ob-metaflow-extensions 1.1.83__py2.py3-none-any.whl → 1.1.86__py2.py3-none-any.whl

metaflow_extensions/outerbounds/config/__init__.py

@@ -14,6 +14,13 @@ DEFAULT_GCP_CLIENT_PROVIDER = "obp"
 FAST_BAKERY_URL = from_conf("FAST_BAKERY_URL", None)
 
 
+###
+# NVCF configuration
+###
+# Maximum number of consecutive heartbeats that can be missed.
+NVIDIA_HEARTBEAT_THRESHOLD = from_conf("NVIDIA_HEARTBEAT_THRESHOLD", "3")
+
+
 ###
 # Snowpark configuration
 ###

metaflow_extensions/outerbounds/plugins/__init__.py

@@ -32,6 +32,126 @@ def hide_access_keys(*args, **kwds):
             os.environ["AWS_SESSION_TOKEN"] = AWS_SESSION_TOKEN
 
 
+# This is a special placeholder value that can be passed as role_arn to
+# get_boto3_session() which makes it use the CSPR role, if its set.
+USE_CSPR_ROLE_ARN_IF_SET = "__cspr__"
+
+
+def get_boto3_session(role_arn=None, session_vars=None):
+    import boto3
+    import botocore
+    from metaflow_extensions.outerbounds.plugins.auth_server import get_token
+
+    from hashlib import sha256
+    from metaflow.util import get_username
+
+    user = get_username()
+
+    token_info = get_token("/generate/aws")
+
+    # Write token to a file. The file name is derived from the user name
+    # so it works with multiple users on the same machine.
+    #
+    # We hash the user name so we don't have to deal with special characters
+    # in the file name and the file name is not exposed to the user
+    # anyways, so it doesn't matter that its a little ugly.
+    token_file = "/tmp/obp_token." + sha256(user.encode("utf-8")).hexdigest()[:16]
+
+    # Write to a temp file then rename to avoid a situation when someone
+    # tries to read the file after it was open for writing (and truncated)
+    # but before the token was written to it.
+    with tempfile.NamedTemporaryFile("w", delete=False) as f:
+        f.write(token_info["token"])
+        tmp_token_file = f.name
+    os.rename(tmp_token_file, token_file)
+
+    cspr_role = None
+    if token_info.get("cspr_role_arn"):
+        cspr_role = token_info["cspr_role_arn"]
+
+    if cspr_role:
+        # If CSPR role is set, we set it as the default role to assume
+        # for the AWS SDK. We do this by writing an AWS config file
+        # with two profiles. One to get credentials for the task role
+        # in exchange for the OIDC token, and second to assume the
+        # CSPR role using the task role credentials.
+        import configparser
+        from io import StringIO
+
+        aws_config = configparser.ConfigParser()
+
+        # Task role profile
+        aws_config["profile task"] = {
+            "role_arn": token_info["role_arn"],
+            "web_identity_token_file": token_file,
+        }
+
+        # CSPR role profile (default)
+        aws_config["profile cspr"] = {
+            "role_arn": cspr_role,
+            "source_profile": "task",
+        }
+
+        aws_config_string = StringIO()
+        aws_config.write(aws_config_string)
+        aws_config_file = (
+            "/tmp/aws_config." + sha256(user.encode("utf-8")).hexdigest()[:16]
+        )
+        with tempfile.NamedTemporaryFile(
+            "w", delete=False, dir=os.path.dirname(aws_config_file)
+        ) as f:
+            f.write(aws_config_string.getvalue())
+            tmp_aws_config_file = f.name
+        os.rename(tmp_aws_config_file, aws_config_file)
+        os.environ["AWS_CONFIG_FILE"] = aws_config_file
+        os.environ["AWS_DEFAULT_PROFILE"] = "cspr"
+    else:
+        os.environ["AWS_WEB_IDENTITY_TOKEN_FILE"] = token_file
+        os.environ["AWS_ROLE_ARN"] = token_info["role_arn"]
+
+    # Enable regional STS endpoints. This is the new recommended way
+    # by AWS [1] and is the more performant way.
+    # [1] https://docs.aws.amazon.com/sdkref/latest/guide/feature-sts-regionalized-endpoints.html
+    os.environ["AWS_STS_REGIONAL_ENDPOINTS"] = "regional"
+    if token_info.get("region"):
+        os.environ["AWS_DEFAULT_REGION"] = token_info["region"]
+
+    with hide_access_keys():
+        if cspr_role:
+            #  The generated AWS config will be used here since we set the
+            #  AWS_CONFIG_FILE environment variable above.
+            if role_arn == USE_CSPR_ROLE_ARN_IF_SET:
+                # Otherwise start from the default profile, assuming CSPR role
+                session = boto3.session.Session(profile_name="default")
+            else:
+                session = boto3.session.Session(profile_name="task")
+        else:
+            # Not using AWS config, just AWS_WEB_IDENTITY_TOKEN_FILE + AWS_ROLE_ARN
+            session = boto3.session.Session()
+
+        if role_arn and role_arn != USE_CSPR_ROLE_ARN_IF_SET:
+            # If the user provided a role_arn, we assume that role
+            # using the task role credentials. CSPR role is not used.
+            fetcher = botocore.credentials.AssumeRoleCredentialFetcher(
+                client_creator=session._session.create_client,
+                source_credentials=session._session.get_credentials(),
+                role_arn=role_arn,
+                extra_args={},
+            )
+            creds = botocore.credentials.DeferredRefreshableCredentials(
+                method="assume-role", refresh_using=fetcher.fetch_credentials
+            )
+            botocore_session = botocore.session.Session(session_vars=session_vars)
+            botocore_session._credentials = creds
+            return boto3.session.Session(botocore_session=botocore_session)
+        else:
+            # If the user didn't provide a role_arn, or if the role_arn
+            # is set to USE_CSPR_ROLE_ARN_IF_SET, we return the default
+            # session which would use the CSPR role if it is set on the
+            # server, and the task role otherwise.
+            return session
+
+
 class ObpAuthProvider(object):
     name = "obp"
 
