skypilot-nightly 1.0.0.dev20240906__py3-none-any.whl → 1.0.0.dev20240907__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 = '2e545b8db70991bc74d112d5070fa908078f9cdb'
+_SKYPILOT_COMMIT_SHA = '363f27bd1fc5444531966bae7ce03e406b03d646'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20240906'
+__version__ = '1.0.0.dev20240907'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/clouds/azure.py

@@ -58,12 +58,11 @@ class Azure(clouds.Cloud):
     # names, so the limit is 64 - 4 - 7 - 10 = 43.
     # Reference: https://azure.github.io/PSRule.Rules.Azure/en/rules/Azure.ResourceGroup.Name/ # pylint: disable=line-too-long
     _MAX_CLUSTER_NAME_LEN_LIMIT = 42
-    _BEST_DISK_TIER = resources_utils.DiskTier.MEDIUM
+    _BEST_DISK_TIER = resources_utils.DiskTier.HIGH
     _DEFAULT_DISK_TIER = resources_utils.DiskTier.MEDIUM
     # Azure does not support high disk and ultra disk tier.
-    _SUPPORTED_DISK_TIERS = (
-        set(resources_utils.DiskTier) -
-        {resources_utils.DiskTier.HIGH, resources_utils.DiskTier.ULTRA})
+    _SUPPORTED_DISK_TIERS = (set(resources_utils.DiskTier) -
+                             {resources_utils.DiskTier.ULTRA})
 
     _INDENT_PREFIX = ' ' * 4
 
@@ -361,7 +360,9 @@ class Azure(clouds.Cloud):
                 start_index += 1
             assert False, 'Low disk tier should always be supported on Azure.'
 
-        return {
+        disk_tier = _failover_disk_tier()
+
+        resources_vars = {
             'instance_type': r.instance_type,
             'custom_resources': custom_resources,
             'num_gpus': acc_count,
@@ -371,12 +372,18 @@ class Azure(clouds.Cloud):
             'zones': None,
             **image_config,
             'need_nvidia_driver_extension': need_nvidia_driver_extension,
-            'disk_tier': Azure._get_disk_type(_failover_disk_tier()),
+            'disk_tier': Azure._get_disk_type(disk_tier),
             'cloud_init_setup_commands': cloud_init_setup_commands,
             'azure_subscription_id': self.get_project_id(dryrun),
             'resource_group': f'{cluster_name.name_on_cloud}-{region_name}',
         }
 
+        # Setting disk performance tier for high disk tier.
+        if disk_tier == resources_utils.DiskTier.HIGH:
+            resources_vars['disk_performance_tier'] = 'P50'
+
+        return resources_vars
+
     def _get_feasible_launchable_resources(
         self, resources: 'resources.Resources'
     ) -> 'resources_utils.FeasibleResources':
@@ -600,10 +607,10 @@ class Azure(clouds.Cloud):
             disk_tier: Optional[resources_utils.DiskTier]) -> Tuple[bool, str]:
         if disk_tier is None or disk_tier == resources_utils.DiskTier.BEST:
             return True, ''
-        if disk_tier == resources_utils.DiskTier.HIGH or disk_tier == resources_utils.DiskTier.ULTRA:
+        if disk_tier == resources_utils.DiskTier.ULTRA:
             return False, (
-                'Azure disk_tier={high, ultra} is not supported now. '
-                'Please use disk_tier={low, medium, best} instead.')
+                'Azure disk_tier=ultra is not supported now. '
+                'Please use disk_tier={low, medium, high, best} instead.')
         # Only S-series supported premium ssd
         # see https://stackoverflow.com/questions/48590520/azure-requested-operation-cannot-be-performed-because-storage-account-type-pre  # pylint: disable=line-too-long
         if cls._get_disk_type(
@@ -611,7 +618,7 @@ class Azure(clouds.Cloud):
         ) == 'Premium_LRS' and not Azure._is_s_series(instance_type):
             return False, (
                 'Azure premium SSDs are only supported for S-series '
-                'instances. To use disk_tier=medium, please make sure '
+                'instances. To use disk_tier>=medium, please make sure '
                 'instance_type is specified to an S-series instance.')
         return True, ''
 
