konduktor-nightly 0.1.0.dev20250821104804__py3-none-any.whl → 0.1.0.dev20250822104721__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 = 'eee38d922bf4c7cb8a2e6e730092dde0ae372500'
+_KONDUKTOR_COMMIT_SHA = 'face26aca22b99192e740dca4875261fcffa2a55'
 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.dev20250821104804'
+__version__ = '1.0.0.dev0.1.0.dev20250822104721'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

konduktor/backends/deployment_utils.py

@@ -47,8 +47,6 @@ _DEPLOYMENT_METADATA_LABELS = {
     'model_name_label': AIBRIX_NAME_LABEL,
 }
 
-_RUN_DURATION_ANNOTATION_KEY = 'kueue.x-k8s.io/maxRunDurationSeconds'
-
 
 def render_specs(
     task: 'konduktor.Task',
@@ -781,7 +779,12 @@ def show_status_table(namespace: str, all_users: bool):
             )
             if port_obj and port_obj.port:
                 port_str = str(port_obj.port)
-        endpoint_str = f'{ip_str}:{port_str}' if port_str else ip_str
+
+        # For vLLM deployments, don't append port since external routing is on port 80
+        if AIBRIX_NAME_LABEL in labels:
+            endpoint_str = ip_str
+        else:
+            endpoint_str = f'{ip_str}:{port_str}' if port_str else ip_str
 
         # Replicas
         ready_replicas = (

konduktor/backends/pod_utils.py

@@ -329,15 +329,6 @@ def inject_deployment_pod_metadata(
     if task.resources and task.resources.labels:
         pod_spec['metadata']['labels'].update(task.resources.labels)
 
-    # Add max run duration annotation
-    assert task.resources is not None and task.resources.labels is not None
-    maxRunDurationSeconds = task.resources.labels.get('maxRunDurationSeconds', None)
-    if not maxRunDurationSeconds:
-        raise ValueError('maxRunDurationSeconds is required')
-    pod_spec['metadata']['annotations'][_RUN_DURATION_ANNOTATION_KEY] = str(
-        maxRunDurationSeconds
-    )
-
     # Set restart policy for deployments
     pod_spec.setdefault('spec', {})
     pod_spec['spec']['restartPolicy'] = 'Always'

konduktor/cli.py

@@ -104,6 +104,7 @@ def _make_task_with_overrides(
     instance_type: Optional[str] = None,
     num_nodes: Optional[int] = None,
     max_restarts: Optional[int] = None,
+    completions: Optional[int] = None,
     image_id: Optional[str] = None,
     disk_size: Optional[int] = None,
     env: Optional[List[Tuple[str, str]]] = None,
@@ -166,7 +167,9 @@ def _make_task_with_overrides(
     if max_restarts is not None:
         assert task.resources is not None
         task.resources.job_config['max_restarts'] = max_restarts
-
+    if completions is not None:
+        assert task.resources is not None
+        task.resources.job_config['completions'] = completions
     if num_nodes is not None:
         task.num_nodes = num_nodes
     if name is not None:
@@ -215,6 +218,16 @@ _TASK_OPTIONS = [
             'supplied.'
         ),
     ),
+    click.option(
+        '--completions',
+        required=False,
+        type=int,
+        help=(
+            'Number of successful completions required. Overrides YAML.'
+            'Overrides the "completions" config in the YAML if both are '
+            'supplied.'
+        ),
+    ),
     click.option(
         '--cpus',
         default=None,
@@ -798,6 +811,7 @@ def launch(
     memory: Optional[str],
     num_nodes: Optional[int],
     max_restarts: Optional[int],
+    completions: Optional[int],
     image_id: Optional[str],
     env_file: Optional[Dict[str, str]],
     env: List[Tuple[str, str]],
@@ -822,6 +836,7 @@ def launch(
         memory=memory,
         num_nodes=num_nodes,
         max_restarts=max_restarts,
+        completions=completions,
         image_id=image_id,
         env=env,
         disk_size=disk_size,
@@ -1686,6 +1701,7 @@ def serve_launch(
     memory: Optional[str],
     num_nodes: Optional[int],
     max_restarts: Optional[int],
+    completions: Optional[int],
     image_id: Optional[str],
     env_file: Optional[Dict[str, str]],
     env: List[Tuple[str, str]],
@@ -1714,6 +1730,7 @@ def serve_launch(
         memory=memory,
         num_nodes=num_nodes,
         max_restarts=max_restarts,
+        completions=completions,
         image_id=image_id,
         env=env,
         disk_size=disk_size,

