File size: 10,054 Bytes
9c6594c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 |
"""The layer between launch sdk user code and the wandb internal process.
If there is an active run this communication is done through the wandb run's
backend interface.
If there is no active run, the messages are staged on the StagedLaunchInputs
singleton and sent when a run is created.
"""
import os
import pathlib
import shutil
import tempfile
from typing import Any, Dict, List, Optional
import wandb
import wandb.data_types
from wandb.sdk.launch.errors import LaunchError
from wandb.sdk.launch.inputs.schema import META_SCHEMA
from wandb.sdk.wandb_run import Run
from wandb.util import get_module
from .files import config_path_is_valid, override_file
PERIOD = "."
BACKSLASH = "\\"
LAUNCH_MANAGED_CONFIGS_DIR = "_wandb_configs"
class ConfigTmpDir:
"""Singleton for managing temporary directories for configuration files.
Any configuration files designated as inputs to a launch job are copied to
a temporary directory. This singleton manages the temporary directory and
provides paths to the configuration files.
"""
_instance = None
def __new__(cls):
if cls._instance is None:
cls._instance = object.__new__(cls)
return cls._instance
def __init__(self):
if not hasattr(self, "_tmp_dir"):
self._tmp_dir = tempfile.mkdtemp()
self._configs_dir = os.path.join(self._tmp_dir, LAUNCH_MANAGED_CONFIGS_DIR)
os.mkdir(self._configs_dir)
@property
def tmp_dir(self):
return pathlib.Path(self._tmp_dir)
@property
def configs_dir(self):
return pathlib.Path(self._configs_dir)
class JobInputArguments:
"""Arguments for the publish_job_input of Interface."""
def __init__(
self,
include: Optional[List[str]] = None,
exclude: Optional[List[str]] = None,
schema: Optional[dict] = None,
file_path: Optional[str] = None,
run_config: Optional[bool] = None,
):
self.include = include
self.exclude = exclude
self.schema = schema
self.file_path = file_path
self.run_config = run_config
class StagedLaunchInputs:
_instance = None
def __new__(cls):
if cls._instance is None:
cls._instance = object.__new__(cls)
return cls._instance
def __init__(self) -> None:
if not hasattr(self, "_staged_inputs"):
self._staged_inputs: List[JobInputArguments] = []
def add_staged_input(
self,
input_arguments: JobInputArguments,
):
self._staged_inputs.append(input_arguments)
def apply(self, run: Run):
"""Apply the staged inputs to the given run."""
for input in self._staged_inputs:
_publish_job_input(input, run)
def _publish_job_input(
input: JobInputArguments,
run: Run,
) -> None:
"""Publish a job input to the backend interface of the given run.
Arguments:
input (JobInputArguments): The arguments for the job input.
run (Run): The run to publish the job input to.
"""
assert run._backend is not None
assert run._backend.interface is not None
assert input.run_config is not None
interface = run._backend.interface
if input.file_path:
config_dir = ConfigTmpDir()
dest = os.path.join(config_dir.configs_dir, input.file_path)
run.save(dest, base_path=config_dir.tmp_dir)
interface.publish_job_input(
include_paths=[_split_on_unesc_dot(path) for path in input.include]
if input.include
else [],
exclude_paths=[_split_on_unesc_dot(path) for path in input.exclude]
if input.exclude
else [],
input_schema=input.schema,
run_config=input.run_config,
file_path=input.file_path or "",
)
def _replace_refs_and_allofs(schema: dict, defs: Optional[dict]) -> dict:
"""Recursively fix JSON schemas with common issues.