@@ -42,67 +162,13 @@ class ObpAuthProvider(object):
         if client_params is None:
             client_params = {}
 
-        import boto3
-        import botocore
         from botocore.exceptions import ClientError
-        from metaflow_extensions.outerbounds.plugins.auth_server import get_token
-
-        from hashlib import sha256
-        from metaflow.util import get_username
-
-        user = get_username()
-
-        token_info = get_token("/generate/aws")
 
-        # Write token to a file. The file name is derived from the user name
-        # so it works with multiple users on the same machine.
-        #
-        # We hash the user name so we don't have to deal with special characters
-        # in the file name and the file name is not exposed to the user
-        # anyways, so it doesn't matter that its a little ugly.
-        token_file = "/tmp/obp_token." + sha256(user.encode("utf-8")).hexdigest()[:16]
-
-        # Write to a temp file then rename to avoid a situation when someone
-        # tries to read the file after it was open for writing (and truncated)
-        # but before the token was written to it.
-        with tempfile.NamedTemporaryFile("w", delete=False) as f:
-            f.write(token_info["token"])
-            tmp_token_file = f.name
-        os.rename(tmp_token_file, token_file)
-
-        os.environ["AWS_WEB_IDENTITY_TOKEN_FILE"] = token_file
-        os.environ["AWS_ROLE_ARN"] = token_info["role_arn"]
-
-        # Enable regional STS endpoints. This is the new recommended way
-        # by AWS [1] and is the more performant way.
-        # [1] https://docs.aws.amazon.com/sdkref/latest/guide/feature-sts-regionalized-endpoints.html
-        os.environ["AWS_STS_REGIONAL_ENDPOINTS"] = "regional"
-        if token_info.get("region"):
-            os.environ["AWS_DEFAULT_REGION"] = token_info["region"]
-
-        with hide_access_keys():
-            if role_arn:
-                session = boto3.session.Session()
-                fetcher = botocore.credentials.AssumeRoleCredentialFetcher(
-                    client_creator=session._session.create_client,
-                    source_credentials=session._session.get_credentials(),
-                    role_arn=role_arn,
-                    extra_args={},
-                )
-                creds = botocore.credentials.DeferredRefreshableCredentials(
-                    method="assume-role", refresh_using=fetcher.fetch_credentials
-                )
-                botocore_session = botocore.session.Session(session_vars=session_vars)
-                botocore_session._credentials = creds
-                session = boto3.session.Session(botocore_session=botocore_session)
-                if with_error:
-                    return session.client(module, **client_params), ClientError
-                else:
-                    return session.client(module, **client_params)
-            if with_error:
-                return boto3.client(module, **client_params), ClientError
-            else:
-                return boto3.client(module, **client_params)
+        session = get_boto3_session(role_arn, session_vars)
+        if with_error:
+            return session.client(module, **client_params), ClientError
+        else:
+            return session.client(module, **client_params)
 
