriggerOnFinishDecorator(FlowDecorator):
             )
         elif self.attributes["flow"]:
             # flow supports the format @trigger_on_finish(flow='FooFlow')
-            if is_stringish(self.attributes["flow"]):
-                self.triggers.append(
-                    {
-                        "fq_name": self.attributes["flow"],
-                    }
-                )
-            elif isinstance(self.attributes["flow"], dict):
-                if "name" not in self.attributes["flow"]:
-                    raise MetaflowException(
-                        "The *flow* attribute for *@trigger_on_finish* is missing the "
-                        "*name* key."
-                    )
-                flow_name = self.attributes["flow"]["name"]
-
-                if not is_stringish(flow_name) or "." in flow_name:
-                    raise MetaflowException(
-                        "The *name* attribute of the *flow* is not a valid string"
-                    )
-                result = {"fq_name": flow_name}
-                if "project" in self.attributes["flow"]:
-                    if is_stringish(self.attributes["flow"]["project"]):
-                        result["project"] = self.attributes["flow"]["project"]
-                    else:
-                        raise MetaflowException(
-                            "The *project* attribute of the *flow* is not a string"
-                        )
-                if "project_branch" in self.attributes["flow"]:
-                    if is_stringish(self.attributes["flow"]["project_branch"]):
-                        result["branch"] = self.attributes["flow"]["project_branch"]
-                    else:
-                        raise MetaflowException(
-                            "The *project_branch* attribute of the *flow* is not a string"
-                        )
-                self.triggers.append(result)
-            elif callable(self.attributes["flow"]) and not isinstance(
+            flow = self.attributes["flow"]
+            if callable(flow) and not isinstance(
                 self.attributes["flow"], DeployTimeField
             ):
-                trig = DeployTimeField(
-                    "fq_name", [str, dict], None, self.attributes["flow"], False
-                )
+                trig = DeployTimeField("fq_name", [str, dict], None, flow, False)
                 self.triggers.append(trig)
             else:
-                raise MetaflowException(
-                    "Incorrect type for *flow* attribute in *@trigger_on_finish* "
-                    " decorator. Supported type is string or Dict[str, str] - \n"
-                    "@trigger_on_finish(flow='FooFlow') or "
-                    "@trigger_on_finish(flow={'name':'FooFlow', 'project_branch': 'branch'})"
-                )
+                self.triggers.extend(self._parse_static_triggers([flow]))
         elif self.attributes["flows"]:
             # flows attribute supports the following formats -
             #     1. flows=['FooFlow', 'BarFlow']
-            if isinstance(self.attributes["flows"], list):
-                for flow in self.attributes["flows"]:
-                    if is_stringish(flow):
-                        self.triggers.append(
-                            {
-                                "fq_name": flow,
-                            }
-                        )
-                    elif isinstance(flow, dict):
-                        if "name" not in flow:
-                            raise MetaflowException(
-                                "One or more flows in the *flows* attribute for "
-                                "*@trigger_on_finish* is missing the "
-                                "*name* key."
-                            )
-                        flow_name = flow["name"]
-
-                        if not is_stringish(flow_name) or "." in flow_name:
-                            raise MetaflowException(
-                                "The *name* attribute '%s' is not a valid string"
-                                % str(flow_name)
-                            )
-                        result = {"fq_name": flow_name}
-                        if "project" in flow:
-                            if is_stringish(flow["project"]):
-                                result["project"] = flow["project"]
-                            else:
-                                raise MetaflowException(
-                                    "The *project* attribute of the *flow* '%s' is not "
-                                    "a string" % flow_name
-                                )
-                        if "project_branch" in flow:
-                            if is_stringish(flow["project_branch"]):
-                                result["branch"] = flow["project_branch"]
-                            else:
-                                raise MetaflowException(
-                                    "The *project_branch* attribute of the *flow* %s "
-                                    "is not a string" % flow_name
-                                )
-                        self.triggers.append(result)
-                    else:
-                        raise MetaflowException(
-                            "One or more flows in *flows* attribute in "
-                            "*@trigger_on_finish* decorator have an incorrect type. "
-                            "Supported type is string or Dict[str, str]- \n"
-                            "@trigger_on_finish(flows=['FooFlow', 'BarFlow']"
-                        )
-            elif callable(self.attributes["flows"]) and not isinstance(
-                self.attributes["flows"], DeployTimeField
-            ):
-                trig = DeployTimeField(
-                    "flows", list, None, self.attributes["flows"], False
-                )
+            flows = self.attributes["flows"]
+            if callable(flows) and not isinstance(flows, DeployTimeField):
+                trig = DeployTimeField("flows", list, None, flows, False)
                 self.triggers.append(trig)
+            elif isinstance(flows, list):
+                self.triggers.extend(self._parse_static_triggers(flows))
             else:
                 raise MetaflowException(
                     "Incorrect type for *flows* attribute in *@trigger_on_finish* "
@@ -519,26 +431,7 @@ class TriggerOnFinishDecorator(FlowDecorator):
         for trigger in self.triggers:
             if isinstance(trigger, DeployTimeField):
                 continue
-            if trigger["fq_name"].count(".") == 0:
-                # fully qualified name is just the flow name
-                trigger["flow"] = trigger["fq_name"]
-            elif trigger["fq_name"].count(".") >= 2:
-                # fully qualified name is of the format - project.branch.flow_name
-                trigger["project"], tail = trigger["fq_name"].split(".", maxsplit=1)
-                trigger["branch"], trigger["flow"] = tail.rsplit(".", maxsplit=1)
-            else:
-                raise MetaflowException(
-                    "Incorrect format for *flow* in *@trigger_on_finish* "
-                    "decorator. Specify either just the *flow_name* or a fully "
-                    "qualified name like *project_name.branch_name.flow_name*."
-                )
-            # TODO: Also sanity check project and branch names
-            if not re.match(r"^[A-Za-z0-9_]+$", trigger["flow"]):
-                raise MetaflowException(
-                    "Invalid flow name *%s* in *@trigger_on_finish* "
-                    "decorator. Only alphanumeric characters and "
-                    "underscores(_) are allowed." % trigger["flow"]
-                )
+            self._parse_fq_name(trigger)
 
