skypilot-nightly 1.0.0.dev20250117__py3-none-any.whl → 1.0.0.dev20250119__py3-none-any.whl

sky/__init__.py

@@ -5,7 +5,7 @@ from typing import Optional
 import urllib.request
 
 # Replaced with the current commit when building the wheels.
-_SKYPILOT_COMMIT_SHA = '9e1b4ddc5fb1cb3fd6c00c106555b919e449e2c9'
+_SKYPILOT_COMMIT_SHA = '2354b818b6faa06ac65e1eb5bfad7e2278ca43f8'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20250117'
+__version__ = '1.0.0.dev20250119'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/jobs/state.py

@@ -313,6 +313,8 @@ _SPOT_STATUS_TO_COLOR = {
 class ManagedJobScheduleState(enum.Enum):
     """Captures the state of the job from the scheduler's perspective.
 
+    A job that predates the introduction of the scheduler will be INVALID.
+
     A newly created job will be INACTIVE.  The following transitions are valid:
     - INACTIVE -> WAITING: The job is "submitted" to the scheduler, and its job
       controller can be started.
@@ -342,6 +344,10 @@ class ManagedJobScheduleState(enum.Enum):
     briefly observe inconsistent states, like a job that just finished but
     hasn't yet transitioned to DONE.
     """
+    # This job may have been created before scheduler was introduced in #4458.
+    # This state is not used by scheduler but just for backward compatibility.
+    # TODO(cooperc): remove this in v0.11.0
+    INVALID = None
     # The job should be ignored by the scheduler.
     INACTIVE = 'INACTIVE'
     # The job is waiting to transition to LAUNCHING for the first time. The

sky/provision/gcp/instance_utils.py

@@ -38,7 +38,7 @@ _FIREWALL_RESOURCE_NOT_FOUND_PATTERN = re.compile(
     r'The resource \'projects/.*/global/firewalls/.*\' was not found')
 
 
-def _retry_on_http_exception(
+def _retry_on_gcp_http_exception(
     regex: Optional[str] = None,
     max_retries: int = GCP_MAX_RETRIES,
     retry_interval_s: int = GCP_RETRY_INTERVAL_SECONDS,
@@ -49,17 +49,18 @@ def _retry_on_http_exception(
 
         @functools.wraps(func)
         def wrapper(*args, **kwargs):
-            exception_type = gcp.http_error_exception()
 
             def try_catch_exc():
                 try:
                     value = func(*args, **kwargs)
                     return value
                 except Exception as e:  # pylint: disable=broad-except
-                    if not isinstance(e, exception_type) or (
-                            regex and not re.search(regex, str(e))):
-                        raise
-                    return e
+                    if (isinstance(e, gcp.http_error_exception()) and
+                        (regex is None or re.search(regex, str(e)))):
+                        logger.error(
+                            f'Retrying for gcp.http_error_exception: {e}')
+                        return e
+                    raise
 
             for _ in range(max_retries):
                 ret = try_catch_exc()
@@ -431,7 +432,7 @@ class GCPComputeInstance(GCPInstance):
         logger.debug(
             f'Waiting GCP operation {operation["name"]} to be ready ...')
 
-        @_retry_on_http_exception(
+        @_retry_on_gcp_http_exception(
             f'Failed to wait for operation {operation["name"]}')
         def call_operation(fn, timeout: int):
             request = fn(
@@ -613,6 +614,11 @@ class GCPComputeInstance(GCPInstance):
         return operation
 
     @classmethod
+    # When there is a cloud function running in parallel to set labels for
+    # newly created instances, it may fail with the following error:
+    #   "Labels fingerprint either invalid or resource labels have changed"
+    # We should retry until the labels are set successfully.
+    @_retry_on_gcp_http_exception('Labels fingerprint either invalid')
     def set_labels(cls, project_id: str, availability_zone: str, node_id: str,
                    labels: dict) -> None:
         node = cls.load_resource().instances().get(
@@ -1211,7 +1217,7 @@ class GCPTPUVMInstance(GCPInstance):
         """Poll for TPU operation until finished."""
         del project_id, region, zone  # unused
 
-        @_retry_on_http_exception(
+        @_retry_on_gcp_http_exception(
             f'Failed to wait for operation {operation["name"]}')
         def call_operation(fn, timeout: int):
             request = fn(name=operation['name'])
@@ -1379,7 +1385,7 @@ class GCPTPUVMInstance(GCPInstance):
                     f'Failed to get VPC name for instance {instance}') from e
 
     @classmethod
-    @_retry_on_http_exception('unable to queue the operation')
+    @_retry_on_gcp_http_exception('unable to queue the operation')
     def set_labels(cls, project_id: str, availability_zone: str, node_id: str,
                    labels: dict) -> None:
         while True:

{skypilot_nightly-1.0.0.dev20250117.dist-info → skypilot_nightly-1.0.0.dev20250119.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: skypilot-nightly
-Version: 1.0.0.dev20250117
+Version: 1.0.0.dev20250119
 Summary: SkyPilot: An intercloud broker for the clouds
 Author: SkyPilot Team
 License: Apache 2.0

{skypilot_nightly-1.0.0.dev20250117.dist-info → skypilot_nightly-1.0.0.dev20250119.dist-info}/RECORD