 
 AWS_CLIENT_PROVIDERS_DESC = [("obp", ".ObpAuthProvider")]

metaflow_extensions/outerbounds/plugins/nvcf/heartbeat_store.py

@@ -0,0 +1,157 @@
+import os
+import sys
+import time
+import signal
+from io import BytesIO
+from datetime import datetime, timezone
+
+from metaflow.exception import MetaflowException
+
+
+class HeartbeatStore(object):
+    def __init__(
+        self,
+        heartbeat_prefix,
+        main_pid=None,
+        storage_backend=None,
+        emit_frequency=30,
+        missed_heartbeat_timeout=60,
+        monitor_frequency=15,
+        max_missed_heartbeats=3,
+    ) -> None:
+        self.heartbeat_prefix = heartbeat_prefix
+        self.main_pid = main_pid
+        self.storage_backend = storage_backend
+        self.emit_frequency = emit_frequency
+        self.monitor_frequency = monitor_frequency
+        self.missed_heartbeat_timeout = missed_heartbeat_timeout
+        self.max_missed_heartbeats = max_missed_heartbeats
+        self.missed_heartbeats = 0
+
+    def emit_heartbeat(self, folder_name=None):
+        heartbeat_key = f"{self.heartbeat_prefix}/heartbeat"
+        if folder_name:
+            heartbeat_key = f"{folder_name}/{heartbeat_key}"
+
+        while True:
+            try:
+                epoch_string = str(datetime.now(timezone.utc).timestamp()).encode(
+                    "utf-8"
+                )
+                self.storage_backend.save_bytes(
+                    [(heartbeat_key, BytesIO(bytes(epoch_string)))], overwrite=True
+                )
+            except Exception as e:
+                print(f"Error writing heartbeat: {e}")
+                sys.exit(1)
+
+            time.sleep(self.emit_frequency)
+
+    def emit_tombstone(self, folder_name=None):
+        tombstone_key = f"{self.heartbeat_prefix}/tombstone"
+        if folder_name:
+            tombstone_key = f"{folder_name}/{tombstone_key}"
+
+        tombstone_string = "tombstone".encode("utf-8")
+        try:
+            self.storage_backend.save_bytes(
+                [(tombstone_key, BytesIO(bytes(tombstone_string)))], overwrite=True
+            )
+        except Exception as e:
+            print(f"Error writing tombstone: {e}")
+            sys.exit(1)
+
+    def __handle_tombstone(self, path):
+        if path is not None:
+            with open(path) as f:
+                contents = f.read()
+                if "tombstone" in contents:
+                    print("[Outerbounds] Tombstone detected. Terminating the task..")
+                    os.kill(self.main_pid, signal.SIGTERM)
+                    sys.exit(1)
+
+    def __handle_heartbeat(self, path):
+        if path is not None:
+            with open(path) as f:
+                contents = f.read()
+                current_timestamp = datetime.now(timezone.utc).timestamp()
+                if current_timestamp - float(contents) > self.missed_heartbeat_timeout:
+                    self.missed_heartbeats += 1
+                else:
+                    self.missed_heartbeats = 0
+        else:
+            self.missed_heartbeats += 1
+
+        if self.missed_heartbeats > self.max_missed_heartbeats:
+            print(
+                f"[Outerbounds] Missed {self.max_missed_heartbeats} consecutive heartbeats. Terminating the task.."
+            )
+            os.kill(self.main_pid, signal.SIGTERM)
+            sys.exit(1)
+
+    def is_main_process_running(self):