         self.options = self.attributes["options"]
 
@@ -593,9 +486,67 @@ class TriggerOnFinishDecorator(FlowDecorator):
                 run_objs.append(run_obj)
             current._update_env({"trigger": Trigger.from_runs(run_objs)})
 
+    @staticmethod
+    def _parse_static_triggers(flows):
+        results = []
+        for flow in flows:
+            if is_stringish(flow):
+                results.append(
+                    {
+                        "fq_name": flow,
+                    }
+                )
+            elif isinstance(flow, dict):
+                if "name" not in flow:
+                    if len(flows) > 1:
+                        raise MetaflowException(
+                            "One or more flows in the *flows* attribute for "
+                            "*@trigger_on_finish* is missing the "
+                            "*name* key."
+                        )
+                    raise MetaflowException(
+                        "The *flow* attribute for *@trigger_on_finish* is missing the "
+                        "*name* key."
+                    )
+                flow_name = flow["name"]
+
+                if not is_stringish(flow_name) or "." in flow_name:
+                    raise MetaflowException(
+                        f"The *name* attribute of the *flow* {flow_name} is not a valid string"
+                    )
+                result = {"fq_name": flow_name}
+                if "project" in flow:
+                    if is_stringish(flow["project"]):
+                        result["project"] = flow["project"]
+                    else:
+                        raise MetaflowException(
+                            f"The *project* attribute of the *flow* {flow_name} is not a string"
+                        )
+                if "project_branch" in flow:
+                    if is_stringish(flow["project_branch"]):
+                        result["branch"] = flow["project_branch"]
+                    else:
+                        raise MetaflowException(
+                            f"The *project_branch* attribute of the *flow* {flow_name} is not a string"
+                        )
+                results.append(result)
+            else:
+                if len(flows) > 1:
+                    raise MetaflowException(
+                        "One or more flows in the *flows* attribute for "
+                        "*@trigger_on_finish* decorator have an incorrect type. "
+                        "Supported type is string or Dict[str, str]- \n"
+                        "@trigger_on_finish(flows=['FooFlow', 'BarFlow']"
+                    )
+                raise MetaflowException(
+                    "Incorrect type for *flow* attribute in *@trigger_on_finish* "
+                    " decorator. Supported type is string or Dict[str, str] - \n"
+                    "@trigger_on_finish(flow='FooFlow') or "
+                    "@trigger_on_finish(flow={'name':'FooFlow', 'project_branch': 'branch'})"
+                )
+        return results
+
     def _parse_fq_name(self, trigger):
-        if isinstance(trigger, DeployTimeField):
-            trigger["fq_name"] = deploy_time_eval(trigger["fq_name"])
         if trigger["fq_name"].count(".") == 0:
             # fully qualified name is just the flow name
             trigger["flow"] = trigger["fq_name"]
@@ -615,32 +566,18 @@ class TriggerOnFinishDecorator(FlowDecorator):
                 "decorator. Only alphanumeric characters and "
                 "underscores(_) are allowed." % trigger["flow"]
             )
-        return trigger
 
