konduktor-nightly 0.1.0.dev20250809104842__py3-none-any.whl → 0.1.0.dev20250811105223__py3-none-any.whl

konduktor/__init__.py

@@ -11,7 +11,7 @@ from konduktor.task import Task
 __all__ = ['launch', 'Resources', 'Task', 'Serving']
 
 # Replaced with the current commit when building the wheels.
-_KONDUKTOR_COMMIT_SHA = '0f0b36c3a67aa7c60d6cb33240631b7c8ccaed03'
+_KONDUKTOR_COMMIT_SHA = '92fe69bd3f29e7b191de663c598dfcf10738f87a'
 os.makedirs(os.path.expanduser('~/.konduktor'), exist_ok=True)
 
 
@@ -45,5 +45,5 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev0.1.0.dev20250809104842'
+__version__ = '1.0.0.dev0.1.0.dev20250811105223'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

konduktor/backends/jobset_utils.py

@@ -232,8 +232,12 @@ def stop_jobset(namespace: str, job_name: str) -> Optional[Dict[str, Any]]:
             'spec': {'suspend': True},
             'metadata': {
                 'annotations': {
-                    backend_constants.STOP_USERID_LABEL: common_utils.user_and_hostname_hash(),
-                    backend_constants.STOP_USERNAME_LABEL: common_utils.get_cleaned_username(),
+                    backend_constants.STOP_USERID_LABEL: (
+                        common_utils.user_and_hostname_hash()
+                    ),
+                    backend_constants.STOP_USERNAME_LABEL: (
+                        common_utils.get_cleaned_username()
+                    ),
                 }
             },
         }

konduktor/cli.py

@@ -855,16 +855,23 @@ def launch(
             'Serving information detected. Use '
             '`konduktor serve launch` instead for serving.'
         )
+    try:
+        _launch_with_confirm(
+            task,
+            dryrun=dryrun,
+            detach_run=detach_run,
+            no_confirm=yes,
+            serving=bool(task.serving),
+        )
+    except KeyboardInterrupt:
+        click.secho(
+            f'Detaching... manage your job {task.name} with the following commands:',
+            fg='yellow',
+            bold=True,
+        )
 
-    job_name = _launch_with_confirm(
-        task,
-        dryrun=dryrun,
-        detach_run=detach_run,
-        no_confirm=yes,
-        serving=bool(task.serving),
-    )
     click.secho(
-        ux_utils.command_hint_messages(ux_utils.CommandHintType.JOB, job_name),
+        ux_utils.command_hint_messages(ux_utils.CommandHintType.JOB, task.name),
         fg='green',
         bold=True,
     )
@@ -1139,6 +1146,12 @@ def stop(
     for job in track(jobs, description='Suspending job(s)...'):
         jobset_utils.stop_jobset(namespace, job)
 
+    click.secho(
+        ux_utils.command_hint_messages(ux_utils.CommandHintType.JOB_STOP, jobs),
+        fg='green',
+        bold=True,
+    )
+
 
 @cli.command(cls=_DocumentedCodeCommand)
 @click.argument(
@@ -1836,7 +1849,7 @@ def main():
         return cli(standalone_mode=False)
     except click.exceptions.Abort:
         click.secho('Detaching...', fg='yellow', bold=True)
-        return
+        return None
 
 
 if __name__ == '__main__':

konduktor/logging.py

@@ -75,12 +75,14 @@ def get_logger(name: str):
     fh.setFormatter(FORMATTER)
     logger.addHandler(fh)
 
-    # --- Console logging: DEBUG level only if KONDUKTOR_DEBUG=1 ---
+    # --- Console logging: INFO level by default, DEBUG if KONDUKTOR_DEBUG=1 ---
+    ch = logging.StreamHandler()
     if os.environ.get('KONDUKTOR_DEBUG') == '1':
-        ch = logging.StreamHandler()
         ch.setLevel(logging.DEBUG)
-        ch.setFormatter(FORMATTER)
-        logger.addHandler(ch)
+    else:
+        ch.setLevel(logging.INFO)
+    ch.setFormatter(FORMATTER)
+    logger.addHandler(ch)
 
     logger.propagate = False
     return logger

konduktor/utils/ux_utils.py

@@ -6,7 +6,7 @@ import os
 import sys
 import traceback
 import typing
-from typing import Callable, Optional, Union
+from typing import Callable, List, Optional, Union
 
 import colorama
 import rich.console as rich_console
@@ -196,27 +196,41 @@ def spinner_message(
 
 class CommandHintType(enum.Enum):
     JOB = 'JOB'
+    JOB_STOP = 'JOB_STOP'
 
