konduktor-nightly 0.1.0.dev20250820104812__tar.gz → 0.1.0.dev20250822104721__tar.gz

{konduktor_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250820104812
+Version: 0.1.0.dev20250822104721
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/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 = '108d7fe47b1bd5db50d555510714d2e204fb7b6f'
+_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.dev20250820104812'
+__version__ = '1.0.0.dev0.1.0.dev20250822104721'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

{konduktor_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/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_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/konduktor/backends/jobset_utils.py

@@ -464,6 +464,41 @@ def _get_end_time_from_conditions(job: Dict[str, Any]) -> str:
     return '-'
 
 
+def _get_time_delta(delta: 'timedelta') -> Tuple[str, 'timedelta']:
+    total_seconds = int(delta.total_seconds())
+
+    days, remainder = divmod(total_seconds, 86400)  # 86400 seconds in a day
+    hours, remainder = divmod(remainder, 3600)  # 3600 seconds in an hour
+    minutes, seconds = divmod(remainder, 60)  # 60 seconds in a minute
+
+    days_str = f'{days} day{"s" if days != 1 else ""}, ' if days > 0 else ''
+    hours_str = f'{hours} hr{"s" if hours != 1 else ""}, ' if hours > 0 else ''
+    minutes_str = (
+        f'{minutes} min{"s" if minutes != 1 else ""}'
+        if minutes > 0 and days == 0
+        else ''
+    )
+
+    seconds_str = (
+        f'{seconds} sec{"s" if seconds != 1 else ""}'
+        if seconds > 0 and days == 0 and hours == 0 and minutes == 0
+        else ''
+    )
+
+    result = f'{days_str}{hours_str}{minutes_str}{seconds_str}'
+    return result if result else '<1 minute', delta
+
+
+def _get_job_length(start_time: str, end_time: str) -> str:
+    if start_time == '-' or end_time == '-':
+        return '-'
+    else:
+        start = datetime.strptime(start_time, '%m/%d/%y %I:%M%p')
+        end = datetime.strptime(end_time, '%m/%d/%y %I:%M%p')
+        delta, _ = _get_time_delta(end - start)
+        return delta
+
+
 def show_status_table(
     namespace: str,
     all_users: bool,
@@ -531,32 +566,6 @@ def show_status_table(
                 f'{JobStatus.PENDING.name}{colorama.Style.RESET_ALL}'
             )
 
-    def _get_time_delta(timestamp: str) -> Tuple[str, 'timedelta']:
-        delta = datetime.now(timezone.utc) - datetime.strptime(
-            timestamp, '%Y-%m-%dT%H:%M:%SZ'
-        ).replace(tzinfo=timezone.utc)
-        total_seconds = int(delta.total_seconds())
-
-        days, remainder = divmod(total_seconds, 86400)  # 86400 seconds in a day
-        hours, remainder = divmod(remainder, 3600)  # 3600 seconds in an hour
-        minutes, seconds = divmod(remainder, 60)  # 60 seconds in a minute
-
-        days_str = f'{days} day{"s" if days != 1 else ""}, ' if days > 0 else ''
-        hours_str = f'{hours} hr{"s" if hours != 1 else ""}, ' if hours > 0 else ''
-        minutes_str = (
-            f'{minutes} min{"s" if minutes != 1 else ""}'
-            if minutes > 0 and days == 0
-            else ''
-        )
-        seconds_str = (
-            f'{seconds} sec{"s" if seconds != 1 else ""}'
-            if seconds > 0 and days == 0 and hours == 0 and minutes == 0
-            else ''
-        )
-
-        result = f'{days_str}{hours_str}{minutes_str}{seconds_str}'
-        return result if result else '<1 minute', delta
-
     def _get_resources(job: Dict[str, Any]) -> str:
         num_pods = int(
             job['spec']['replicatedJobs'][0]['template']['spec']['parallelism']
@@ -580,9 +589,18 @@ def show_status_table(
             'SUBMITTED',
             'START TIME',
             'END TIME',
+            'DURATION',
         ]
     else:
-        columns = ['NAME', 'STATUS', 'RESOURCES', 'SUBMITTED', 'START TIME', 'END TIME']
+        columns = [
+            'NAME',
+            'STATUS',
+            'RESOURCES',
+            'SUBMITTED',
+            'START TIME',
+            'END TIME',
+            'DURATION',
+        ]
     job_table = log_utils.create_table(columns)
     job_specs = list_jobset(namespace)
     assert job_specs is not None, 'Retrieving jobs failed'
@@ -621,13 +639,18 @@ def show_status_table(
             start_time = _format_timestamp(start_time)
 
         # Get submitted time (how long ago)
-        submitted_time, _ = _get_time_delta(job['metadata']['creationTimestamp'])
+        time_delta = datetime.now(timezone.utc) - datetime.strptime(
+            job['metadata']['creationTimestamp'], '%Y-%m-%dT%H:%M:%SZ'
+        ).replace(tzinfo=timezone.utc)
+        submitted_time, _ = _get_time_delta(time_delta)
 
         # Get end time (from JobSet conditions)
         end_time = _get_end_time_from_conditions(job)
         if end_time != '-':
             end_time = _format_timestamp(end_time)
 