     def format_deploytime_value(self):
-        for trigger in self.triggers:
-            # Case were trigger is a function that returns a list
-            # Need to do this bc we need to iterate over list and process
-            if isinstance(trigger, DeployTimeField):
-                deploy_value = deploy_time_eval(trigger)
-                if isinstance(deploy_value, list):
-                    self.triggers = deploy_value
+        if len(self.triggers) == 1 and isinstance(self.triggers[0], DeployTimeField):
+            deploy_value = deploy_time_eval(self.triggers[0])
+            if isinstance(deploy_value, list):
+                self.triggers = deploy_value
             else:
-                break
-        for trigger in self.triggers:
-            # Entire trigger is a function (returns either string or dict)
-            old_trig = trigger
-            if isinstance(trigger, DeployTimeField):
-                trigger = deploy_time_eval(trigger)
-            if isinstance(trigger, dict):
-                trigger["fq_name"] = trigger.get("name")
-                trigger["project"] = trigger.get("project")
-                trigger["branch"] = trigger.get("project_branch")
-            # We also added this bc it won't be formatted yet
-            if isinstance(trigger, str):
-                trigger = {"fq_name": trigger}
-                trigger = self._parse_fq_name(trigger)
-            self.triggers[self.triggers.index(old_trig)] = trigger
+                self.triggers = [deploy_value]
+            triggers = self._parse_static_triggers(self.triggers)
+            for trigger in triggers:
+                self._parse_fq_name(trigger)
+            self.triggers = triggers
 
     def get_top_level_options(self):
         return list(self._option_values.items())

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow/plugins/kubernetes/kubernetes.py

@@ -669,15 +669,6 @@ class Kubernetes(object):
         for name, value in system_annotations.items():
             job.annotation(name, value)
 
-        (
-            job.annotation("metaflow/run_id", run_id)
-            .annotation("metaflow/step_name", step_name)
-            .annotation("metaflow/task_id", task_id)
-            .annotation("metaflow/attempt", attempt)
-            .label("app.kubernetes.io/name", "metaflow-task")
-            .label("app.kubernetes.io/part-of", "metaflow")
-        )
-
         return job
 
     def create_k8sjob(self, job):

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow/plugins/kubernetes/kubernetes_cli.py

@@ -190,7 +190,7 @@ def step(
     executable = ctx.obj.environment.executable(step_name, executable)
 
     # Set environment
-    env = {}
+    env = {"METAFLOW_FLOW_FILENAME": os.path.basename(sys.argv[0])}
     env_deco = [deco for deco in node.decorators if deco.name == "environment"]
     if env_deco:
         env = env_deco[0].attributes["vars"]

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow/plugins/kubernetes/kubernetes_decorator.py

@@ -547,6 +547,13 @@ class KubernetesDecorator(StepDecorator):
             self._save_logs_sidecar = Sidecar("save_logs_periodically")
             self._save_logs_sidecar.start()
 
