konduktor-nightly 0.1.0.dev20250813105033__tar.gz → 0.1.0.dev20250814105131__tar.gz

{konduktor_nightly-0.1.0.dev20250813105033 → konduktor_nightly-0.1.0.dev20250814105131}/PKG-INFO

@@ -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 → konduktor_nightly-0.1.0.dev20250814105131}/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_nightly-0.1.0.dev20250813105033 → konduktor_nightly-0.1.0.dev20250814105131}/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_nightly-0.1.0.dev20250813105033 → konduktor_nightly-0.1.0.dev20250814105131}/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 → konduktor_nightly-0.1.0.dev20250814105131}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "konduktor-nightly"
-version = "0.1.0.dev20250813105033"
+version = "0.1.0.dev20250814105131"
 description = "GPU Cluster Health Management"
 packages = [
     {include = "konduktor"}