konduktor-nightly 0.1.0.dev20250623105113__py3-none-any.whl → 0.1.0.dev20250625105100__py3-none-any.whl

konduktor/__init__.py

@@ -14,7 +14,7 @@ __all__ = [
 ]
 
 # Replaced with the current commit when building the wheels.
-_KONDUKTOR_COMMIT_SHA = 'd7cf0201d5498c86113628b414d2112670a356d6'
+_KONDUKTOR_COMMIT_SHA = '249f48b15859d1a210ca9b28e6e9cd85ac19ac68'
 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.dev20250623105113'
+__version__ = '1.0.0.dev0.1.0.dev20250625105100'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

konduktor/backends/jobset_utils.py

@@ -10,6 +10,7 @@ from datetime import datetime, timezone
 from typing import Any, Dict, Optional, Tuple
 from urllib.parse import urlparse
 
+import click
 import colorama
 
 if typing.TYPE_CHECKING:
@@ -272,8 +273,10 @@ def create_pod_spec(task: 'konduktor.Task') -> Dict[str, Any]:
     )
 
     # validate pod spec using json schema
-    # schema: https://github.com/instrumenta/kubernetes-json-schema/blob/master/v1.9.8-standalone-strict/podspec.json
-    validator.validate_pod_spec(pod_config['kubernetes']['pod_config']['spec'])
+    try:
+        validator.validate_pod_spec(pod_config['kubernetes']['pod_config']['spec'])
+    except ValueError as e:
+        raise click.UsageError(str(e))
 
     return pod_config
 

konduktor/logging.py

@@ -17,6 +17,7 @@ _FORMAT = '[%(levelname).1s %(asctime)s %(filename)s:%(lineno)d] %(message)s'
 _DATE_FORMAT = '%m-%d %H:%M:%S'
 
 _logging_config = threading.local()
+_log_path = None
 
 
 class NewLineFormatter(logging.Formatter):
@@ -51,19 +52,36 @@ def set_logging_level(logger: str, level: int):
 
 
 def get_logger(name: str):
-    # Determine the logging level based on the KONDUKTOR_DEBUG environment variable
-    log_level = logging.INFO
-    if os.environ.get('KONDUKTOR_DEBUG', None) == '1':
-        log_level = logging.DEBUG
+    global _log_path
 
-    # Configure the logger
     logger = logging.getLogger(name)
-    if not logger.hasHandlers():  # Check if the logger already has handlers
-        logger.setLevel(log_level)
+
+    # Avoid duplicate handlers
+    if logger.hasHandlers():
+        return logger
+
+    logger.setLevel(logging.DEBUG)  # Always capture all levels internally
+
+    # --- File logging: Always enabled ---
+    if not _log_path:
+        log_dir = os.path.expanduser('~/.konduktor/logs')
+        os.makedirs(log_dir, exist_ok=True)
+        timestamp = datetime.now().strftime('%Y%m%d-%H%M%S')
+        _log_path = os.path.join(log_dir, f'konduktor-logs-{timestamp}.log')
+        print(f'Log file: {_log_path}')
+
+    fh = logging.FileHandler(_log_path)
+    fh.setLevel(logging.DEBUG)
+    fh.setFormatter(FORMATTER)
+    logger.addHandler(fh)
+
+    # --- Console logging: DEBUG level only if KONDUKTOR_DEBUG=1 ---
+    if os.environ.get('KONDUKTOR_DEBUG') == '1':
         ch = logging.StreamHandler()
-        ch.setLevel(log_level)
+        ch.setLevel(logging.DEBUG)
         ch.setFormatter(FORMATTER)
         logger.addHandler(ch)
+
     logger.propagate = False
     return logger
 

konduktor/resource.py

@@ -426,7 +426,7 @@ class Resources:
             common_utils.validate_schema(
                 config['job_config'],
                 schemas.get_job_schema(),
-                'Invalid job config YAML',
+                'Invalid job config YAML: ',
             )
 
         def _override_resources(

konduktor/utils/validator.py

@@ -14,12 +14,16 @@ import requests
 from colorama import Fore, Style
 from filelock import FileLock
 
+from konduktor import logging
+
 SCHEMA_VERSION = 'v1.32.0-standalone-strict'
 SCHEMA_URL = f'https://raw.githubusercontent.com/yannh/kubernetes-json-schema/master/{SCHEMA_VERSION}/podspec.json'
 SCHEMA_CACHE_PATH = Path.home() / '.konduktor/schemas/podspec.json'
 SCHEMA_LOCK_PATH = SCHEMA_CACHE_PATH.with_suffix('.lock')
 CACHE_MAX_AGE_SECONDS = 86400  # 24 hours
 