+            # Start spot termination monitor sidecar.
+            current._update_env(
+                {"spot_termination_notice": "/tmp/spot_termination_notice"}
+            )
+            self._spot_monitor_sidecar = Sidecar("spot_termination_monitor")
+            self._spot_monitor_sidecar.start()
+
         num_parallel = None
         if hasattr(flow, "_parallel_ubf_iter"):
             num_parallel = flow._parallel_ubf_iter.num_parallel
@@ -605,6 +612,7 @@ class KubernetesDecorator(StepDecorator):
 
         try:
             self._save_logs_sidecar.terminate()
+            self._spot_monitor_sidecar.terminate()
         except:
             # Best effort kill
             pass

metaflow-2.13.6/metaflow/plugins/kubernetes/spot_metadata_cli.py

@@ -0,0 +1,69 @@
+from metaflow._vendor import click
+from datetime import datetime, timezone
+from metaflow.tagging_util import validate_tags
+from metaflow.metadata_provider import MetaDatum
+
+
+@click.group()
+def cli():
+    pass
+
+
+@cli.group(help="Commands related to spot metadata.")
+def spot_metadata():
+    pass
+
+
+@spot_metadata.command(help="Record spot termination metadata for a task.")
+@click.option(
+    "--run-id",
+    required=True,
+    help="Run ID for which metadata is to be recorded.",
+)
+@click.option(
+    "--step-name",
+    required=True,
+    help="Step Name for which metadata is to be recorded.",
+)
+@click.option(
+    "--task-id",
+    required=True,
+    help="Task ID for which metadata is to be recorded.",
+)
+@click.option(
+    "--termination-notice-time",
+    required=True,
+    help="Spot termination notice time.",
+)
+@click.option(
+    "--tag",
+    "tags",
+    multiple=True,
+    required=False,
+    default=None,
+    help="List of tags.",
+)
+@click.pass_obj
+def record(obj, run_id, step_name, task_id, termination_notice_time, tags=None):
+    validate_tags(tags)
+
+    tag_list = list(tags) if tags else []
+
+    entries = [
+        MetaDatum(
+            field="spot-termination-received-at",
+            value=datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
+            type="spot-termination-received-at",
+            tags=tag_list,
+        ),
+        MetaDatum(
+            field="spot-termination-time",
+            value=termination_notice_time,
+            type="spot-termination-time",
+            tags=tag_list,
+        ),
+    ]
+
+    obj.metadata.register_metadata(
+        run_id=run_id, step_name=step_name, task_id=task_id, metadata=entries
+    )

