konduktor-nightly 0.1.0.dev20250813105033__py3-none-any.whl → 0.1.0.dev20250814105131__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 = 'f4ba2084fac1c1030245b475323f4f3a57fd3fa3'
+_KONDUKTOR_COMMIT_SHA = '0301ec982f1bd0a04115661b6e139b2964c329b8'
 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.dev20250813105033'
+__version__ = '1.0.0.dev0.1.0.dev20250814105131'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

konduktor/cli.py

@@ -693,11 +693,11 @@ def status(
 @click.option(
     '--num-lines',
     '--num_lines' '-n',
-    default=1000,
+    default=-1,
     type=int,
     help=(
         'The number of lines to display from the end of the log file. '
-        'Default is 1000.'
+        'Default is -1 (no limit).'
     ),
 )
 @click.option(
@@ -732,18 +732,19 @@ def logs(
     # Verify the job exists before attempting to tail logs
     # TODO(asaiacai): unify the 404 logic under jobset_utils
     try:
-        jobset_response = jobset_utils.get_jobset(namespace, job_id)
+        _ = jobset_utils.get_jobset(namespace, job_id)
     except jobset_utils.JobNotFoundError:
-        raise click.UsageError(
-            f"Job '{job_id}' not found in namespace "
-            f"'{namespace}'. Check your jobs with "
+        click.secho(
+            f"Job '{job_id}' not found in namespace '{namespace}'. "
+            f'Job may have been `konduktor down`. '
+            f'Check your jobs with '
             f'{colorama.Style.BRIGHT}`konduktor status`'
-            f'{colorama.Style.RESET_ALL}.'
+            f'{colorama.Style.RESET_ALL}.',
+            fg='yellow',
         )
 
-    assert isinstance(jobset_response, dict), f'jobset_response: {jobset_response}'
     log_utils.tail_logs(
-        jobset_response,
+        job_id,
         worker_id=node_rank,
         follow=follow,
         num_logs=num_lines,

konduktor/utils/log_utils.py

@@ -337,12 +337,11 @@ def tail_loki_logs_ws(
 
 
 def tail_vicky_logs(
-    jobset_response: Dict[str, Any],
+    job_name: str,
     worker_id: int = 0,
     num_logs: int = -1,
     follow: bool = True,
 ):
-    job_name = jobset_response['metadata']['name']
     context = kubernetes_utils.get_current_kube_config_context_name()
     namespace = kubernetes_utils.get_kube_config_context_namespace(context)
     query: Dict[str, Any] = {}
@@ -357,11 +356,11 @@ def tail_vicky_logs(
         query = {'limit': num_logs}
     if follow:
         logger.info(
-            'No end time found, tailing logs from 1 hour ago. '
+            'Tailing logs from 1 hour ago. '
             'If logs come up empty, there might be logs just earlier '
             'than the past hour, check Grafana or use:\n'
             f'{colorama.Style.BRIGHT}{colorama.Fore.YELLOW}'
-            f'`konduktor tail --no-follow {job_name}`'
+            f'`konduktor logs --no-follow {job_name}`'
             f'{colorama.Style.RESET_ALL}'
         )
         query['start_offset'] = '1h'
@@ -379,7 +378,7 @@ def tail_vicky_logs(
             vicky_url = f'http://localhost:{port}/select/logsql/tail'
         else:
             vicky_url = f'http://localhost:{port}/select/logsql/query'
-            timeout = 1
+            timeout = 5
         logger.debug(f'Vicky URL: {vicky_url}')
 
         try:
@@ -418,17 +417,16 @@ def tail_vicky_logs(
 
 
 def tail_logs(
-    jobset_response: Dict[str, Any],
+    job_name: str,
     worker_id: int = 0,
     num_logs: int = 1000,
     follow: bool = True,
 ):
-    job_name = jobset_response['metadata']['name']
     logs_backend = config.get_nested(('logs', 'backend'), None)
     if logs_backend == LogBackend.VICTORIA:
-        tail_vicky_logs(jobset_response, worker_id, num_logs, follow)
+        tail_vicky_logs(job_name, worker_id, num_logs, follow)
     elif logs_backend == LogBackend.LOKI:
         tail_loki_logs_ws(job_name, worker_id, num_logs, follow)
     else:
         logger.info('Defaulting to VictoriaLogs')
-        tail_vicky_logs(jobset_response, worker_id, num_logs, follow)
+        tail_vicky_logs(job_name, worker_id, num_logs, follow)