1. Replaces any instances of $ref with their associated definition in defs
2. Removes any "allOf" lists that only have one item, "lifting" the item up
See test_internal.py for examples
"""
ret: Dict[str, Any] = {}
if "$ref" in schema and defs:
# Reference found, replace it with its definition
def_key = schema.pop("$ref").split("#/$defs/")[1]
# Also run recursive replacement in case a ref contains more refs
ret = _replace_refs_and_allofs(defs[def_key], defs)
for key, val in schema.items():
if isinstance(val, dict):
# Step into dicts recursively
new_val_dict = _replace_refs_and_allofs(val, defs)
ret[key] = new_val_dict
elif isinstance(val, list):
# Step into each item in the list
new_val_list = []
for item in val:
if isinstance(item, dict):
new_val_list.append(_replace_refs_and_allofs(item, defs))
else:
new_val_list.append(item)
# Lift up allOf blocks with only one item
if (
key == "allOf"
and len(new_val_list) == 1
and isinstance(new_val_list[0], dict)
):
ret.update(new_val_list[0])
else:
ret[key] = new_val_list
else:
# For anything else (str, int, etc) keep it as-is
ret[key] = val
return ret
def _validate_schema(schema: dict) -> None:
jsonschema = get_module(
"jsonschema",
required="Setting job schema requires the jsonschema package. Please install it with `pip install 'wandb[launch]'`.",
lazy=False,
)
validator = jsonschema.Draft202012Validator(META_SCHEMA)
errs = sorted(validator.iter_errors(schema), key=str)
if errs:
wandb.termwarn(f"Schema includes unhandled or invalid configurations:\n{errs}")
def handle_config_file_input(
path: str,
include: Optional[List[str]] = None,
exclude: Optional[List[str]] = None,
schema: Optional[Any] = None,
):
"""Declare an overridable configuration file for a launch job.
The configuration file is copied to a temporary directory and the path to
the copy is sent to the backend interface of the active run and used to
configure the job builder.
If there is no active run, the configuration file is staged and sent when a
run is created.
"""
config_path_is_valid(path)
override_file(path)
tmp_dir = ConfigTmpDir()
dest = os.path.join(tmp_dir.configs_dir, path)
dest_dir = os.path.dirname(dest)
if not os.path.exists(dest_dir):
os.makedirs(dest_dir)
shutil.copy(
path,
dest,
)
if schema:
# This supports both an instance of a pydantic BaseModel class (e.g. schema=MySchema(...))
# or the BaseModel class itself (e.g. schema=MySchema)
if hasattr(schema, "model_json_schema") and callable(
schema.model_json_schema # type: ignore
):
schema = schema.model_json_schema()
if not isinstance(schema, dict):
raise LaunchError(
"schema must be a dict, Pydantic model instance, or Pydantic model class."
)
defs = schema.pop("$defs", None)
schema = _replace_refs_and_allofs(schema, defs)
_validate_schema(schema)
arguments = JobInputArguments(
include=include,
exclude=exclude,
schema=schema,
file_path=path,
run_config=False,
)
if wandb.run is not None:
_publish_job_input(arguments, wandb.run)
else:
staged_inputs = StagedLaunchInputs()
staged_inputs.add_staged_input(arguments)
def handle_run_config_input(
include: Optional[List[str]] = None,
exclude: Optional[List[str]] = None,
schema: Optional[Any] = None,
):
"""Declare wandb.config as an overridable configuration for a launch job.
The include and exclude paths are sent to the backend interface of the
active run and used to configure the job builder.
If there is no active run, the include and exclude paths are staged and sent
when a run is created.
"""
if schema:
# This supports both an instance of a pydantic BaseModel class (e.g. schema=MySchema(...))
# or the BaseModel class itself (e.g. schema=MySchema)
if hasattr(schema, "model_json_schema") and callable(
schema.model_json_schema # type: ignore
):
schema = schema.model_json_schema()
if not isinstance(schema, dict):
raise LaunchError(
"schema must be a dict, Pydantic model instance, or Pydantic model class."
)
defs = schema.pop("$defs", None)
schema = _replace_refs_and_allofs(schema, defs)
_validate_schema(schema)
arguments = JobInputArguments(
include=include,
exclude=exclude,
schema=schema,
run_config=True,
file_path=None,
)
if wandb.run is not None:
_publish_job_input(arguments, wandb.run)
else:
stage_inputs = StagedLaunchInputs()
stage_inputs.add_staged_input(arguments)
def _split_on_unesc_dot(path: str) -> List[str]:
r"""Split a string on unescaped dots.
Arguments:
path (str): The string to split.
Raises:
ValueError: If the path has a trailing escape character.
Returns:
List[str]: The split string.
"""
parts = []
part = ""
i = 0
while i < len(path):
if path[i] == BACKSLASH:
if i == len(path) - 1:
raise LaunchError(
f"Invalid config path {path}: trailing {BACKSLASH}.",
)
if path[i + 1] == PERIOD:
part += PERIOD
i += 2
elif path[i] == PERIOD:
parts.append(part)
part = ""
i += 1
else:
part += path[i]
i += 1
if part:
parts.append(part)
return parts
|