konduktor-nightly 0.1.0.dev20250909104711__tar.gz → 0.1.0.dev20250910100913__tar.gz

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: konduktor-nightly
-Version: 0.1.0.dev20250909104711
+Version: 0.1.0.dev20250910100913
 Summary: GPU Cluster Health Management
 Author: Andrew Aikawa
 Author-email: asai@berkeley.edu

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/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 = 'b772c32b1fa9e082d58d857dd655abc132c7bab4'
+_KONDUKTOR_COMMIT_SHA = 'ca586c26e1db34e0b199c9fa57e84b39983db15e'
 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.dev20250909104711'
+__version__ = '1.0.0.dev0.1.0.dev20250910100913'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/konduktor/backends/jobset_utils.py

@@ -117,7 +117,12 @@ def create_jobset(
         )
         logger.info(
             f'task {colorama.Fore.CYAN}{colorama.Style.BRIGHT}'
-            f'{task.name}{colorama.Style.RESET_ALL} created'
+            f'{colorama.Fore.CYAN}{colorama.Style.BRIGHT}{task.name}'
+            f'{colorama.Style.RESET_ALL} created in context '
+            f'{colorama.Fore.YELLOW}{colorama.Style.BRIGHT}{context}'
+            f'{colorama.Style.RESET_ALL}, namespace '
+            f'{colorama.Fore.GREEN}{colorama.Style.BRIGHT}{namespace}'
+            f'{colorama.Style.RESET_ALL}'
         )
         return jobset
     except kube_client.api_exception() as err:

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/konduktor/kube_client.py

@@ -1,10 +1,11 @@
 import logging
 import os
-from typing import Any, Callable, Optional
+from typing import Any, Callable, List, Optional
 
 import kubernetes
 import urllib3
 
+from konduktor import config
 from konduktor import logging as konduktor_logging
 from konduktor.utils import annotations, ux_utils
 
@@ -20,6 +21,10 @@ DEFAULT_IN_CLUSTER_REGION = 'in-cluster'
 # set to DEFAULT_IN_CLUSTER_REGION.
 IN_CLUSTER_CONTEXT_NAME_ENV_VAR = 'KONDUKTOR_IN_CLUSTER_CONTEXT_NAME'
 
+# Tracks the most recently selected/loaded context name.
+# None means no explicit context was resolved/loaded yet.
+_ACTIVE_CONTEXT: Optional[str] = None
+
 
 def _decorate_methods(obj: Any, decorator: Callable, decoration_type: str):
     for attr_name in dir(obj):
@@ -54,12 +59,37 @@ def _api_logging_decorator(logger: str, level: int):
 def _load_config(context: Optional[str] = None):
     urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
 
+    # If no context explicitly provided, prefer the configured allowed context
+    # (first element) when present. This ensures the client defaults to the
+    # user-specified context instead of kubeconfig's current-context.
+    effective_context = context
+    allowed_contexts: List[str] = config.get_nested(
+        ('kubernetes', 'allowed_contexts'), []
+    )
+
+    is_allowed_selected = False
+    if effective_context is None and allowed_contexts:
+        effective_context = allowed_contexts[0]
+        is_allowed_selected = True
+        logger.info(
+            'Detected kubernetes.allowed_contexts in config; using context: %s',
+            effective_context,
+        )
+
     def _load_config_from_kubeconfig(context: Optional[str] = None):
         try:
             kubernetes.config.load_kube_config(context=context)
         except kubernetes.config.config_exception.ConfigException as e:
-            # Check if exception was due to no current-context
-            if 'Expected key current-context' in str(e):
+            # Improve error when a configured allowed context cannot be loaded
+            msg = str(e)
+            if is_allowed_selected and context is not None:
+                err_str = (
+                    'Configured Kubernetes context not usable: '
+                    f'kubernetes.allowed_contexts[0] = {context!r}. '
+                    'Please ensure this context exists and is valid in your '
+                    'kubeconfig (typically at ~/.kube/config).'
+                )
+            elif 'Expected key current-context' in msg:
                 err_str = (
                     f'Failed to load Kubernetes configuration for {context!r}. '
                     'Kubeconfig does not contain any valid context(s).\n'
@@ -73,7 +103,8 @@ def _load_config(context: Optional[str] = None):
             with ux_utils.print_exception_no_traceback():
                 raise ValueError(err_str) from None
 