+        try:
+            # Check if the process is running
+            os.kill(self.main_pid, 0)
+        except ProcessLookupError:
+            return False
+        return True
+
+    def monitor(self, folder_name=None):
+        heartbeat_key = f"{self.heartbeat_prefix}/heartbeat"
+        if folder_name:
+            heartbeat_key = f"{folder_name}/{heartbeat_key}"
+
+        tombstone_key = f"{self.heartbeat_prefix}/tombstone"
+        if folder_name:
+            tombstone_key = f"{folder_name}/{tombstone_key}"
+
+        while self.is_main_process_running():
+            with self.storage_backend.load_bytes(
+                [heartbeat_key, tombstone_key]
+            ) as results:
+                for key, path, _ in results:
+                    if key == tombstone_key:
+                        self.__handle_tombstone(path)
+                    elif key == heartbeat_key:
+                        self.__handle_heartbeat(path)
+
+            time.sleep(self.monitor_frequency)
+
+
+if __name__ == "__main__":
+    from metaflow.plugins import DATASTORES
+    from metaflow.metaflow_config import NVIDIA_HEARTBEAT_THRESHOLD
+
+    if len(sys.argv) != 4:
+        print("Usage: heartbeat_store.py <main_pid> <datastore_type> <folder_name>")
+        sys.exit(1)
+    _, main_pid, datastore_type, folder_name = sys.argv
+
+    if datastore_type not in ("azure", "gs", "s3"):
+        print(f"Datastore unsupported for type: {datastore_type}")
+        sys.exit(1)
+
+    datastores = [d for d in DATASTORES if d.TYPE == datastore_type]
+    datastore_sysroot = datastores[0].get_datastore_root_from_config(
+        lambda *args, **kwargs: None
+    )
+    if datastore_sysroot is None:
+        raise MetaflowException(
+            msg="METAFLOW_DATASTORE_SYSROOT_{datastore_type} must be set!".format(
+                datastore_type=datastore_type.upper()
+            )
+        )
+
+    storage = datastores[0](datastore_sysroot)
+
+    heartbeat_prefix = f"{os.getenv('MF_PATHSPEC')}/{os.getenv('MF_ATTEMPT')}"
+
+    store = HeartbeatStore(
+        heartbeat_prefix=heartbeat_prefix,
+        main_pid=int(main_pid),
+        storage_backend=storage,
+        max_missed_heartbeats=int(NVIDIA_HEARTBEAT_THRESHOLD),
+    )
+
+    store.monitor(folder_name=folder_name)

metaflow_extensions/outerbounds/plugins/nvcf/nvcf.py

@@ -1,6 +1,7 @@
 import json
 import os
-import time
+import threading
+from urllib.parse import urlparse
 from urllib.request import HTTPError, Request, URLError, urlopen
 
 from metaflow import util
@@ -14,6 +15,7 @@ from metaflow.mflog import (
 )
 import requests
 from metaflow.metaflow_config_funcs import init_config
+from metaflow_extensions.outerbounds.plugins.nvcf.heartbeat_store import HeartbeatStore
 
 
 class NvcfException(MetaflowException):
@@ -58,10 +60,11 @@ class Nvcf(object):
             code_package_url, code_package_ds
         )
         init_expr = " && ".join(init_cmds)
+        heartbeat_expr = f'python -m metaflow_extensions.outerbounds.plugins.nvcf.heartbeat_store "$MAIN_PID" {code_package_ds} nvcf_heartbeats 1>> $MFLOG_STDOUT 2>> $MFLOG_STDERR'
         step_expr = bash_capture_logs(
             " && ".join(
                 self.environment.bootstrap_commands(step_name, code_package_ds)
-                + [step_cli]
+                + [step_cli + " & MAIN_PID=$!; " + heartbeat_expr]
             )
         )
 