+        job_length = _get_job_length(start_time, end_time)
+
         if all_users:
             rows.append(
                 [
@@ -638,6 +661,7 @@ def show_status_table(
                     submitted_time,
                     start_time,
                     end_time,
+                    job_length,
                     job['metadata']['creationTimestamp'],
                 ]
             )
@@ -654,6 +678,7 @@ def show_status_table(
                     submitted_time,
                     start_time,
                     end_time,
+                    job_length,
                     job['metadata']['creationTimestamp'],
                 ]
             )

{konduktor_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/konduktor/backends/pod_utils.py

@@ -1,6 +1,7 @@
 """Pod utils: handles pod spec creation and manipulation"""
 
 import base64
+import json
 import os
 import tempfile
 import typing
@@ -284,7 +285,7 @@ def create_pod_spec(task: 'konduktor.Task') -> Dict[str, Any]:
     pod_config['kubernetes']['pod_config']['spec']['containers'][0]['env'] = list(
         env_map.values()
     )
-    logger.debug(f'rendered pod spec: \n\t{pod_config}')
+    logger.debug(f'rendered pod spec: \n\t{json.dumps(pod_config, indent=2)}')
 
     # validate pod spec using json schema
     try:
@@ -328,15 +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_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/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_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/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_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/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_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/konduktor/templates/pod.yaml.j2

@@ -161,6 +161,10 @@ kubernetes:
           - name: git-ssh-secret
             mountPath: /run/konduktor/git-ssh-secret
           {% endif %}
+          {% if tailscale_secret %}
+          - name: tailscale-state
+            mountPath: /var/lib/tailscale
+          {% endif %}
           command: ["bash", "-c"]
           args:
             - |
@@ -317,18 +321,19 @@ kubernetes:
               export TS_HOSTNAME=$(echo "$POD_NAME" | sed 's/-[^-]*$//')
               $(prefix_cmd) echo "TS_HOSTNAME=${TS_HOSTNAME}" >> /etc/environment
               function InstallTailscale {
-                while ! tailscale status >/dev/null 2>&1; do
+                if ! command -v tailscale >/dev/null 2>&1; then
+                  $(prefix_cmd) curl -fsSL https://tailscale.com/install.sh | DEBIAN_FRONTEND=noninteractive $(prefix_cmd) sh > ~/.konduktor/tmp/tailscale-install.log 2>&1
+                fi
+                if ! tailscale status >/dev/null 2>&1; then
                   $(prefix_cmd) mkdir -p /var/run/tailscale /var/cache/tailscale /var/lib/tailscale
-                  if ! command -v tailscale >/dev/null 2>&1; then
-                    $(prefix_cmd) curl -fsSL https://tailscale.com/install.sh | DEBIAN_FRONTEND=noninteractive $(prefix_cmd) sh > ~/.konduktor/tmp/tailscale-install.log 2>&1
-                  fi
-                  $(prefix_cmd) tailscaled --tun=userspace-networking --state=mem: >~/.konduktor/tmp/tailscaled.log 2>&1 &
-                  $(prefix_cmd) sleep 2
-                  $(prefix_cmd) timeout 5 tailscale up --auth-key=${TS_AUTHKEY} --ssh --hostname=${TS_HOSTNAME} || echo "tailscale up failed retrying"
-                  $(prefix_cmd) sleep 2
+                  $(prefix_cmd) nohup tailscaled --tun=userspace-networking >~/.konduktor/tmp/tailscaled.log 2>&1 &
+                fi
+                until tailscale status >/dev/null 2>&1; do
+                  $(prefix_cmd) tailscale up --auth-key=${TS_AUTHKEY} --ssh --hostname=${TS_HOSTNAME} --accept-dns=false || echo "tailscale up failed retrying"
                 done
                 $(prefix_cmd) echo "Tailscale is up"
                 $(prefix_cmd) tailscale status
+                $(prefix_cmd) tailscale netcheck
               }
               InstallTailscale | tee ~/.konduktor/tmp/tailscale-out.log
               {% if konduktor_debug %}
@@ -413,6 +418,10 @@ kubernetes:
         emptyDir:
           medium: "Memory"
           sizeLimit: 4Gi
+      {% if tailscale_secret %}
+      - name: tailscale-state
+        emptyDir: {}
+      {% endif %}
       - name: sync
         emptyDir: {}
       {% for secret_type, secret_name in mount_secrets.items() %}

{konduktor_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/konduktor/utils/log_utils.py

@@ -397,6 +397,8 @@ def tail_vicky_logs(
                 for line in response.iter_lines(decode_unicode=True):
                     if line:
                         payload = json.loads(line)
+                        if 'missing _msg field' in payload['_msg']:
+                            payload['_msg'] = ''
                         print(
                             f"{colorama.Fore.CYAN}{colorama.Style.BRIGHT} "
                             f"(job_name={job_name} worker_id={worker_id})"

{konduktor_nightly-0.1.0.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/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.dev20250820104812 → konduktor_nightly-0.1.0.dev20250822104721}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "konduktor-nightly"
-version = "0.1.0.dev20250820104812"
+version = "0.1.0.dev20250822104721"
 description = "GPU Cluster Health Management"
 packages = [
     {include = "konduktor"}