 
-def command_hint_messages(hint_type: CommandHintType, job_id: str) -> str:
+def command_hint_messages(
+    hint_type: CommandHintType,
+    job_id: Union[str, List[str]],
+) -> str:
     """Gets the command hint messages for the given job id."""
+    hint_str = '\n📋 Useful Commands'
     if hint_type == CommandHintType.JOB:
         job_hint_str = (
             f'\nJob ID: {job_id}'
-            f'\n{INDENT_SYMBOL}To cancel the job:\t\t'
-            f'{BOLD}konduktor down {job_id} {RESET_BOLD}'
             f'\n{INDENT_SYMBOL}To stream job logs:\t\t'
             f'{BOLD}konduktor logs {job_id} {RESET_BOLD}'
             f'\n{INDENT_SYMBOL}To list all jobs:\t\t'
             f'{BOLD}konduktor status{RESET_BOLD}'
+            f'\n{INDENT_SYMBOL}To suspend the job:\t\t'
+            f'{BOLD}konduktor stop {job_id} {RESET_BOLD}'
+            f'\n{INDENT_SYMBOL}{colorama.Fore.RED}To delete the job:\t\t'
+            f'{BOLD}konduktor down {job_id} {RESET_BOLD}{colorama.Style.RESET_ALL}'
         )
-        hint_str = '\n📋 Useful Commands'
         hint_str += f'{job_hint_str}'
-        if config.get_nested(('tailscale', 'secret_name'), None) is not None:
-            hint_str += (
-                f'\n{INDENT_SYMBOL}To tailscale ssh:\t\t'
-                f'{BOLD}ssh root@{job_id}-workers-0-0 {RESET_BOLD}'
-            )
-        return hint_str
+    elif hint_type == CommandHintType.JOB_STOP:
+        assert isinstance(job_id, list), 'job_id must be a list of strings'
+        job_ids_str = ' '.join(job_id)
+        hint_str += (
+            f'\n{INDENT_SYMBOL}To resume the following jobs:\t\t'
+            f'{BOLD}konduktor start {job_ids_str} {RESET_BOLD}'
+        )
     else:
         raise ValueError(f'Invalid hint type: {hint_type}')
+
+    if config.get_nested(('tailscale', 'secret_name'), None) is not None:
+        hint_str += (
+            f'\n{INDENT_SYMBOL}To tailscale ssh:\t\t'
+            f'{BOLD}ssh root@{job_id}-workers-0-0 {RESET_BOLD}'
+        )
+    return hint_str

{konduktor_nightly-0.1.0.dev20250809104842.dist-info → konduktor_nightly-0.1.0.dev20250811105223.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250809104842
+Version: 0.1.0.dev20250811105223
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250809104842.dist-info → konduktor_nightly-0.1.0.dev20250811105223.dist-info}/RECORD