@@ -87,7 +90,9 @@ class Nvcf(object):
             '${METAFLOW_INIT_SCRIPT:+eval \\"${METAFLOW_INIT_SCRIPT}\\"} && %s'
             % cmd_str
         )
-        self.job = Job('bash -c "%s"' % cmd_str, env)
+        self.job = Job(
+            'bash -c "%s"' % cmd_str, env, task_spec, self.datastore._storage_impl
+        )
         self.job.submit()
 
     def wait(self, stdout_location, stderr_location, echo=None):
@@ -137,8 +142,7 @@ result_endpoint = f"{nvcf_url}/v2/nvcf/pexec/status"
 
 
 class Job(object):
-    def __init__(self, command, env):
-
+    def __init__(self, command, env, task_spec, backend):
         self._payload = {
             "command": command,
             "env": {k: v for k, v in env.items() if v is not None},
@@ -170,6 +174,27 @@ class Job(object):
             if f["model_key"] == "metaflow_task_executor":
                 self._function_id = f["id"]
 
+        flow_name = task_spec.get("flow_name")
+        run_id = task_spec.get("run_id")
+        step_name = task_spec.get("step_name")
+        task_id = task_spec.get("task_id")
+        retry_count = task_spec.get("retry_count")
+
+        heartbeat_prefix = "/".join(
+            (flow_name, str(run_id), step_name, str(task_id), str(retry_count))
+        )
+
+        store = HeartbeatStore(
+            heartbeat_prefix=heartbeat_prefix,
+            main_pid=None,
+            storage_backend=backend,
+        )
+
+        self.heartbeat_thread = threading.Thread(
+            target=store.emit_heartbeat, args=("nvcf_heartbeats",), daemon=True
+        )
+        self.heartbeat_thread.start()
+
     def submit(self):
         try:
             headers = {
@@ -224,7 +249,6 @@ class Job(object):
 
     def _poll(self):
         try:
-            invocation_id = self._invocation_id
             headers = {
                 "Authorization": f"Bearer {self._ngc_api_key}",
                 "Content-Type": "application/json",

metaflow_extensions/outerbounds/plugins/nvcf/nvcf_cli.py

@@ -4,7 +4,7 @@ import sys
 import time
 import traceback
 
-from metaflow import util
+from metaflow import util, Run
 from metaflow._vendor import click
 from metaflow.exception import METAFLOW_EXIT_DISALLOW_RETRY
 from metaflow.metadata.util import sync_local_metadata_from_datastore
@@ -27,6 +27,7 @@ from metaflow.metaflow_config import (
     CARD_GSROOT,
     KUBERNETES_SANDBOX_INIT_SCRIPT,
     OTEL_ENDPOINT,
+    NVIDIA_HEARTBEAT_THRESHOLD,
 )
 from metaflow.mflog import TASK_LOG_SOURCE
 