@@ -631,7 +638,7 @@ class Azure(clouds.Cloud):
         # cannot be used as OS disks so we might need data disk support
         tier2name = {
             resources_utils.DiskTier.ULTRA: 'Disabled',
-            resources_utils.DiskTier.HIGH: 'Disabled',
+            resources_utils.DiskTier.HIGH: 'Premium_LRS',
             resources_utils.DiskTier.MEDIUM: 'Premium_LRS',
             resources_utils.DiskTier.LOW: 'Standard_LRS',
         }

sky/provision/azure/instance.py

@@ -18,6 +18,7 @@ from sky.adaptors import azure
 from sky.provision import common
 from sky.provision import constants
 from sky.utils import common_utils
+from sky.utils import subprocess_utils
 from sky.utils import ux_utils
 
 if typing.TYPE_CHECKING:
@@ -274,6 +275,17 @@ def _create_instances(
         deployment_name=vm_name,
         parameters=parameters,
     ).wait()
+
+    performance_tier = node_config.get('disk_performance_tier', None)
+    if performance_tier is not None:
+        disks = compute_client.disks.list_by_resource_group(resource_group)
+        for disk in disks:
+            name = disk.name
+            # TODO(tian): Investigate if we can use Python SDK to update this.
+            subprocess_utils.run_no_outputs(
+                f'az disk update -n {name} -g {resource_group} '
+                f'--set tier={performance_tier}')
+
     filters = {
         constants.TAG_RAY_CLUSTER_NAME: cluster_name_on_cloud,
         _TAG_SKYPILOT_VM_ID: vm_id

sky/provision/runpod/instance.py

@@ -11,6 +11,7 @@ from sky.utils import resources_utils
 from sky.utils import ux_utils
 
 POLL_INTERVAL = 5
+QUERY_PORTS_TIMEOUT_SECONDS = 30
 
 logger = sky_logging.init_logger(__name__)
 
@@ -224,11 +225,25 @@ def query_ports(
 ) -> Dict[int, List[common.Endpoint]]:
     """See sky/provision/__init__.py"""
     del head_ip, provider_config  # Unused.
-    instances = _filter_instances(cluster_name_on_cloud, None, head_only=True)
-    assert len(instances) == 1
-    head_inst = list(instances.values())[0]
-    return {
-        port: [common.SocketEndpoint(**endpoint)]
-        for port, endpoint in head_inst['port2endpoint'].items()
-        if port in resources_utils.port_ranges_to_set(ports)
-    }
+    # RunPod ports sometimes take a while to be ready.
+    start_time = time.time()
+    ports_to_query = resources_utils.port_ranges_to_set(ports)
+    while True:
+        instances = _filter_instances(cluster_name_on_cloud,
+                                      None,
+                                      head_only=True)
+        assert len(instances) == 1
+        head_inst = list(instances.values())[0]
+        ready_ports: Dict[int, List[common.Endpoint]] = {
+            port: [common.SocketEndpoint(**endpoint)]
+            for port, endpoint in head_inst['port2endpoint'].items()
+            if port in ports_to_query
+        }
+        not_ready_ports = ports_to_query - set(ready_ports.keys())
+        if not not_ready_ports:
+            return ready_ports
+        if time.time() - start_time > QUERY_PORTS_TIMEOUT_SECONDS:
+            logger.warning(f'Querying ports {ports} timed out. Ports '
+                           f'{not_ready_ports} are not ready.')
+            return ready_ports
+        time.sleep(1)

sky/templates/azure-ray.yml.j2

@@ -81,6 +81,9 @@ available_node_types:
               {{ cmd }}
               {%- endfor %}
         need_nvidia_driver_extension: {{need_nvidia_driver_extension}}
+        {%- if disk_performance_tier is not none %}
+        disk_performance_tier: {{disk_performance_tier}}
+        {%- endif %}
         # TODO: attach disk
 
 head_node_type: ray.head.default

sky/utils/kubernetes/generate_kubeconfig.sh

@@ -169,7 +169,9 @@ roleRef:
   kind: Role
   name: skypilot-system-service-account-role
   apiGroup: rbac.authorization.k8s.io
----
+EOF
+# Apply optional ingress-related roles, but don't make the script fail if it fails
+kubectl apply -f - <<EOF || echo "Failed to apply optional ingress-related roles. Nginx ingress is likely not installed. This is not critical and the script will continue."
 # Optional: Role for accessing ingress resources
 apiVersion: rbac.authorization.k8s.io/v1
 kind: Role

{skypilot_nightly-1.0.0.dev20240906.dist-info → skypilot_nightly-1.0.0.dev20240907.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: skypilot-nightly
-Version: 1.0.0.dev20240906
+Version: 1.0.0.dev20240907
 Summary: SkyPilot: An intercloud broker for the clouds
 Author: SkyPilot Team
 License: Apache 2.0

{skypilot_nightly-1.0.0.dev20240906.dist-info → skypilot_nightly-1.0.0.dev20240907.dist-info}/RECORD