+logger = logging.get_logger(__name__)
+
 
 def case_insensitive_enum(validator, enums, instance, schema):
     del validator, schema  # Unused.
@@ -65,6 +69,16 @@ def validate_pod_spec(pod_spec: dict) -> None:
         return
 
     formatted = [
+        f'- {error.message}'
+        + (f" at path: {' → '.join(str(p) for p in error.path)}" if error.path else '')
+        for error in errors
+    ]
+
+    # Clean log
+    logger.debug('Invalid k8s pod spec/config:\n%s', '\n'.join(formatted))
+
+    # Color only in CLI
+    formatted_colored = [
         f'{Fore.RED}- {error.message}'
         + (f" at path: {' → '.join(str(p) for p in error.path)}" if error.path else '')
         + Style.RESET_ALL
@@ -72,7 +86,6 @@ def validate_pod_spec(pod_spec: dict) -> None:
     ]
 
     raise ValueError(
-        f'\n{Fore.RED}Invalid k8s pod spec/config: \
-        {Style.RESET_ALL}\n'
-        + '\n'.join(formatted)
+        f'\n{Fore.RED}Invalid k8s pod spec/config: {Style.RESET_ALL}\n'
+        + '\n'.join(formatted_colored)
     )

{konduktor_nightly-0.1.0.dev20250623105113.dist-info → konduktor_nightly-0.1.0.dev20250625105100.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250623105113
+Version: 0.1.0.dev20250625105100
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu
@@ -68,9 +68,6 @@ resources:
     kueue.x-k8s.io/queue-name: user-queue
     kueue.x-k8s.io/priority-class: low-priority
 
-setup: |
-  pip install torch torchvision
-
 run: |
   torchrun \
   --nproc_per_node 8 \

{konduktor_nightly-0.1.0.dev20250623105113.dist-info → konduktor_nightly-0.1.0.dev20250625105100.dist-info}/RECORD