@@ -43,6 +44,65 @@ def nvcf():
     pass
 
 
+@nvcf.command(help="List steps / tasks running as an NVCF job.")
+@click.argument("run-id")
+@click.pass_context
+def list(ctx, run_id):
+    flow_name = ctx.obj.flow.name
+    run_obj = Run(pathspec=f"{flow_name}/{run_id}", _namespace_check=False)
+    running_invocations = []
+    for each_step in run_obj:
+        if (
+            not each_step.task.finished
+            and "nvcf-function-id" in each_step.task.metadata_dict
+        ):
+            task_pathspec = each_step.task.pathspec
+            attempt = each_step.task.metadata_dict.get("attempt")
+            flow_name, run_id, step_name, task_id = task_pathspec.split("/")
+            running_invocations.append(
+                f"Flow Name: {flow_name}, Run ID: {run_id}, Step Name: {step_name}, Task ID: {task_id}, Retry Count: {attempt}"
+            )
+
+    if running_invocations:
+        for each_invocation in running_invocations:
+            print(each_invocation)
+    else:
+        print("No running NVCF invocations for Run ID: %s" % run_id)
+
+
+@nvcf.command(help="Kill steps / tasks running as an NVCF job.")
+@click.argument("run-id")
+@click.pass_context
+def kill(ctx, run_id):
+    from metaflow_extensions.outerbounds.plugins.nvcf.heartbeat_store import (
+        HeartbeatStore,
+    )
+
+    flow_name = ctx.obj.flow.name
+    run_obj = Run(pathspec=f"{flow_name}/{run_id}", _namespace_check=False)
+
+    for each_step in run_obj:
+        if (
+            not each_step.task.finished
+            and "nvcf-function-id" in each_step.task.metadata_dict
+        ):
+            task_pathspec = each_step.task.pathspec
+            attempt = each_step.task.metadata_dict.get("attempt")
+            heartbeat_prefix = "{task_pathspec}/{attempt}".format(
+                task_pathspec=task_pathspec, attempt=attempt
+            )
+
+            datastore_root = ctx.obj.datastore_impl.datastore_root
+            store = HeartbeatStore(
+                heartbeat_prefix=heartbeat_prefix,
+                main_pid=None,
+                storage_backend=ctx.obj.datastore_impl(datastore_root),
+            )
+            store.emit_tombstone(folder_name="nvcf_heartbeats")
+        else:
+            print("No running NVCF invocations for Run ID: %s" % run_id)
+
+
 @nvcf.command(
     help="Execute a single task using NVCF. This command calls the "
     "top-level step command inside a NVCF job with the given options. "
@@ -139,6 +199,7 @@ def step(ctx, step_name, code_package_sha, code_package_url, **kwargs):
         "METAFLOW_CARD_GSROOT": CARD_GSROOT,
         "METAFLOW_INIT_SCRIPT": KUBERNETES_SANDBOX_INIT_SCRIPT,
         "METAFLOW_OTEL_ENDPOINT": OTEL_ENDPOINT,
+        "METAFLOW_NVIDIA_HEARTBEAT_THRESHOLD": str(NVIDIA_HEARTBEAT_THRESHOLD),
     }
 
     env_deco = [deco for deco in node.decorators if deco.name == "environment"]

metaflow_extensions/outerbounds/toplevel/global_aliases_for_metaflow_package.py

@@ -5,6 +5,49 @@
 __version__ = "v1"
 __mf_extensions__ = "ob"
 
-# To support "from metaflow import get_aws_client"
-from metaflow.plugins.aws.aws_client import get_aws_client
+
+# Must match the signature of metaflow.plugins.aws.aws_client.get_aws_client
+# This function is called by the "userland" code inside tasks. Metaflow internals
+# will call the function in metaflow.plugins.aws.aws_client.get_aws_client directly.
+#
+# Unlike the original function, this wrapper will use the CSPR role if both of the following
+# conditions are met:
+#
+#  1. CSPR is set
+#  2. user didn't provide a role to assume explicitly.
+#
+def get_aws_client(
+    module, with_error=False, role_arn=None, session_vars=None, client_params=None
+):
+    import metaflow.plugins.aws.aws_client
+
+    from metaflow_extensions.outerbounds.plugins import USE_CSPR_ROLE_ARN_IF_SET
+
+    return metaflow.plugins.aws.aws_client.get_aws_client(
+        module,
+        with_error=with_error,
+        role_arn=role_arn or USE_CSPR_ROLE_ARN_IF_SET,
+        session_vars=session_vars,
+        client_params=client_params,
+    )
+
+
+# This should match the signature of metaflow.plugins.datatools.s3.S3.
+#
+# This assumes that "userland" code inside tasks will call this, while Metaflow
+# internals will call metaflow.plugins.datatools.s3.S3 directly.
+#
+# This wrapper will make S3() use the CSPR role if its set, and user didn't provide
+# a role to assume explicitly.
+def S3(*args, **kwargs):
+    import sys
+    import metaflow.plugins.datatools.s3
+    from metaflow_extensions.outerbounds.plugins import USE_CSPR_ROLE_ARN_IF_SET
+
+    if "role" not in kwargs or kwargs["role"] is None:
+        kwargs["role"] = USE_CSPR_ROLE_ARN_IF_SET
+
+    return metaflow.plugins.datatools.s3.S3(*args, **kwargs)
+
+
 from .. import profilers

