skypilot-nightly 1.0.0.dev20250514__py3-none-any.whl → 1.0.0.dev20250516__py3-none-any.whl

sky/jobs/server/core.py

@@ -172,9 +172,6 @@ def launch(
         controller_resources = controller_utils.get_controller_resources(
             controller=controller,
             task_resources=sum([list(t.resources) for t in dag.tasks], []))
-        controller_idle_minutes_to_autostop, controller_down = (
-            controller_utils.get_controller_autostop_config(
-                controller=controller))
 
         vars_to_fill = {
             'remote_user_yaml_path': remote_user_yaml_path,
@@ -216,15 +213,12 @@ def launch(
         # Launch with the api server's user hash, so that sky status does not
         # show the owner of the controller as whatever user launched it first.
         with common.with_server_user_hash():
-            return execution.launch(
-                task=controller_task,
-                cluster_name=controller_name,
-                stream_logs=stream_logs,
-                idle_minutes_to_autostop=controller_idle_minutes_to_autostop,
-                down=controller_down,
-                retry_until_up=True,
-                fast=True,
-                _disable_controller_check=True)
+            return execution.launch(task=controller_task,
+                                    cluster_name=controller_name,
+                                    stream_logs=stream_logs,
+                                    retry_until_up=True,
+                                    fast=True,
+                                    _disable_controller_check=True)
 
 
 def queue_from_kubernetes_pod(

sky/optimizer.py

@@ -671,7 +671,7 @@ class Optimizer:
         plan: Dict[task_lib.Task, resources_lib.Resources],
     ) -> float:
         """Estimates the total cost of running the DAG by the plan."""
-        total_cost = 0
+        total_cost = 0.
         for node in topo_order:
             resources = plan[node]
             if node.time_estimator_func is None:
@@ -777,10 +777,13 @@ class Optimizer:
             accelerators = resources.get_accelerators_str()
             spot = resources.get_spot_str()
             cloud = resources.cloud
-            vcpus, mem = cloud.get_vcpus_mem_from_instance_type(
+            assert cloud is not None, 'Cloud must be specified'
+            assert (resources.instance_type is not None), \
+                'Instance type must be specified'
+            vcpus_, mem_ = cloud.get_vcpus_mem_from_instance_type(
                 resources.instance_type)
 
-            def format_number(x):
+            def format_number(x: Optional[float]) -> str:
                 if x is None:
                     return '-'
                 elif x.is_integer():
@@ -788,8 +791,8 @@ class Optimizer:
                 else:
                     return f'{x:.1f}'
 
-            vcpus = format_number(vcpus)
-            mem = format_number(mem)
+            vcpus = format_number(vcpus_)
+            mem = format_number(mem_)
 
             if resources.zone is None:
                 region_or_zone = resources.region
@@ -814,11 +817,12 @@ class Optimizer:
 
             accelerators = resources.get_accelerators_str()
             spot = resources.get_spot_str()
+            resources = resources.assert_launchable()
             cloud = resources.cloud
-            vcpus, mem = cloud.get_vcpus_mem_from_instance_type(
+            vcpus_, mem_ = cloud.get_vcpus_mem_from_instance_type(
                 resources.instance_type)
 
-            def format_number(x):
+            def format_number(x: Optional[float]) -> str:
                 if x is None:
                     return '-'
                 elif x.is_integer():
@@ -826,8 +830,8 @@ class Optimizer:
                 else:
                     return f'{x:.1f}'
 
-            vcpus = format_number(vcpus)
-            mem = format_number(mem)
+            vcpus = format_number(vcpus_)
+            mem = format_number(mem_)
 
             if resources.zone is None:
                 region_or_zone = resources.region
@@ -1195,10 +1199,12 @@ def _check_specified_clouds(dag: 'dag_lib.Dag') -> None:
             all_clouds_specified.add(cloud_str)
 
         # Explicitly check again to update the enabled cloud list.
-        sky_check.check_capability(sky_cloud.CloudCapability.COMPUTE,
-                                   quiet=True,
-                                   clouds=list(clouds_need_recheck -
-                                               global_disabled_clouds))
+        clouds_to_check_again = list(clouds_need_recheck -
+                                     global_disabled_clouds)
+        if len(clouds_to_check_again) > 0:
+            sky_check.check_capability(sky_cloud.CloudCapability.COMPUTE,
+                                       quiet=True,
+                                       clouds=clouds_to_check_again)
         enabled_clouds = sky_check.get_cached_enabled_clouds_or_refresh(
             capability=sky_cloud.CloudCapability.COMPUTE,
             raise_if_no_cloud_access=True)

sky/provision/kubernetes/utils.py

@@ -1736,9 +1736,16 @@ class KubernetesInstanceType:
     @staticmethod
     def is_valid_instance_type(name: str) -> bool:
         """Returns whether the given name is a valid instance type."""
+        # Before https://github.com/skypilot-org/skypilot/pull/4756,
+        # the accelerators are appended with format "--{a}{type}",
+        # e.g. "4CPU--16GB--1V100".
+        # Check both patterns to keep backward compatibility.
+        # TODO(romilb): Backward compatibility, remove after 0.11.0.
+        prev_pattern = re.compile(
+            r'^(\d+(\.\d+)?CPU--\d+(\.\d+)?GB)(--\d+\S+)?$')
         pattern = re.compile(
             r'^(\d+(\.\d+)?CPU--\d+(\.\d+)?GB)(--[\w\d-]+:\d+)?$')
-        return bool(pattern.match(name))
+        return bool(pattern.match(name)) or bool(prev_pattern.match(name))
 
     @classmethod
     def _parse_instance_type(
@@ -1755,6 +1762,11 @@ class KubernetesInstanceType:
             r'^(?P<cpus>\d+(\.\d+)?)CPU--(?P<memory>\d+(\.\d+)?)GB(?:--(?P<accelerator_type>[\w\d-]+):(?P<accelerator_count>\d+))?$'  # pylint: disable=line-too-long
         )
         match = pattern.match(name)
+        # TODO(romilb): Backward compatibility, remove after 0.11.0.
+        prev_pattern = re.compile(
+            r'^(?P<cpus>\d+(\.\d+)?)CPU--(?P<memory>\d+(\.\d+)?)GB(?:--(?P<accelerator_count>\d+)(?P<accelerator_type>\S+))?$'  # pylint: disable=line-too-long
+        )
+        prev_match = prev_pattern.match(name)
         if match:
             cpus = float(match.group('cpus'))
             memory = float(match.group('memory'))
@@ -1769,6 +1781,19 @@ class KubernetesInstanceType:
                 accelerator_count = None
                 accelerator_type = None
             return cpus, memory, accelerator_count, accelerator_type
+        # TODO(romilb): Backward compatibility, remove after 0.11.0.
+        elif prev_match:
+            cpus = float(prev_match.group('cpus'))
+            memory = float(prev_match.group('memory'))
+            accelerator_count = prev_match.group('accelerator_count')
+            accelerator_type = prev_match.group('accelerator_type')
+            if accelerator_count:
+                accelerator_count = int(accelerator_count)
+                accelerator_type = str(accelerator_type).replace('_', ' ')
+            else:
+                accelerator_count = None
+                accelerator_type = None
+            return cpus, memory, accelerator_count, accelerator_type
         else:
             raise ValueError(f'Invalid instance name: {name}')