konduktor-nightly 0.1.0.dev20250414104837__tar.gz → 0.1.0.dev20250415104823__tar.gz

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: konduktor-nightly
-Version: 0.1.0.dev20250414104837
+Version: 0.1.0.dev20250415104823
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/konduktor/__init__.py

@@ -14,7 +14,7 @@ __all__ = [
 ]
 
 # Replaced with the current commit when building the wheels.
-_KONDUKTOR_COMMIT_SHA = 'c7becbf4b64d992f56aad47b69770f3345a37297'
+_KONDUKTOR_COMMIT_SHA = 'db987d020cad5afaafe4fd3c7ff858fd9bcb264d'
 os.makedirs(os.path.expanduser('~/.konduktor'), exist_ok=True)
 
 
@@ -48,5 +48,5 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev0.1.0.dev20250414104837'
+__version__ = '1.0.0.dev0.1.0.dev20250415104823'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/konduktor/adaptors/common.py

@@ -14,9 +14,12 @@
 
 import functools
 import importlib
+import os
 import threading
 from typing import Any, Callable, Optional, Tuple
 
+from filelock import FileLock
+
 
 class LazyImport:
     """Lazy importer for heavy modules or cloud modules only when enabled.
@@ -86,3 +89,31 @@ def load_lazy_modules(modules: Tuple[LazyImport, ...]):
         return wrapper
 
     return decorator
+
+
+class LockedClientProxy:
+    """Proxy for GCP client that locks access to the client."""
+
+    def __init__(
+        self,
+        client,
+        lock_path=os.path.expanduser('~/.konduktor/gcs_storage.lock'),
+        timeout=10,
+    ):
+        self._client = client
+        self._lock = FileLock(lock_path, timeout=timeout)
+
+    def __getattr__(self, attr):
+        target = getattr(self._client, attr)
+
+        if callable(target):
+
+            @functools.wraps(target)
+            def locked_method(*args, **kwargs):
+                with self._lock:
+                    return target(*args, **kwargs)
+
+            return locked_method
+        else:
+            # Attribute (not method) access just passes through
+            return target

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/konduktor/adaptors/gcp.py

@@ -14,9 +14,6 @@
 
 import json
 import os
-from functools import wraps
-
-from filelock import FileLock
 
 from konduktor.adaptors import common
 
@@ -29,33 +26,7 @@ googleapiclient = common.LazyImport(
 google = common.LazyImport('google', import_error_message=_IMPORT_ERROR_MESSAGE)
 _LAZY_MODULES = (google, googleapiclient)
 
-
-class LockedClientProxy:
-    """Proxy for GCP client that locks access to the client."""
-
-    def __init__(
-        self,
-        client,
-        lock_path=os.path.expanduser('~/.konduktor/gcs_storage.lock'),
-        timeout=10,
-    ):
-        self._client = client
-        self._lock = FileLock(lock_path, timeout=timeout)
-
-    def __getattr__(self, attr):
-        target = getattr(self._client, attr)
-
-        if callable(target):
-
-            @wraps(target)
-            def locked_method(*args, **kwargs):
-                with self._lock:
-                    return target(*args, **kwargs)
-
-            return locked_method
-        else:
-            # Attribute (not method) access just passes through
-            return target
+_LOCK_PATH = '~/.konduktor/gcs_storage.lock'
 
 
 @common.load_lazy_modules(_LAZY_MODULES)
@@ -75,7 +46,9 @@ def storage_client():
     """Helper that connects to GCS Storage Client for GCS Bucket"""
     from google.cloud import storage
 
-    return LockedClientProxy(storage.Client())
+    return common.LockedClientProxy(
+        storage.Client(), lock_path=os.path.expanduser(_LOCK_PATH)
+    )
 
 
 @common.load_lazy_modules(_LAZY_MODULES)
@@ -83,7 +56,9 @@ def anonymous_storage_client():
     """Helper that connects to GCS Storage Client for Public GCS Buckets"""
     from google.cloud import storage
 