{ob_metaflow_extensions-1.1.83.dist-info → ob_metaflow_extensions-1.1.86.dist-info}/METADATA

@@ -1,13 +1,13 @@
 Metadata-Version: 2.1
 Name: ob-metaflow-extensions
-Version: 1.1.83
+Version: 1.1.86
 Summary: Outerbounds Platform Extensions for Metaflow
 Author: Outerbounds, Inc.
 License: Commercial
 Description-Content-Type: text/markdown
 Requires-Dist: boto3
 Requires-Dist: kubernetes
-Requires-Dist: ob-metaflow (==2.12.18.1)
+Requires-Dist: ob-metaflow (==2.12.18.2)
 
 # Outerbounds platform package
 

{ob_metaflow_extensions-1.1.83.dist-info → ob_metaflow_extensions-1.1.86.dist-info}/RECORD

@@ -1,7 +1,7 @@
 metaflow_extensions/outerbounds/__init__.py,sha256=TRGvIUMjkfneWtYUFSWoubu_Kf2ekAL4WLbV3IxOj9k,499
 metaflow_extensions/outerbounds/remote_config.py,sha256=Zpfpjgz68_ZgxlXezjzlsDLo4840rkWuZgwDB_5H57U,4059
-metaflow_extensions/outerbounds/config/__init__.py,sha256=MwC9dK3A5waKt-DOdHJMw-7sA5Zrl89uLmYJiM3mucc,1057
-metaflow_extensions/outerbounds/plugins/__init__.py,sha256=Y6Y2RlZFW5RwZjXa5QrKptht-u1p8faJxFFaA2n9Jy8,10074
+metaflow_extensions/outerbounds/config/__init__.py,sha256=JsQGRuGFz28fQWjUvxUgR8EKBLGRdLUIk_buPLJplJY,1225
+metaflow_extensions/outerbounds/plugins/__init__.py,sha256=g07Xj7YifwLfTTa94cyf3kLebHezeLWNG9HoGnpDyMo,12519
 metaflow_extensions/outerbounds/plugins/auth_server.py,sha256=1v2GBqoMBxp5E7Lejz139w-jxJtPnLDvvHXP0HhEIHI,2361
 metaflow_extensions/outerbounds/plugins/perimeters.py,sha256=QXh3SFP7GQbS-RAIxUOPbhPzQ7KDFVxZkTdKqFKgXjI,2697
 metaflow_extensions/outerbounds/plugins/fast_bakery/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -14,8 +14,9 @@ metaflow_extensions/outerbounds/plugins/kubernetes/kubernetes_client.py,sha256=g
 metaflow_extensions/outerbounds/plugins/nim/__init__.py,sha256=GVnvSTjqYVj5oG2yh8KJFt7iZ33cEadDD5HbdmC9hJ0,1457
 metaflow_extensions/outerbounds/plugins/nim/nim_manager.py,sha256=SWieODDxtIaeZwdMYtObDi57Kjyfw2DUuE6pJtU750w,9206
 metaflow_extensions/outerbounds/plugins/nvcf/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-metaflow_extensions/outerbounds/plugins/nvcf/nvcf.py,sha256=ftxC5SCo64P5Ycpv5vudluTnQi3-VCZW0umdsPP326A,7926
-metaflow_extensions/outerbounds/plugins/nvcf/nvcf_cli.py,sha256=ow3lonclEDoZEUQCDV_L8lEr6HopXqjNXzubRrfdIm4,7219
+metaflow_extensions/outerbounds/plugins/nvcf/heartbeat_store.py,sha256=wIlPBzsTszkHpftK1x7zBgaQ_7d3tNURqh4ez71Ra7A,5416
