[UX] minor optimizations for launch and introduce py-spy (#4495)

* [UX] minor optimizations for launch and introduce py-spy

Signed-off-by: Aylei <[email protected]>

* fix

Signed-off-by: Aylei <[email protected]>

---------

Signed-off-by: Aylei <[email protected]>

CONTRIBUTING.md

@@ -50,9 +50,10 @@ pytest tests/test_smoke.py --generic-cloud azure
 
 For profiling code, use:
 ```
-pip install tuna # Tuna is used for visualization of profiling data.
-python3 -m cProfile -o sky.prof -m sky.cli status # Or some other command
-tuna sky.prof
+pip install py-spy # py-spy is a sampling profiler for Python programs
+py-spy record -t -o sky.svg -- python -m sky.cli status # Or some other command
+py-spy top -- python -m sky.cli status # Get a live top view
+py-spy -h # For more options
 ```
 
 #### Testing in a container

sky/adaptors/common.py

@@ -1,6 +1,7 @@
 """Lazy import for modules to avoid import error when not used."""
 import functools
 import importlib
+import threading
 from typing import Any, Callable, Optional, Tuple
 
 
@@ -24,17 +25,22 @@ def __init__(self,
         self._module = None
         self._import_error_message = import_error_message
         self._set_loggers = set_loggers
+        self._lock = threading.RLock()
 
     def load_module(self):
-        if self._module is None:
-            try:
-                self._module = importlib.import_module(self._module_name)
-                if self._set_loggers is not None:
-                    self._set_loggers()
-            except ImportError as e:
-                if self._import_error_message is not None:
-                    raise ImportError(self._import_error_message) from e
-                raise
+        # Avoid extra imports when multiple threads try to import the same
+        # module. The overhead is minor since import can only run in serial
+        # due to GIL even in multi-threaded environments.
+        with self._lock:
+            if self._module is None:
+                try:
+                    self._module = importlib.import_module(self._module_name)
+                    if self._set_loggers is not None:
+                        self._set_loggers()
+                except ImportError as e:
+                    if self._import_error_message is not None:
+                        raise ImportError(self._import_error_message) from e
+                    raise
         return self._module
 
     def __getattr__(self, name: str) -> Any:

sky/usage/usage_lib.py

@@ -3,7 +3,6 @@
 import contextlib
 import datetime
 import enum
-import inspect
 import json
 import os
 import time
@@ -12,19 +11,28 @@
 from typing import Any, Callable, Dict, List, Optional, Union
 
 import click
-import requests
 
 import sky
 from sky import sky_logging
+from sky.adaptors import common as adaptors_common
 from sky.usage import constants
 from sky.utils import common_utils
 from sky.utils import env_options
 from sky.utils import ux_utils
 
 if typing.TYPE_CHECKING:
+    import inspect
+
+    import requests
+
     from sky import resources as resources_lib
     from sky import status_lib
     from sky import task as task_lib
+else:
+    # requests and inspect cost ~100ms to load, which can be postponed to
+    # collection phase or skipped if user specifies no collection
+    requests = adaptors_common.LazyImport('requests')
+    inspect = adaptors_common.LazyImport('inspect')
 
 logger = sky_logging.init_logger(__name__)
 

sky/utils/log_utils.py

@@ -5,6 +5,8 @@
 from typing import Callable, Iterator, List, Optional, TextIO, Type
 
 import colorama
+# slow due to https://github.com/python-pendulum/pendulum/issues/808
+# FIXME(aylei): bump pendulum if it get fixed
 import pendulum
 import prettytable