-    return LockedClientProxy(storage.Client.create_anonymous_client())
+    return common.LockedClientProxy(
+        storage.Client(), lock_path=os.path.expanduser(_LOCK_PATH)
+    )
 
 
 @common.load_lazy_modules(_LAZY_MODULES)

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/konduktor/backends/jobset.py

@@ -185,7 +185,9 @@ class JobsetBackend(backend.Backend):
         if not dryrun and not detach_run:
             with ux_utils.print_exception_no_traceback():
                 with rich_utils.safe_status(
-                    ux_utils.spinner_message('waiting for job to start\n')
+                    ux_utils.spinner_message(
+                        'waiting for job to start. ' 'Press Ctrl+C to detach. \n'
+                    )
                 ):
                     _wait_for_jobset_start(namespace, task.name)
                 try:

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/konduktor/data/aws/s3.py

@@ -24,11 +24,11 @@ import typing
 import uuid
 from typing import Any, Dict, List, Optional, Tuple
 
-import boto3
 import colorama
 
 from konduktor import config, logging
 from konduktor.adaptors import aws
+from konduktor.adaptors.aws import boto3
 from konduktor.data import constants, data_utils, storage_utils
 from konduktor.utils import (
     annotations,
@@ -40,6 +40,9 @@ from konduktor.utils import (
     ux_utils,
 )
 
+if typing.TYPE_CHECKING:
+    from konduktor.adaptors.aws import boto3
+
 logger = logging.get_logger(__name__)
 
 # Maximum number of concurrent rsync upload processes
@@ -54,6 +57,8 @@ DEFAULT_AWS_CREDENTIALS_DIR = '~/.aws/'
 DEFAULT_AWS_CREDENTIAL_PATH = '~/.aws/credentials'
 DEFAULT_AWS_CONFIG_PATH = '~/.aws/config'
 
+_LOCK_PATH = '~/.konduktor/s3_storage.lock'
+
 
 class AWSIdentityType(enum.Enum):
     """AWS identity type.
@@ -148,7 +153,7 @@ class S3Store(storage_utils.AbstractStore):
         sync_on_reconstruction: Optional[bool] = True,
         _bucket_sub_path: Optional[str] = None,
     ):
-        self.client: 'boto3.client.Client'
+        self.client: 'boto3.client.Client'  # type: ignore[name-defined]
         self.bucket: 'constants.StorageHandle'
         if region in self._CUSTOM_ENDPOINT_REGIONS:
             logger.warning(
@@ -908,7 +913,7 @@ class S3Store(storage_utils.AbstractStore):
                 )
 
         try:
-            s3 = boto3.client('s3')
+            s3 = aws.client('s3')
 
             suffix = uuid.uuid4().hex[:6]
             test_bucket = f'konduktor-check-s3-{int(time.time())}-{suffix}'
@@ -1039,7 +1044,7 @@ class S3Store(storage_utils.AbstractStore):
             )
 
         try:
-            identity = boto3.client('sts').get_caller_identity()
+            identity = aws.client('sts').get_caller_identity()
             logger.info(
                 f"Using AWS credentials for ARN: {identity['Arn']} "
                 f"(UserId: {identity['UserId']}, Account: {identity['Account']})"

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/konduktor/utils/accelerator_registry.py

@@ -4,6 +4,7 @@ _ACCELERATORS = [
     'A100',
     'A100-80GB',
     'H100',
+    'H200',
     'L40S',
 ]
 

{konduktor_nightly-0.1.0.dev20250414104837 → konduktor_nightly-0.1.0.dev20250415104823}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "konduktor-nightly"
-version = "0.1.0.dev20250414104837"
+version = "0.1.0.dev20250415104823"
 description = "GPU Cluster Health Management"
 packages = [
     {include = "konduktor"}