metaflow-2.13.6/metaflow/plugins/kubernetes/spot_monitor_sidecar.py

@@ -0,0 +1,109 @@
+import os
+import sys
+import time
+import signal
+import requests
+import subprocess
+from multiprocessing import Process
+from datetime import datetime, timezone
+from metaflow.sidecar import MessageTypes
+
+
+class SpotTerminationMonitorSidecar(object):
+    EC2_TYPE_URL = "http://169.254.169.254/latest/meta-data/instance-life-cycle"
+    METADATA_URL = "http://169.254.169.254/latest/meta-data/spot/termination-time"
+    TOKEN_URL = "http://169.254.169.254/latest/api/token"
+    POLL_INTERVAL = 5  # seconds
+
+    def __init__(self):
+        self.is_alive = True
+        self._process = None
+        self._token = None
+        self._token_expiry = 0
+
+        if self._is_aws_spot_instance():
+            self._process = Process(target=self._monitor_loop)
+            self._process.start()
+
+    def process_message(self, msg):
+        if msg.msg_type == MessageTypes.SHUTDOWN:
+            self.is_alive = False
+            if self._process:
+                self._process.terminate()
+
+    @classmethod
+    def get_worker(cls):
+        return cls
+
+    def _get_imds_token(self):
+        current_time = time.time()
+        if current_time >= self._token_expiry - 60:  # Refresh 60s before expiry
+            try:
+                response = requests.put(
+                    url=self.TOKEN_URL,
+                    headers={"X-aws-ec2-metadata-token-ttl-seconds": "300"},
+                    timeout=1,
+                )
+                if response.status_code == 200:
+                    self._token = response.text
+                    self._token_expiry = current_time + 240  # Slightly less than TTL
+            except requests.exceptions.RequestException:
+                pass
+        return self._token
+
+    def _make_ec2_request(self, url, timeout):
+        token = self._get_imds_token()
+        headers = {"X-aws-ec2-metadata-token": token} if token else {}
+        response = requests.get(url=url, headers=headers, timeout=timeout)
+        return response
+
+    def _is_aws_spot_instance(self):
+        try:
+            response = self._make_ec2_request(url=self.EC2_TYPE_URL, timeout=1)
+            return response.status_code == 200 and response.text == "spot"
+        except (requests.exceptions.RequestException, requests.exceptions.Timeout):
+            return False
+
+    def _monitor_loop(self):
+        while self.is_alive:
+            try:
+                response = self._make_ec2_request(url=self.METADATA_URL, timeout=1)
+                if response.status_code == 200:
+                    termination_time = response.text
+                    self._emit_termination_metadata(termination_time)
+                    os.kill(os.getppid(), signal.SIGTERM)
+                    break
+            except (requests.exceptions.RequestException, requests.exceptions.Timeout):
+                pass
+            time.sleep(self.POLL_INTERVAL)
+
+    def _emit_termination_metadata(self, termination_time):
+        flow_filename = os.getenv("METAFLOW_FLOW_FILENAME")
+        pathspec = os.getenv("MF_PATHSPEC")
+        _, run_id, step_name, task_id = pathspec.split("/")
+        retry_count = os.getenv("MF_ATTEMPT")
+
+        with open("/tmp/spot_termination_notice", "w") as fp:
+            fp.write(termination_time)
+
+        command = [
+            sys.executable,
+            f"/metaflow/{flow_filename}",
+            "spot-metadata",
+            "record",
+            "--run-id",
+            run_id,
+            "--step-name",
+            step_name,
+            "--task-id",
+            task_id,
+            "--termination-notice-time",
+            termination_time,
+            "--tag",
+            "attempt_id:{}".format(retry_count),
+        ]
+
+        result = subprocess.run(command, capture_output=True, text=True)
+
+        if result.returncode != 0:
+            print(f"Failed to record spot termination metadata: {result.stderr}")

metaflow-2.13.6/metaflow/version.py

@@ -0,0 +1 @@
+metaflow_version = "2.13.6"

{metaflow-2.13.4 → metaflow-2.13.6/metaflow.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: metaflow
-Version: 2.13.4
+Version: 2.13.6
 Summary: Metaflow: More Data Science, Less Engineering
 Author: Metaflow Developers
 Author-email: help@metaflow.org
@@ -26,7 +26,7 @@ License-File: LICENSE
 Requires-Dist: requests
 Requires-Dist: boto3
 Provides-Extra: stubs
-Requires-Dist: metaflow-stubs==2.13.4; extra == "stubs"
+Requires-Dist: metaflow-stubs==2.13.6; extra == "stubs"
 Dynamic: author
 Dynamic: author-email
 Dynamic: classifier

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow.egg-info/SOURCES.txt

@@ -305,6 +305,8 @@ metaflow/plugins/kubernetes/kubernetes_client.py
 metaflow/plugins/kubernetes/kubernetes_decorator.py
 metaflow/plugins/kubernetes/kubernetes_job.py
 metaflow/plugins/kubernetes/kubernetes_jobsets.py
+metaflow/plugins/kubernetes/spot_metadata_cli.py
+metaflow/plugins/kubernetes/spot_monitor_sidecar.py
 metaflow/plugins/metadata_providers/__init__.py
 metaflow/plugins/metadata_providers/local.py
 metaflow/plugins/metadata_providers/service.py

{metaflow-2.13.4 → metaflow-2.13.6}/metaflow.egg-info/requires.txt

@@ -2,4 +2,4 @@ requests
 boto3
 
 [stubs]
-metaflow-stubs==2.13.4
+metaflow-stubs==2.13.6

metaflow-2.13.4/metaflow/version.py

@@ -1 +0,0 @@
-metaflow_version = "2.13.4"