@@ -1,4 +1,4 @@
-konduktor/__init__.py,sha256=PHCpQXy4kf7PS0Hrl2ofcVkSL6fHhE4vWV6zWnkwi3A,1540
+konduktor/__init__.py,sha256=uc7vr7keZUYeHbZs1BjiGOtnD8DknRUw5AdryM8F9lg,1540
 konduktor/adaptors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 konduktor/adaptors/aws.py,sha256=s47Ra-GaqCQibzVfmD0pmwEWHif1EGO5opMbwkLxTCU,8244
 konduktor/adaptors/common.py,sha256=ZIqzjx77PIHUwpjfAQ1uX8B2aX78YMuGj4Bppd-MdyM,4183
@@ -7,7 +7,7 @@ konduktor/authentication.py,sha256=_mVy3eqoKohicHostFiGwG1-2ybxP-l7ouofQ0LRlCY,4
 konduktor/backends/__init__.py,sha256=1Q6sqqdeMYarpTX_U-QVywJYf7idiUTRsyP-E4BQSOw,129
 konduktor/backends/backend.py,sha256=qh0bp94lzoTYZkzyQv2-CVrB5l91FkG2vclXg24UFC0,2910
 konduktor/backends/jobset.py,sha256=UdhwAuZODLMbLY51Y2zOBsh6wg4Pb84oHVvUKzx3Z2w,8434
-konduktor/backends/jobset_utils.py,sha256=mOjK3oFgmNacpg956r0qtR7cdZ0PPzXYVVuIr3QqKuI,22170
+konduktor/backends/jobset_utils.py,sha256=esudKceD7iFjzYokRGEkAQd21GxsHvgQHTEBMU0rsdM,22145
 konduktor/check.py,sha256=JennyWoaqSKhdyfUldd266KwVXTPJpcYQa4EED4a_BA,7569
 konduktor/cli.py,sha256=qtktD8N17IRC5MYEdaE0o3pv8EI36cvyyQkYUFi5_nQ,35590
 konduktor/config.py,sha256=J50JxC6MsXMnlrJPXdDUMr38C89xvOO7mR8KJ6fyils,15520
@@ -63,12 +63,12 @@ konduktor/data/storage.py,sha256=o2So-bY9glvgbGdoN7AQNYmNnvGf1AUDPpImtadRL90,352
 konduktor/data/storage_utils.py,sha256=n4GivkN0KMqmyOTDznF0Z-hzsJvm7KCEh5i5HgFAT-4,20806
 konduktor/execution.py,sha256=NCl2bgo5p1ZZl8HLaXT-juAe9PXr-iCJv0md2sT7A20,18395
 konduktor/kube_client.py,sha256=lC-U_1hLRG3mDN8tBxYc4VZ3BS5BzKm8hlt-lE3505A,5938
-konduktor/logging.py,sha256=mBCoCTNhDEkUxd4tsse4mw-aVzSGohhXYf16ViR0ch4,2722
+konduktor/logging.py,sha256=mA1JCCWPCqQMRqEpE4l6D6vOYdbtbQXr0BuEk9RR790,3177
 konduktor/manifests/controller_deployment.yaml,sha256=6p3oSLkEVONZsvKZGqVop0Dhn4bo3lrigRmhf8NXBHE,1730
 konduktor/manifests/dashboard_deployment.yaml,sha256=xJLd4FbPMAosI0fIv5_8y7dV9bw0Vsf81l-w4MB_aU8,2837
 konduktor/manifests/dmesg_daemonset.yaml,sha256=pSWt7YOeTYjS0l0iki1fvHOs7MhY-sH-RQfVW6JJyno,1391
 konduktor/manifests/pod_cleanup_controller.yaml,sha256=hziL1Ka1kCAEL9R7Tjvpb80iw1vcq9_3gwHCu75Bi0A,3939
-konduktor/resource.py,sha256=Fg4kon7jQ9xDo9Iz8Q0J8doIRmTkSwIhYXLH6jbtRO8,19610
+konduktor/resource.py,sha256=nHgPWXCbWj5sWyslNngrFypMN1K0Dksb0yHbJqWaei8,19612
 konduktor/task.py,sha256=ofwd8WIhfD6C3ThLcv6X3GUzQHyZ6ddjUagE-umF4K0,35207
 konduktor/templates/jobset.yaml.j2,sha256=rdURknodtgLp4zoA2PX86Nn4wPpi3tr5l4IG55aWBRg,1059
 konduktor/templates/pod.yaml.j2,sha256=SlK6XKSwjuFJtBimlrUiFTcx7G_00XDtEopIKXBg5SI,16635
@@ -90,9 +90,9 @@ konduktor/utils/rich_utils.py,sha256=ycADW6Ij3wX3uT8ou7T8qxX519RxlkJivsLvUahQaJo
 konduktor/utils/schemas.py,sha256=VGPERAso2G4sVAznsJ80qT2Q-I_EFxXw6Rfcw-vkYgQ,16535
 konduktor/utils/subprocess_utils.py,sha256=WoFkoFhGecPR8-rF8WJxbIe-YtV94LXz9UG64SDhCY4,9448
 konduktor/utils/ux_utils.py,sha256=czCwiS1bDqgeKtzAJctczpLwFZzAse7WuozdvzEFYJ4,7437
-konduktor/utils/validator.py,sha256=K-eEmwq4qgYcOhMv6SqgIPcrpWqDusH0f8EBkzv827Q,2429
-konduktor_nightly-0.1.0.dev20250623105113.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
-konduktor_nightly-0.1.0.dev20250623105113.dist-info/METADATA,sha256=NnIKhdyHu3zwwBVBwnW7PM46sJHJbbYrPrpS5QuNPMY,4289
-konduktor_nightly-0.1.0.dev20250623105113.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-konduktor_nightly-0.1.0.dev20250623105113.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
-konduktor_nightly-0.1.0.dev20250623105113.dist-info/RECORD,,
+konduktor/utils/validator.py,sha256=uCRlScO1NYxsbTNKY9dkoqvlO8S0ISIIB8XmX2ItcO8,2793
+konduktor_nightly-0.1.0.dev20250625105100.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
+konduktor_nightly-0.1.0.dev20250625105100.dist-info/METADATA,sha256=9rBtjMCrltjV0GGyTPOpp_bwVQdC1uO3RvgScWUNlLo,4247
+konduktor_nightly-0.1.0.dev20250625105100.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+konduktor_nightly-0.1.0.dev20250625105100.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
+konduktor_nightly-0.1.0.dev20250625105100.dist-info/RECORD,,