-    if context == in_cluster_context_name() or context is None:
+    global _ACTIVE_CONTEXT
+    if effective_context == in_cluster_context_name() or effective_context is None:
         try:
             # Load in-cluster config if running in a pod and context is None.
             # Kubernetes set environment variables for service discovery do not
@@ -83,10 +114,22 @@ def _load_config(context: Optional[str] = None):
             os.environ['KUBERNETES_SERVICE_HOST'] = 'kubernetes.default.svc'
             os.environ['KUBERNETES_SERVICE_PORT'] = '443'
             kubernetes.config.load_incluster_config()
+            _ACTIVE_CONTEXT = in_cluster_context_name()
         except kubernetes.config.config_exception.ConfigException:
-            _load_config_from_kubeconfig()
+            # If allowed_contexts was specified, do not fall back silently.
+            if is_allowed_selected:
+                _load_config_from_kubeconfig(effective_context)
+            else:
+                _load_config_from_kubeconfig()
+            # Best effort: set active context to current-context from kubeconfig
+            try:
+                _, current_ctx = kubernetes.config.list_kube_config_contexts()
+                _ACTIVE_CONTEXT = current_ctx.get('name') if current_ctx else None
+            except kubernetes.config.config_exception.ConfigException:
+                _ACTIVE_CONTEXT = None
     else:
-        _load_config_from_kubeconfig(context)
+        _load_config_from_kubeconfig(effective_context)
+        _ACTIVE_CONTEXT = effective_context
 
 
 @_api_logging_decorator('urllib3', logging.ERROR)
@@ -183,3 +226,12 @@ def in_cluster_context_name() -> Optional[str]:
     context name.
     """
     return os.environ.get(IN_CLUSTER_CONTEXT_NAME_ENV_VAR) or DEFAULT_IN_CLUSTER_REGION
+
+
+def get_active_context() -> Optional[str]:
+    """Returns the last context selected by the client loader.
+
+    This reflects the effective context used by the most recent client init.
+    May be None if no client has been initialized yet.
+    """
+    return _ACTIVE_CONTEXT

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/konduktor/utils/kubernetes_utils.py

@@ -264,11 +264,31 @@ AUTOSCALER_TO_LABEL_FORMATTER = {
 
 @functools.lru_cache()
 def get_current_kube_config_context_name() -> Optional[str]:
-    """Get the current kubernetes context from the kubeconfig file
+    """Get the active Kubernetes context name.
+
+    Precedence:
+    1) The first entry in `kubernetes.allowed_contexts` (if configured).
+    2) kubeconfig's current-context (fallback when not configured).
 
     Returns:
-        str | None: The current kubernetes context if it exists, None otherwise
+        str | None: The selected context if it exists, None otherwise.
     """
+    # 1) Prefer a user-configured allowed context if provided.
+    try:
+        allowed_contexts: Optional[List[str]] = config.get_nested(
+            ('kubernetes', 'allowed_contexts'), None
+        )
+        if allowed_contexts:
+            context = allowed_contexts[0]
+            logger.info(
+                'Detected kubernetes.allowed_contexts in config; using context: %s',
+                context,
+            )
+            return context
+    except Exception:  # fallback safely if config loading fails unexpectedly
+        pass
+
+    # 2) Fall back to kubeconfig's current context
     k8s = kubernetes
     try:
         _, current_context = k8s.config.list_kube_config_contexts()
@@ -336,13 +356,24 @@ def get_kube_config_context_namespace(context_name: Optional[str] = None) -> str
     """
     k8s = kubernetes
     ns_path = '/var/run/secrets/kubernetes.io/serviceaccount/namespace'
-    # If using in-cluster context, get the namespace from the service account
-    # namespace file. Uses the same logic as adaptors.kubernetes._load_config()
-    # to stay consistent with in-cluster config loading.
+
+    # If no explicit context provided, prefer configured allowed context first.
+    if context_name is None:
+        try:
+            allowed_contexts: Optional[List[str]] = config.get_nested(
+                ('kubernetes', 'allowed_contexts'), None
+            )
+            if allowed_contexts:
+                context_name = allowed_contexts[0]
+        except Exception:
+            pass
+
+    # If using in-cluster context, get the namespace from the SA namespace file.
     if context_name == kube_client.in_cluster_context_name() or context_name is None:
         if os.path.exists(ns_path):
             with open(ns_path, encoding='utf-8') as f:
                 return f.read().strip()
+
     # If not in-cluster, get the namespace from kubeconfig
     try:
         contexts, current_context = k8s.config.list_kube_config_contexts()

