skypilot-nightly 1.0.0.dev20250905__py3-none-any.whl → 1.0.0.dev20251203__py3-none-any.whl

sky/skylet/log_lib.pyi

@@ -4,7 +4,7 @@ overloaded type hints for run_with_log(), as we need to determine
 the return type based on the value of require_outputs.
 """
 import typing
-from typing import Dict, List, Optional, Tuple, Union
+from typing import Dict, Iterable, Iterator, List, Optional, Tuple, Union
 
 from typing_extensions import Literal
 
@@ -42,7 +42,7 @@ class _ProcessingArgs:
         ...
 
 
-def _get_context() -> Optional[context.Context]:
+def _get_context() -> Optional[context.SkyPilotContext]:
     ...
 
 
@@ -129,8 +129,46 @@ def run_bash_command_with_log(bash_command: str,
     ...
 
 
+def run_bash_command_with_log_and_return_pid(
+        bash_command: str,
+        log_path: str,
+        env_vars: Optional[Dict[str, str]] = ...,
+        stream_logs: bool = ...,
+        with_ray: bool = ...):
+    ...
+
+
 def tail_logs(job_id: int,
               log_dir: Optional[str],
               managed_job_id: Optional[int] = ...,
               follow: bool = ...) -> None:
     ...
+
+
+def tail_logs_iter(job_id: Optional[int],
+                   log_dir: Optional[str],
+                   managed_job_id: Optional[int] = ...,
+                   follow: bool = ...,
+                   tail: int = ...) -> Iterator[str]:
+    ...
+
+
+class LogBuffer:
+    max_chars: int
+
+    def __init__(self, max_chars: int = ...):
+        ...
+
+    def flush(self) -> str:
+        ...
+
+    def write(self, line: str) -> bool:
+        ...
+
+    def close(self):
+        ...
+
+
+def buffered_iter_with_timeout(buffer: LogBuffer, iterable: Iterable[str],
+                               timeout: float) -> Iterable[str]:
+    ...

sky/skylet/providers/ibm/node_provider.py

@@ -24,7 +24,7 @@ import socket
 import threading
 import time
 from pathlib import Path
-from pprint import pprint
+from pprint import pformat, pprint
 from typing import Any, Dict, List, Optional
 from uuid import uuid4
 
@@ -67,13 +67,13 @@ def log_in_out(func):
         logger.debug(
             f"\n\nEnter {name} from {inspect.stack()[0][3]} "
             f"{inspect.stack()[1][3]} {inspect.stack()[2][3]} with args: "
-            f"entered with args:\n{pprint(args)} and kwargs {pprint(kwargs)}"
+            f"entered with args:\n{pformat(args)} and kwargs {pformat(kwargs)}"
         )
         try:
             result = func(*args, **kwargs)
             logger.debug(
                 f"Leave {name} from {inspect.stack()[1][3]} with result "
-                f"Func Result:{pprint(result)}\n\n"
+                f"Func Result:{pformat(result)}\n\n"
             )
         except Exception:
             cli_logger.error(f"Error in {name}")
@@ -445,7 +445,7 @@ class IBMVPCNodeProvider(NodeProvider):
         """returns the worker's node private ip address"""
         node = self._get_cached_node(node_id)
 
-        # if a bug ocurred, or node data was fetched before primary_ip
+        # if a bug occurred, or node data was fetched before primary_ip
         # was assigned, refetch node data from cloud.
         try:
             primary_ip = node["network_interfaces"][0].get("primary_ip")["address"]
@@ -502,8 +502,12 @@ class IBMVPCNodeProvider(NodeProvider):
 
         logger.info(f"Creating new VM instance {name}")
 
-        security_group_identity_model = {"id": self.vpc_tags["security_group_id"]}
-        subnet_identity_model = {"id": self.vpc_tags["subnet_id"]}
+        if self.vpc_tags is None:
+            raise ValueError("vpc_tags must be initialized before creating instances")
+        vpc_tags = self.vpc_tags  # Help mypy with type narrowing
+
+        security_group_identity_model = {"id": vpc_tags["security_group_id"]}
+        subnet_identity_model = {"id": vpc_tags["subnet_id"]}
         primary_network_interface = {
             "name": "eth0",
             "subnet": subnet_identity_model,
@@ -536,7 +540,7 @@ class IBMVPCNodeProvider(NodeProvider):
         instance_prototype["keys"] = [key_identity_model]
         instance_prototype["profile"] = {"name": profile_name}
         instance_prototype["resource_group"] = {"id": self.resource_group_id}
-        instance_prototype["vpc"] = {"id": self.vpc_tags["vpc_id"]}
+        instance_prototype["vpc"] = {"id": vpc_tags["vpc_id"]}
         instance_prototype["image"] = {"id": base_config["image_id"]}
 