@@ -1,4 +1,4 @@
-sky/__init__.py,sha256=KUaBF_zJk4J7rLYkxjkEXZHIkZOyas42Bgy5v5cjx3s,5588
+sky/__init__.py,sha256=mOsHkFpaIrSqMgc4H57tcwwApJLz_ChgpL_3m8lgqGY,5588
 sky/authentication.py,sha256=PXKsabUKnvhoTNoNwZzo66qjCLAsuc5pQ8esVu0IYh8,20424
 sky/check.py,sha256=sqUCow5Gqqtp2ouk7tZ3whwDOK23wqyUJvcxMBsich8,9080
 sky/cli.py,sha256=2cOw3lXzRA-uLlEH-eK7N_1VmUpf1LR4Ztu-ZaKu3Is,201673
@@ -40,7 +40,7 @@ sky/benchmark/benchmark_state.py,sha256=X8CXmuU9KgsDRhKedhFgjeRMUFWtQsjFs1qECvPG
 sky/benchmark/benchmark_utils.py,sha256=oJOzJ4fs2sruxYh4Tl1NZ5fi2-3oWfXtoeCIAq2hgjw,26136
 sky/clouds/__init__.py,sha256=WuNIJEnZmBO72tU5awgaaL3rdvFRSkgaYNNeuY68dXo,1356
 sky/clouds/aws.py,sha256=piC1CJnY7rvBxpllW9aMB3SbTLt4JDrSUJL3moyYn4Q,48168
-sky/clouds/azure.py,sha256=VqD9TF90u3YBb91ZEKpv-K_UMYa_lLbhn-hmFMcger4,28350
+sky/clouds/azure.py,sha256=fo81PYEgSDcAG8fvFYAu0brrCL4WOHFZzsaNWtkW6Y0,28533
 sky/clouds/cloud.py,sha256=ufoep7SYzxHzphEFILJA0ggCwKGIqQMXAmRjb3X3QnU,33498
 sky/clouds/cloud_registry.py,sha256=4yQMv-iBSgyN5aNL4Qxbn0JVE-dkVoEUIgj7S1z9S_Q,955
 sky/clouds/cudo.py,sha256=otfYWN7jHqbKJU9meUEOZcjmCC_9_QLEQ4BZ5vHDyWo,13147
@@ -118,7 +118,7 @@ sky/provision/azure/__init__.py,sha256=87cgk1_Ws7n9rqaDDPv-HpfrkVeSQMdFQnhnXwyx9
 sky/provision/azure/azure-config-template.json,sha256=dwTO-DG70UtBifN59NzsJwPZHZ4uCs7_oLooHgXG_N8,4349
 sky/provision/azure/azure-vm-template.json,sha256=k0A4xclQ-xQbAdcjo516sZqMDsU-n8_jeRuPxtGKBFI,10901
 sky/provision/azure/config.py,sha256=Fx_N_F9912q2gUHt0FhAxVDwlTbd9cLxrP5rii0KJOM,6772
-sky/provision/azure/instance.py,sha256=JXZRWTeYojInFVEhizwhrzA1IImjLppgQygSvrgNmBg,32443
+sky/provision/azure/instance.py,sha256=-95c__W4d0mgKweUPhT3tPLAdNtdruroHrfIvNTFQtY,32962
 sky/provision/cudo/__init__.py,sha256=KAEl26MVPsk7IoP9Gg-MOJJRIV6-X9B0fbyHdyJWdLo,741
 sky/provision/cudo/config.py,sha256=RYOVkV0MoUqVBJRZiKhBZhjFygeyFs7eUdVMdPg1vds,327
 sky/provision/cudo/cudo_machine_type.py,sha256=_VNXWPELmlFXbtdcnPvkuLuyE9CZ923BUCdiac-ClDY,696
@@ -150,7 +150,7 @@ sky/provision/paperspace/instance.py,sha256=8qkvZt2-gBYCiDpT9-lztaP2_DgDdYQAAHFv
 sky/provision/paperspace/utils.py,sha256=Bl3POslZjtZU_wbBIXid7ubhRy2j5kpsesR85q7MN5w,9428
 sky/provision/runpod/__init__.py,sha256=6HYvHI27EaLrX1SS0vWVhdLu5HDBeZCdvAeDJuwM5pk,556
 sky/provision/runpod/config.py,sha256=9ulZJVL7nHuxhTdoj8D7lNn7SdicJ5zc6FIcHIG9tcg,321
-sky/provision/runpod/instance.py,sha256=kKEIY4ENSIRo8CRGVnThneVLfogdhrv1xwwCrtXwEj0,8874
+sky/provision/runpod/instance.py,sha256=ucmFQEzapbxylsl6K9EUo7bHTZYzvfECo6tpJc-MFrw,9577
 sky/provision/runpod/utils.py,sha256=ZjrcpjKzwS2nXQ21dW405PLxBl_V9awcfRjucGB3alw,6795
 sky/provision/vsphere/__init__.py,sha256=5icB8-kfs926S9DVfNJSCBVr7z7cmCEDr04-YHX89_4,788
 sky/provision/vsphere/config.py,sha256=f_ojGmi_vbnwJ8Ri48cqhZHBOuIkj41j9bFbq-ldPOo,504
@@ -217,7 +217,7 @@ sky/skylet/ray_patches/resource_demand_scheduler.py.patch,sha256=AVV-Hw-Rxw16aFm