+metaflow_extensions/outerbounds/plugins/nvcf/nvcf.py,sha256=NIt1kJHuYpnCF7n73A90ZITWsk5QWtsbiHfzvdVjgqk,8997
+metaflow_extensions/outerbounds/plugins/nvcf/nvcf_cli.py,sha256=AhlFa8UVzVKQRUilQwtwp7ZVWJ0WNitPU0vp29i7WuY,9545
 metaflow_extensions/outerbounds/plugins/nvcf/nvcf_decorator.py,sha256=0xNA4aRTPJ4SKpRIFKZzlL9a7lf367KGTrVWVXd-uGE,6052
 metaflow_extensions/outerbounds/plugins/snowpark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 metaflow_extensions/outerbounds/plugins/snowpark/snowpark.py,sha256=vzgpVLCKvHjzHNfJvmH0jcxefYNsVggw_vof_y_U_a8,10643
@@ -28,11 +29,11 @@ metaflow_extensions/outerbounds/plugins/snowpark/snowpark_service_spec.py,sha256
 metaflow_extensions/outerbounds/profilers/__init__.py,sha256=wa_jhnCBr82TBxoS0e8b6_6sLyZX0fdHicuGJZNTqKw,29
 metaflow_extensions/outerbounds/profilers/gpu.py,sha256=a5YZAepujuP0uDqG9UpXBlZS3wjUt4Yv8CjybXqeT2c,24342
 metaflow_extensions/outerbounds/toplevel/__init__.py,sha256=qWUJSv_r5hXJ7jV_On4nEasKIfUCm6_UjkjXWA_A1Ts,90
-metaflow_extensions/outerbounds/toplevel/global_aliases_for_metaflow_package.py,sha256=efl01b4O7mRXerjOtLUD-CQ2l7ZGG78iyEXRGMAJYsU,412
+metaflow_extensions/outerbounds/toplevel/global_aliases_for_metaflow_package.py,sha256=Zq3OuL1bOod8KJra-Zk8B3gNhSHoWEGteM9T7g0pp6E,1881
 metaflow_extensions/outerbounds/toplevel/plugins/azure/__init__.py,sha256=WUuhz2YQfI4fz7nIcipwwWq781eaoHEk7n4GAn1npDg,63
 metaflow_extensions/outerbounds/toplevel/plugins/gcp/__init__.py,sha256=BbZiaH3uILlEZ6ntBLKeNyqn3If8nIXZFq_Apd7Dhco,70
 metaflow_extensions/outerbounds/toplevel/plugins/kubernetes/__init__.py,sha256=5zG8gShSj8m7rgF4xgWBZFuY3GDP5n1T0ktjRpGJLHA,69
-ob_metaflow_extensions-1.1.83.dist-info/METADATA,sha256=jjlyH-CO-VHAH8Q3T8THUvoPPfRf6OIreJNk8Otjj1c,520
-ob_metaflow_extensions-1.1.83.dist-info/WHEEL,sha256=bb2Ot9scclHKMOLDEHY6B2sicWOgugjFKaJsT7vwMQo,110
-ob_metaflow_extensions-1.1.83.dist-info/top_level.txt,sha256=NwG0ukwjygtanDETyp_BUdtYtqIA_lOjzFFh1TsnxvI,20
-ob_metaflow_extensions-1.1.83.dist-info/RECORD,,
+ob_metaflow_extensions-1.1.86.dist-info/METADATA,sha256=LKi6v-1FMvAWE2u9eW2PKoVBi1xl6Wor1TjhwYVzPNk,520
+ob_metaflow_extensions-1.1.86.dist-info/WHEEL,sha256=bb2Ot9scclHKMOLDEHY6B2sicWOgugjFKaJsT7vwMQo,110
+ob_metaflow_extensions-1.1.86.dist-info/top_level.txt,sha256=NwG0ukwjygtanDETyp_BUdtYtqIA_lOjzFFh1TsnxvI,20
+ob_metaflow_extensions-1.1.86.dist-info/RECORD,,