         instance_prototype["zone"] = {"name": self.zone}
@@ -584,7 +588,7 @@ class IBMVPCNodeProvider(NodeProvider):
         floating_ip_name = f"{RAY_RECYCLABLE}-{uuid4().hex[:4]}"
         # create a new floating ip
         logger.debug(f"Creating floating IP {floating_ip_name}")
-        floating_ip_prototype = {}
+        floating_ip_prototype: Dict[str, Any] = {}
         floating_ip_prototype["name"] = floating_ip_name
         floating_ip_prototype["zone"] = {"name": self.zone}
         floating_ip_prototype["resource_group"] = {"id": self.resource_group_id}

sky/skylet/providers/ibm/vpc_provider.py

@@ -10,6 +10,7 @@ import textwrap
 import time
 import uuid
 from concurrent.futures import ThreadPoolExecutor
+from typing import Any, Dict
 
 import requests
 
@@ -173,7 +174,7 @@ class IBMVPCProvider:
                 "a subnet"
             )
 
-        subnet_prototype = {}
+        subnet_prototype: Dict[str, Any] = {}
         subnet_prototype["zone"] = {"name": zone_name}
         subnet_prototype["ip_version"] = "ipv4"
         subnet_prototype["name"] = subnet_name
@@ -186,7 +187,7 @@ class IBMVPCProvider:
 
     def create_public_gateway(self, vpc_id, zone_name, subnet_data):
 
-        gateway_prototype = {}
+        gateway_prototype: Dict[str, Any] = {}
         gateway_prototype["vpc"] = {"id": vpc_id}
         gateway_prototype["zone"] = {"name": zone_name}
         gateway_prototype["name"] = f"{subnet_data['name']}-gw"
@@ -345,7 +346,7 @@ class IBMVPCProvider:
                     return True
                 tries -= 1
                 time.sleep(sleep_interval)
-            logger.error("Failed to delete instance within the alloted time\n")
+            logger.error("Failed to delete instance within the allotted time\n")
             return False
 
         for subnet_id in self.get_vpc_subnets(vpc_data, region, field="id"):
@@ -522,7 +523,7 @@ class ClusterCleaner:
             if e.code == 404:
                 print(("VPC doesn't exist."))
                 return None
-            else: raise 
+            else: raise
 
     def delete_subnets(vpc_data):
         def _poll_subnet_exists(subnet_id):
@@ -560,12 +561,12 @@ class ClusterCleaner:
                     deleting_resource = False
                 except ibm_cloud_sdk_core.ApiException as e:
                     if e.code == 404:
-                        print("gateway doesn't exist.") 
+                        print("gateway doesn't exist.")
                         deleting_resource = False
                     if e.code == 409:
                         print("gateway still in use.")
-                        # will retry until cloud functions timeout. 
-                        time.sleep(5) 
+                        # will retry until cloud functions timeout.
+                        time.sleep(5)
 
     def delete_vms(vpc_id):
         def _poll_vpc_contains_vms(vpc_id):
@@ -586,7 +587,7 @@ class ClusterCleaner:
             )
 
         def _del_instance(vm_data):
-            # first delete ips created by node_provider 
+            # first delete ips created by node_provider
             nic_id = vm_data["network_interfaces"][0]["id"]
             res = ibm_vpc_client.list_instance_network_interface_floating_ips(
                 vm_data["id"], nic_id
@@ -598,7 +599,7 @@ class ClusterCleaner:
                     ibm_vpc_client.delete_floating_ip(ip["id"])
             print(f"Deleting VM: {vm_data['id']}")
             ibm_vpc_client.delete_instance(id=vm_data["id"])
-            
+
         res = ibm_vpc_client.list_instances(vpc_id=vpc_id).get_result()
         num_instances = res["total_count"]
 
@@ -619,12 +620,12 @@ class ClusterCleaner:
                 deleting_resource = False
             except ibm_cloud_sdk_core.ApiException as e:
                 if e.code == 404:
-                    print("VPC doesn't exist.") 
+                    print("VPC doesn't exist.")
                     deleting_resource = False
                 if e.code == 409:
                     print("VPC still in use.")
-                    # will retry until cloud functions timeout. 
-                    time.sleep(5) 
+                    # will retry until cloud functions timeout.
+                    time.sleep(5)
 
     def delete_vpc(vpc_id):
         vpc_data = get_vpc_data(vpc_id)

sky/skylet/runtime_utils.py

@@ -0,0 +1,21 @@
+"""Runtime utilities for SkyPilot."""
+import os
+
+from sky.skylet import constants
+
+
+def get_runtime_dir_path(path_suffix: str = '') -> str:
+    """Get an expanded path within the SkyPilot runtime directory.
+
+    Args:
+        path_suffix: Path suffix to join with the runtime dir
+        (e.g., '.sky/jobs.db').
+
+    Returns:
+        The full expanded path.
+    """
+    runtime_dir = os.path.expanduser(
+        os.environ.get(constants.SKY_RUNTIME_DIR_ENV_VAR_KEY, '~'))
+    if path_suffix:
+        return os.path.join(runtime_dir, path_suffix)
+    return runtime_dir