@@ -1,4 +1,4 @@
-konduktor/__init__.py,sha256=bTZfonclg2KPH9idiGwsQys1xaoPFVgSruIPRdy8GSQ,1574
+konduktor/__init__.py,sha256=DYiQ-TfOdYUqIzT8psdjnWcjtjMI4sbldAghArAY5e0,1574
 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
@@ -10,10 +10,10 @@ konduktor/backends/constants.py,sha256=NfdhY1PQnewvDCjgRKXj6EZDcVH8k_0GGxnMo7w6H
 konduktor/backends/deployment.py,sha256=EHfB2uLeKFQ3maek9tx6XL4_sjQ-ax59DZA79Q3EkVs,5519
 konduktor/backends/deployment_utils.py,sha256=VGuL01rKe7p7PoVRI_cP4tiZRxHZ13nnTMG-bmDf7P0,28975
 konduktor/backends/jobset.py,sha256=OwgDog9nH-FoUmNU_H--C3U5jx70reTKL1l849M1k5A,8430
-konduktor/backends/jobset_utils.py,sha256=YPbGxbM9FIPNLlvu3_189iGDopWrGLqL_kJO17McRUU,24567
+konduktor/backends/jobset_utils.py,sha256=O983a78D411go_F0K2mijZAE1dXAFF7i6aQ7rOrfH7A,24663
 konduktor/backends/pod_utils.py,sha256=Jfv_CY8suF0e7QEaeQiNRRxRnOueLgPR8SfLEO7lnwc,15260
 konduktor/check.py,sha256=JennyWoaqSKhdyfUldd266KwVXTPJpcYQa4EED4a_BA,7569
-konduktor/cli.py,sha256=lEZmfrswuxMAyU5hmndMHqk4GkJyohk_TOHBx-0h90M,56316
+konduktor/cli.py,sha256=9S6DEsK_qlD34UM6CwFah0FmJgQ4lVaV-LViKp9fJ6o,56687
 konduktor/config.py,sha256=9upqgCCYvcu6fKw7tovEYC1MWTkAAir0_WHPdayylbI,15536
 konduktor/constants.py,sha256=T3AeXXxuQHINW_bAWyztvDeS8r4g8kXBGIwIq13cys0,1814
 konduktor/controller/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -67,7 +67,7 @@ konduktor/data/storage.py,sha256=o2So-bY9glvgbGdoN7AQNYmNnvGf1AUDPpImtadRL90,352
 konduktor/data/storage_utils.py,sha256=n4GivkN0KMqmyOTDznF0Z-hzsJvm7KCEh5i5HgFAT-4,20806
 konduktor/execution.py,sha256=d0EP79iSrW2uFsoqn0YV_4kgIupPIqpMOParXx0y3kg,18519
 konduktor/kube_client.py,sha256=WELs9jClRW9r-imNJF3gJi3Z7ygkFDnYDmMXu5nJhEM,6213
-konduktor/logging.py,sha256=mA1JCCWPCqQMRqEpE4l6D6vOYdbtbQXr0BuEk9RR790,3177
+konduktor/logging.py,sha256=xtcCdnecmC3rqMTyunK-klQRINojI7NI4Apag78i9jM,3221
 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
@@ -95,10 +95,10 @@ konduktor/utils/loki_utils.py,sha256=h2ZvZQr1nE_wXXsKsGMjhG2s2MXknNd4icydTR_ruKU
 konduktor/utils/rich_utils.py,sha256=ycADW6Ij3wX3uT8ou7T8qxX519RxlkJivsLvUahQaJo,3583
 konduktor/utils/schemas.py,sha256=tBrKhnkfn9uKDYdlb4L2KgooW-muuhww7U8fu9zX-ms,18336
 konduktor/utils/subprocess_utils.py,sha256=WoFkoFhGecPR8-rF8WJxbIe-YtV94LXz9UG64SDhCY4,9448
-konduktor/utils/ux_utils.py,sha256=czCwiS1bDqgeKtzAJctczpLwFZzAse7WuozdvzEFYJ4,7437
+konduktor/utils/ux_utils.py,sha256=7-Lt3QbDVvBQUli5_U9lOdXKeC-ip8rZBpO9gQ6vPJw,7955
 konduktor/utils/validator.py,sha256=5C1kE57Eyj1OPnAbvojqMNHHtf5fnl47FK_vEttd8aw,4331
-konduktor_nightly-0.1.0.dev20250809104842.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
-konduktor_nightly-0.1.0.dev20250809104842.dist-info/METADATA,sha256=cOmfOuxpidR03qJMivceD6nkUWRwmNzm_D3OVKGF2q8,4247
-konduktor_nightly-0.1.0.dev20250809104842.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-konduktor_nightly-0.1.0.dev20250809104842.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
-konduktor_nightly-0.1.0.dev20250809104842.dist-info/RECORD,,
+konduktor_nightly-0.1.0.dev20250811105223.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
+konduktor_nightly-0.1.0.dev20250811105223.dist-info/METADATA,sha256=Gp9W_UVyGtpg-hU24Hm1DU-RKr-Hkzx10JQmBsrJMdQ,4247
+konduktor_nightly-0.1.0.dev20250811105223.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+konduktor_nightly-0.1.0.dev20250811105223.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
+konduktor_nightly-0.1.0.dev20250811105223.dist-info/RECORD,,