 sky/skylet/ray_patches/updater.py.patch,sha256=ZNMGVYICPBB44jLbEx2KvCgIY7BWYdDv3-2b2HJWmAQ,289
 sky/skylet/ray_patches/worker.py.patch,sha256=_OBhibdr3xOy5Qje6Tt8D1eQVm_msi50TJbCJmOTxVU,565
 sky/templates/aws-ray.yml.j2,sha256=wijRSqqAoX8YalAvZCQf1DmfmyJgPy2DrfC33OEqFxM,7933
-sky/templates/azure-ray.yml.j2,sha256=wJM98G8sp9J63LclZp4bm30GFWwFSoBQgtN1B9j6nNU,6005
+sky/templates/azure-ray.yml.j2,sha256=RtYAcAmFQd6TB3j-pbxi7ekjWhznqFhJtzdkqH_nXqM,6135
 sky/templates/cudo-ray.yml.j2,sha256=SEHVY57iBauCOE2HYJtYVFEKlriAkdwQu_p86a1n_bA,3548
 sky/templates/fluidstack-ray.yml.j2,sha256=t8TCULgiErCZdtFmBZVsA8ZdcqR7ccwsmQhuDFTBEAU,3541
 sky/templates/gcp-ray.yml.j2,sha256=q2xSWxxYI8MVAq_mA__8FF6PwEqXCAW1SOEOGTt0qPw,9591
@@ -264,15 +264,15 @@ sky/utils/kubernetes/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hS
 sky/utils/kubernetes/create_cluster.sh,sha256=rv5Lz6AR00yBJMRyScfMSQiGKptMhtHWRsvyG20-u9c,7764
 sky/utils/kubernetes/delete_cluster.sh,sha256=BSccHF43GyepDNf-FZcenzHzpXXATkVD92vgn1lWPgk,927
 sky/utils/kubernetes/generate_kind_config.py,sha256=_TNLnifA_r7-CRq083IP1xjelYqiLjzQX9ohuqYpDH8,3187
-sky/utils/kubernetes/generate_kubeconfig.sh,sha256=jrHZRfy2oyXnyCNGy38eLUhKdjJG76CMpMQDNLhjrcQ,9018
+sky/utils/kubernetes/generate_kubeconfig.sh,sha256=AcYhuuG5jXWGHUmyRuH-oKy5qcn92gXhu6bXOt6eD6g,9274
 sky/utils/kubernetes/gpu_labeler.py,sha256=MEUv0U4ACDcNwtFVltlv017XJMjxx1Bndf6fL0i6eqg,6960
 sky/utils/kubernetes/k8s_gpu_labeler_job.yaml,sha256=KPqp23B-zQ2SZK03jdHeF9fLTogMEFxotE35IosIpRg,1186
 sky/utils/kubernetes/k8s_gpu_labeler_setup.yaml,sha256=VLKT2KKimZu1GDg_4AIlIt488oMQvhRZWwsj9vBbPUg,3812
 sky/utils/kubernetes/rsync_helper.sh,sha256=1ad-m4s8QTPxaBQxNIL8AGDfX4un6T0dxZgk-R7OLyE,154
 sky/utils/kubernetes/ssh_jump_lifecycle_manager.py,sha256=RFLJ3k7MR5UN4SKHykQ0lV9SgXumoULpKYIAt1vh-HU,6560
-skypilot_nightly-1.0.0.dev20240906.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
-skypilot_nightly-1.0.0.dev20240906.dist-info/METADATA,sha256=A4MH55bftdna18x7OpwOOY1MwWXbbPx2soeW-pMK5JI,18870
-skypilot_nightly-1.0.0.dev20240906.dist-info/WHEEL,sha256=cVxcB9AmuTcXqmwrtPhNK88dr7IR_b6qagTj0UvIEbY,91
-skypilot_nightly-1.0.0.dev20240906.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
-skypilot_nightly-1.0.0.dev20240906.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
-skypilot_nightly-1.0.0.dev20240906.dist-info/RECORD,,
+skypilot_nightly-1.0.0.dev20240907.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
+skypilot_nightly-1.0.0.dev20240907.dist-info/METADATA,sha256=Vyls5rRkU9g88Qa5qjN-C0ELMUVtX918YLMskF_FY0k,18870
+skypilot_nightly-1.0.0.dev20240907.dist-info/WHEEL,sha256=cVxcB9AmuTcXqmwrtPhNK88dr7IR_b6qagTj0UvIEbY,91
+skypilot_nightly-1.0.0.dev20240907.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
+skypilot_nightly-1.0.0.dev20240907.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
+skypilot_nightly-1.0.0.dev20240907.dist-info/RECORD,,