{konduktor_nightly-0.1.0.dev20250813105033.dist-info → konduktor_nightly-0.1.0.dev20250814105131.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250813105033
+Version: 0.1.0.dev20250814105131
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250813105033.dist-info → konduktor_nightly-0.1.0.dev20250814105131.dist-info}/RECORD

@@ -1,4 +1,4 @@
-konduktor/__init__.py,sha256=_hY0EbT0p1RyIpToDmvNWpK2QTJNNplI92T0AlW7tck,1574
+konduktor/__init__.py,sha256=XyFuy_BPt7mWYrkEh_gtnqtClkMLgAdFkP4SpjNbw9s,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
@@ -13,7 +13,7 @@ konduktor/backends/jobset.py,sha256=E9THHmcpxTohsx6Goi9mKF4dy_mYpR2DHloSwGVr9jA,
 konduktor/backends/jobset_utils.py,sha256=7fB8X4b2Q5BKFCIGME72dyeCfi-EemoMeJVnwtzcjq4,25184
 konduktor/backends/pod_utils.py,sha256=Jfv_CY8suF0e7QEaeQiNRRxRnOueLgPR8SfLEO7lnwc,15260
 konduktor/check.py,sha256=JennyWoaqSKhdyfUldd266KwVXTPJpcYQa4EED4a_BA,7569
-konduktor/cli.py,sha256=YD9gMH2ZJykdfrHvzY-DkPQgD-cltahEW141wdI8eiI,56674
+konduktor/cli.py,sha256=YW9vNlbayDIcHk5D8-qsVGwoOdRV1o9kFmF2jtBf3vE,56641
 konduktor/config.py,sha256=9upqgCCYvcu6fKw7tovEYC1MWTkAAir0_WHPdayylbI,15536
 konduktor/constants.py,sha256=T3AeXXxuQHINW_bAWyztvDeS8r4g8kXBGIwIq13cys0,1814
 konduktor/controller/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -90,15 +90,15 @@ konduktor/utils/env_options.py,sha256=T41Slzf4Mzl-n45CGXXqdy2fCrYhPNZQ7RP5vmnN4x
 konduktor/utils/exceptions.py,sha256=5IFnN5bIUSBJv4KRRrCepk5jyY9EG5vWWQqbjCmP3NU,6682
 konduktor/utils/kubernetes_enums.py,sha256=SabUueF6Bpzbpa57gyH5VB65xla2N9l8CZmAeYTfGmM,176
 konduktor/utils/kubernetes_utils.py,sha256=7RThCOiyaALRqbwHZ40qMnBsbAgt669k0NHkxtfx7Bs,26205
-konduktor/utils/log_utils.py,sha256=xg5-NM1l3oodRTkiKihuzwe82g7XnfTzprFPndSF1A8,17032
+konduktor/utils/log_utils.py,sha256=jxm9ovPcJPOGfd2wOwPDEThRO25ETIV5a1DmAfLhqJc,16861
 konduktor/utils/loki_utils.py,sha256=h2ZvZQr1nE_wXXsKsGMjhG2s2MXknNd4icydTR_ruKU,3539
 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=7-Lt3QbDVvBQUli5_U9lOdXKeC-ip8rZBpO9gQ6vPJw,7955
 konduktor/utils/validator.py,sha256=5C1kE57Eyj1OPnAbvojqMNHHtf5fnl47FK_vEttd8aw,4331
-konduktor_nightly-0.1.0.dev20250813105033.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
-konduktor_nightly-0.1.0.dev20250813105033.dist-info/METADATA,sha256=SG5D0d2YiNKbLh9yKpMif5NucNJgwSYp-lU55QXL00c,4247
-konduktor_nightly-0.1.0.dev20250813105033.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-konduktor_nightly-0.1.0.dev20250813105033.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
-konduktor_nightly-0.1.0.dev20250813105033.dist-info/RECORD,,
+konduktor_nightly-0.1.0.dev20250814105131.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
+konduktor_nightly-0.1.0.dev20250814105131.dist-info/METADATA,sha256=I57VbLFYr7rStkmzKNbH5QUOmd0JvX1CDdENYtCjPJs,4247
+konduktor_nightly-0.1.0.dev20250814105131.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+konduktor_nightly-0.1.0.dev20250814105131.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
+konduktor_nightly-0.1.0.dev20250814105131.dist-info/RECORD,,