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/konduktor/utils/log_utils.py

@@ -170,6 +170,29 @@ def create_table(field_names: List[str], **kwargs) -> prettytable.PrettyTable:
     return table
 
 
+def _get_kr8s_api():
+    """Return a kr8s API configured for the selected context.
+
+    Honors kubernetes.allowed_contexts via
+    get_current_kube_config_context_name(). Fails with a clear error if a
+    configured context cannot be initialized.
+    """
+    ctx = kubernetes_utils.get_current_kube_config_context_name()
+    try:
+        # kr8s exposes a factory function `api(...)` that accepts context.
+        api = kr8s.api(context=ctx) if ctx else kr8s.api()
+        if ctx:
+            logger.debug('Initialized kr8s API for context: %s', ctx)
+        return api
+    except Exception as e:  # defensive: surface a clear error if context fails
+        if ctx:
+            raise ValueError(
+                'Failed to initialize kr8s client for context '
+                f'{ctx!r}. Ensure the context exists and your kubeconfig is valid.'
+            ) from e
+        raise
+
+
 def run_with_log(
     cmd: Union[List[str], str],
     log_path: str,
@@ -311,7 +334,8 @@ def tail_loki_logs_ws(
         # until we reach the end of the log and then invoke tail again.
         # Also include checks that the job is running/ever ran.
         raise ValueError('num_logs must be less than or equal to 5000')
-    loki_svc = kr8s.objects.Service.get('loki', namespace='loki')
+    api = _get_kr8s_api()
+    loki_svc = kr8s.objects.Service.get('loki', namespace='loki', api=api)
     with kr8s.portforward.PortForward(
         loki_svc, LOKI_REMOTE_PORT, local_port='auto'
     ) as port:
@@ -345,8 +369,9 @@ def tail_vicky_logs(
     context = kubernetes_utils.get_current_kube_config_context_name()
     namespace = kubernetes_utils.get_kube_config_context_namespace(context)
     query: Dict[str, Any] = {}
+    api = _get_kr8s_api()
     vicky_svc = kr8s.objects.Service.get(
-        'vls-victoria-logs-single-server', namespace='victoria-logs'
+        'vls-victoria-logs-single-server', namespace='victoria-logs', api=api
     )
 
     if num_logs == -1:

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/konduktor/utils/loki_utils.py

@@ -11,6 +11,7 @@ import kr8s
 import websockets
 
 from konduktor import logging
+from konduktor.utils import kubernetes_utils
 
 logger = logging.get_logger(__name__)
 
@@ -59,7 +60,18 @@ def tail_loki_logs_ws(
         # until we reach the end of the log and then invoke tail again.
         # Also include checks that the job is running/ever ran.
         raise ValueError('num_logs must be less than or equal to 5000')
-    loki_svc = kr8s.objects.Service.get('loki', namespace='loki')
+    # Initialize kr8s API honoring allowed_contexts if configured.
+    ctx = kubernetes_utils.get_current_kube_config_context_name()
+    try:
+        api = kr8s.api(context=ctx) if ctx else kr8s.api()
+    except Exception as e:
+        if ctx:
+            raise ValueError(
+                'Failed to initialize kr8s client for context '
+                f'{ctx!r}. Ensure the context exists and your kubeconfig is valid.'
+            ) from e
+        raise
+    loki_svc = kr8s.objects.Service.get('loki', namespace='loki', api=api)
     with kr8s.portforward.PortForward(
         loki_svc, LOKI_REMOTE_PORT, local_port='auto'
     ) as port:

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/konduktor/utils/schemas.py

@@ -504,6 +504,13 @@ def get_config_schema():
                         'anyOf': [{'required': ['name']}, {'required': ['namespace']}]
                     },
                 },
+                'allowed_contexts': {
+                    'type': 'array',
+                    'items': {
+                        'type': 'string',
+                    },
+                    'maxItems': 1,
+                },
                 'provision_timeout': {
                     'type': 'integer',
                 },

{konduktor_nightly-0.1.0.dev20250909104711 → konduktor_nightly-0.1.0.dev20250910100913}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "konduktor-nightly"
-version = "0.1.0.dev20250909104711"
+version = "0.1.0.dev20250910100913"
 description = "GPU Cluster Health Management"
 packages = [
     {include = "konduktor"}