@@ -1,4 +1,4 @@
-sky/__init__.py,sha256=kA2ieB1SawvfYteqYNXCfJepPEo1gELmFmJNfKXgdkM,5944
+sky/__init__.py,sha256=aC5XkRJ6V-eJ02IR9TVtzIZdM6O3RIr3pg2xI_Qb3jM,5944
 sky/admin_policy.py,sha256=hPo02f_A32gCqhUueF0QYy1fMSSKqRwYEg_9FxScN_s,3248
 sky/authentication.py,sha256=LXUDABKP1FJCS256xTTDJa40WXwHKF5x49S-4hZbD1M,21501
 sky/check.py,sha256=s8deMVL-k9y8gd519K7NWZc3DqWsEySwiAr0uH3Vvcc,9459
@@ -102,7 +102,7 @@ sky/jobs/controller.py,sha256=Qv7vOj4OXkbnZF0F9wKrlJsYhkSNJjJ1Mgrn2FyQyaM,28250
 sky/jobs/core.py,sha256=2_Q9thiBPnd3i2nDqyUtQY-dsGZ1kRgAdnLcXHoycYo,19938
 sky/jobs/recovery_strategy.py,sha256=m-EA-MWXPFrgx2CYFPr6MmgeUoDTEBmY2xruD2PRSGY,26365
 sky/jobs/scheduler.py,sha256=WAvNb8-vBk8q1zFordFdpH7gxqWDjPHDGZZay6aodOk,12028
-sky/jobs/state.py,sha256=DK-tQnN200SpCXzDllRRibfQOjDCCFo7VRGSdi-FKA4,37856
+sky/jobs/state.py,sha256=Cjv2UEKfk3j7enXaCkU9CDqsvUfYZ3FWnYEH5HMachs,38153
 sky/jobs/utils.py,sha256=waKmLbUNRXeuYKBn_U7sekSFGAEgoPp9QemUULK4Y9k,49491
 sky/jobs/dashboard/dashboard.py,sha256=KMSarpVcfnc-ELPFvy1M9_I1k4kSeXubTk3ibQC67Tg,3219
 sky/jobs/dashboard/static/favicon.ico,sha256=uYlvgxSM7gjBmXpZ8wydvZUPAbJiiix-rc2Xe5mma9s,15086
@@ -142,7 +142,7 @@ sky/provision/gcp/__init__.py,sha256=zlgjR2JoaGD7sStGStMRu9bJ62f-8NKEIyb-bFHBlzM
 sky/provision/gcp/config.py,sha256=rNpnRFNZqqvEHjzjSdwMoI7Fq7RW9w_dL2vIaubj3Dc,33319
 sky/provision/gcp/constants.py,sha256=9eLZatVSW2uGqxrvFGaWK2A_Ab0o_4S4GdgUuI7mfsk,7441
 sky/provision/gcp/instance.py,sha256=AMnJz6xDwYqIHmDfJfENTG_ID6uhjD_2VhlWw8FJp_s,24934
-sky/provision/gcp/instance_utils.py,sha256=veRBr6Oziv0KaUdC4acuWeaOremNV0gMYCCHaSvY7c8,70943
+sky/provision/gcp/instance_utils.py,sha256=T0AVT8lMn128snPp3MvqmhXOihlZSC8-c1QpgYT4_FA,71377
 sky/provision/gcp/mig_utils.py,sha256=oFpcFZoapHMILSE4iIm8V5bxP1RhbMHRF7cciqq8qAk,7883
 sky/provision/kubernetes/__init__.py,sha256=y6yVfii81WYG3ROxv4hiIj-ydinS5-xGxLvXnARVQoI,719
 sky/provision/kubernetes/config.py,sha256=bXwOGdSAnXCkDreew0KsSUqSv3ZrptNeevqat76LLts,29012
@@ -289,9 +289,9 @@ sky/utils/kubernetes/k8s_gpu_labeler_job.yaml,sha256=k0TBoQ4zgf79-sVkixKSGYFHQ7Z
 sky/utils/kubernetes/k8s_gpu_labeler_setup.yaml,sha256=VLKT2KKimZu1GDg_4AIlIt488oMQvhRZWwsj9vBbPUg,3812
 sky/utils/kubernetes/rsync_helper.sh,sha256=h4YwrPFf9727CACnMJvF3EyK_0OeOYKKt4su_daKekw,1256
 sky/utils/kubernetes/ssh_jump_lifecycle_manager.py,sha256=Kq1MDygF2IxFmu9FXpCxqucXLmeUrvs6OtRij6XTQbo,6554
-skypilot_nightly-1.0.0.dev20250117.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
-skypilot_nightly-1.0.0.dev20250117.dist-info/METADATA,sha256=NeM0Gfty_C94SeoCFNc7G5N0Q2Gb4tNCxgwZv_e_jnQ,20884
-skypilot_nightly-1.0.0.dev20250117.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-skypilot_nightly-1.0.0.dev20250117.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
-skypilot_nightly-1.0.0.dev20250117.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
-skypilot_nightly-1.0.0.dev20250117.dist-info/RECORD,,
+skypilot_nightly-1.0.0.dev20250119.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
+skypilot_nightly-1.0.0.dev20250119.dist-info/METADATA,sha256=EMtxUVksl6PZt-fJt5k6PvgcWoaDL1pHzaQEc5K9iCU,20884
+skypilot_nightly-1.0.0.dev20250119.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+skypilot_nightly-1.0.0.dev20250119.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
+skypilot_nightly-1.0.0.dev20250119.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
+skypilot_nightly-1.0.0.dev20250119.dist-info/RECORD,,