konduktor/resource.py

@@ -387,7 +387,13 @@ class Resources:
 
     def get_completions(self) -> Optional[int]:
         value = self.job_config.get('completions')
-        return int(value) if value is not None else None
+        if value is not None:
+            value = int(value)
+            if value <= 0:
+                with ux_utils.print_exception_no_traceback():
+                    raise ValueError('completions must be a positive integer')
+            return value
+        return None
 
     def get_max_restarts(self) -> Optional[int]:
         value = self.job_config.get('max_restarts')

konduktor/templates/jobset.yaml.j2

@@ -19,7 +19,7 @@ jobset:
     annotations: {}
   spec:
     ttlSecondsAfterFinished: 31536000 # 1 year (365 days)
-    {% if max_restarts %}
+    {% if max_restarts is not none %}
     failurePolicy:
       maxRestarts: {{ max_restarts }}
     {% endif %}
@@ -29,7 +29,11 @@ jobset:
         spec:
           ttlSecondsAfterFinished: 600 # 5 minutes
           parallelism: {{ num_nodes }}
+          {% if completions %}
+          completions: {{ completions }}
+          {% else %}
           completions: {{ num_nodes }}
+          {% endif %}
           backoffLimit: 0
           template: {}
           podFailurePolicy:

konduktor/utils/schemas.py

@@ -467,7 +467,8 @@ def get_job_schema():
         'additionalProperties': False,
         'properties': {
             'completions': {
-                'type': 'number',
+                'type': 'integer',
+                'minimum': 1,
             },
             'max_restarts': {
                 'type': 'integer',

{konduktor_nightly-0.1.0.dev20250821104804.dist-info → konduktor_nightly-0.1.0.dev20250822104721.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250821104804
+Version: 0.1.0.dev20250822104721
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250821104804.dist-info → konduktor_nightly-0.1.0.dev20250822104721.dist-info}/RECORD

@@ -1,4 +1,4 @@
-konduktor/__init__.py,sha256=vSTzrg0X_r9sPmZDztRtaQbXDhuOpoRTQWuv4z1wybA,1574
+konduktor/__init__.py,sha256=cRGSzBwIktM510ZBH1rt7DLOACBZjXpolNWjtbpVgR4,1574
 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
@@ -8,12 +8,12 @@ konduktor/backends/__init__.py,sha256=usWJ8HdZJEyg7MIsN8Zcz9rk9e2Lq5dWJ8dv6hCN3y
 konduktor/backends/backend.py,sha256=qh0bp94lzoTYZkzyQv2-CVrB5l91FkG2vclXg24UFC0,2910
 konduktor/backends/constants.py,sha256=nt9G9AmFCOMwO4GuKgRQSzJJuKapOmaROp4_Y0tMF5A,732
 konduktor/backends/deployment.py,sha256=EHfB2uLeKFQ3maek9tx6XL4_sjQ-ax59DZA79Q3EkVs,5519
-konduktor/backends/deployment_utils.py,sha256=VGuL01rKe7p7PoVRI_cP4tiZRxHZ13nnTMG-bmDf7P0,28975
+konduktor/backends/deployment_utils.py,sha256=Z58o9I3XDxqQpssZ3j-31n0sRRYo64TV-KNkX44Admc,29084
 konduktor/backends/jobset.py,sha256=OwgDog9nH-FoUmNU_H--C3U5jx70reTKL1l849M1k5A,8430
 konduktor/backends/jobset_utils.py,sha256=8YUIFg7mi33A5p_K9GhVdjllTdXRMj_GEg-iQ5Nj_iU,25708
-konduktor/backends/pod_utils.py,sha256=EkoyN4mMfnrwEG656J1kxPtNZQClo4X7JEyzV1F8oHg,15294
+konduktor/backends/pod_utils.py,sha256=KP_PAgsdNHFgt4Od-5gAtpifAKIL7DMBg7NJ44uqikg,14885
 konduktor/check.py,sha256=JennyWoaqSKhdyfUldd266KwVXTPJpcYQa4EED4a_BA,7569
-konduktor/cli.py,sha256=YW9vNlbayDIcHk5D8-qsVGwoOdRV1o9kFmF2jtBf3vE,56641
+konduktor/cli.py,sha256=zdcdqS0avYzxTGDuE040vt4s2IgC23uG3aVRmOK3YOY,57235
 konduktor/config.py,sha256=9upqgCCYvcu6fKw7tovEYC1MWTkAAir0_WHPdayylbI,15536
 konduktor/constants.py,sha256=T3AeXXxuQHINW_bAWyztvDeS8r4g8kXBGIwIq13cys0,1814
 konduktor/controller/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -72,11 +72,11 @@ konduktor/manifests/controller_deployment.yaml,sha256=6p3oSLkEVONZsvKZGqVop0Dhn4
 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=qQhMlI6gvTaoGfYb9NNgSrUavgNqfcYVfb9V_oC5pLE,20411
+konduktor/resource.py,sha256=epOkJngNGtSBvMqiQeO8j0C7g0PzYhwjTMbG9VapvyQ,20628
 konduktor/serving.py,sha256=sh8TPAUXg23Bkt0ByatIMdxFFqzRm18HJTEkt3wHzdo,5147
 konduktor/task.py,sha256=97iLCo62qpN9wLGNPeFw64E8k1nch7AyySY3BUXHPWY,37496
 konduktor/templates/deployment.yaml.j2,sha256=uXFjDQaimbpFdAn2RJGaIvS_PzDY136cw_L3QMjz3ZA,3452
-konduktor/templates/jobset.yaml.j2,sha256=gURWl6uQv_OLni-LFy2E7ttjGOtuRDt5Vfs4ALH7fpI,1196
+konduktor/templates/jobset.yaml.j2,sha256=NQcVeRNsTLLmTnJRnkL1vr45mSeth-b11YShXn_RoSg,1323
 konduktor/templates/pod.yaml.j2,sha256=owid7Wpo-kmQ-Gvx1zVoyG_feOiKviGDsnWLTlReXPo,19086
 konduktor/usage/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 konduktor/usage/constants.py,sha256=gCL8afIHZhO0dcxbJGpESE9sCC1cBSbeRnQ8GwNOY4M,612
@@ -93,12 +93,12 @@ konduktor/utils/kubernetes_utils.py,sha256=7RThCOiyaALRqbwHZ40qMnBsbAgt669k0NHkx
 konduktor/utils/log_utils.py,sha256=k43eGpSwIdGhNotC8w7_Hq-bbyQQTrHwMOAMct_gr9M,16978
 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/schemas.py,sha256=M4O2uISyIYYFCrdQ_b2RKQNFoo81YG-GQdmOLOiWU14,18367
 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.dev20250821104804.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
-konduktor_nightly-0.1.0.dev20250821104804.dist-info/METADATA,sha256=qQPp3F_j37Mx4Ios1of0b_961SL9jkOJd-5sZ-_mFwE,4247
-konduktor_nightly-0.1.0.dev20250821104804.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-konduktor_nightly-0.1.0.dev20250821104804.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
-konduktor_nightly-0.1.0.dev20250821104804.dist-info/RECORD,,
+konduktor_nightly-0.1.0.dev20250822104721.dist-info/LICENSE,sha256=MuuqTZbHvmqXR_aNKAXzggdV45ANd3wQ5YI7tnpZhm0,6586
+konduktor_nightly-0.1.0.dev20250822104721.dist-info/METADATA,sha256=VG5_2yweBen0IWkOhw4PQWmuSbQ9-aYQ33IsxwyKRDE,4247
+konduktor_nightly-0.1.0.dev20250822104721.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+konduktor_nightly-0.1.0.dev20250822104721.dist-info/entry_points.txt,sha256=k3nG5wDFIJhNqsZWrHk4d0irIB2Ns9s47cjRWYsTCT8,48
+konduktor_nightly-0.1.0.dev20250822104721.dist-info/RECORD,,