snowflake-ml-python 1.15.0__py3-none-any.whl → 1.17.0__py3-none-any.whl

snowflake/ml/jobs/_interop/protocols.py

@@ -0,0 +1,471 @@
+import base64
+import logging
+import pickle
+import posixpath
+import sys
+from typing import Any, Callable, Optional, Protocol, Union, cast, runtime_checkable
+
+from snowflake import snowpark
+from snowflake.ml.jobs._interop import data_utils
+from snowflake.ml.jobs._interop.dto_schema import (
+    BinaryManifest,
+    ParquetManifest,
+    ProtocolInfo,
+)
+
+Condition = Union[type, tuple[type, ...], Callable[[Any], bool], None]
+
+logger = logging.getLogger(__name__)
+
+
+class SerializationError(TypeError):
+    """Exception raised when a serialization protocol fails."""
+
+
+class DeserializationError(ValueError):
+    """Exception raised when a serialization protocol fails."""
+
+
+class InvalidPayloadError(DeserializationError):
+    """Exception raised when the payload is invalid."""
+
+
+class ProtocolMismatchError(DeserializationError):
+    """Exception raised when the protocol of the serialization protocol is incompatible."""
+
+
+class VersionMismatchError(ProtocolMismatchError):
+    """Exception raised when the version of the serialization protocol is incompatible."""
+
+
+class ProtocolNotFoundError(SerializationError):
+    """Exception raised when no suitable serialization protocol is available."""
+
+
+@runtime_checkable
+class SerializationProtocol(Protocol):
+    """
+    More advanced protocol which supports more flexibility in how results are saved or loaded.
+    Results can be saved as one or more files, or directly inline in the PayloadManifest.
+    If saving as files, the PayloadManifest can save arbitrary "manifest" information.
+    """
+
+    @property
+    def supported_types(self) -> Condition:
+        """The types that the protocol supports."""
+
+    @property
+    def protocol_info(self) -> ProtocolInfo:
+        """The information about the protocol."""
+
+    def save(self, obj: Any, dest_dir: str, session: Optional[snowpark.Session] = None) -> ProtocolInfo:
+        """Save the object to the destination directory."""
+
+    def load(
+        self,
+        payload_info: ProtocolInfo,
+        session: Optional[snowpark.Session] = None,
+        path_transform: Optional[Callable[[str], str]] = None,
+    ) -> Any:
+        """Load the object from the source directory."""
+
+
+class CloudPickleProtocol(SerializationProtocol):
+    """
+    CloudPickle serialization protocol.
+    Uses BinaryManifest for manifest schema.
+    """
+
+    DEFAULT_PATH = "mljob_extra.pkl"
+
+    def __init__(self) -> None:
+        import cloudpickle as cp
+
+        self._backend = cp
+
+    def _get_compatibility_error(self, payload_info: ProtocolInfo) -> Optional[Exception]:
+        """Check compatibility and attempt load, raising helpful errors on failure."""
+        version_error = python_error = None
+
+        # Check cloudpickle version compatibility
+        if payload_info.version:
+            try:
+                from packaging import version
+
+                payload_major, current_major = (
+                    version.parse(payload_info.version).major,
+                    version.parse(self._backend.__version__).major,
+                )
+                if payload_major != current_major:
+                    version_error = "cloudpickle version mismatch: payload={}, current={}".format(
+                        payload_info.version, self._backend.__version__
+                    )
+            except Exception:
+                if payload_info.version != self.protocol_info.version:
+                    version_error = "cloudpickle version mismatch: payload={}, current={}".format(
+                        payload_info.version, self.protocol_info.version
+                    )
+
+        # Check Python version compatibility
+        if payload_info.metadata and "python_version" in payload_info.metadata:
+            payload_py, current_py = (
+                payload_info.metadata["python_version"],
+                f"{sys.version_info.major}.{sys.version_info.minor}",
+            )
+            if payload_py != current_py:
+                python_error = f"Python version mismatch: payload={payload_py}, current={current_py}"
+
+        if version_error or python_error:
+            errors = [err for err in [version_error, python_error] if err]
+            return VersionMismatchError(f"Load failed due to incompatibility: {'; '.join(errors)}")
+        return None
+
+    @property
+    def supported_types(self) -> Condition:
+        return None  # All types are supported
+
+    @property
+    def protocol_info(self) -> ProtocolInfo:
+        return ProtocolInfo(
+            name="cloudpickle",
+            version=self._backend.__version__,
+            metadata={
+                "python_version": f"{sys.version_info.major}.{sys.version_info.minor}",
+            },
+        )
+
+    def save(self, obj: Any, dest_dir: str, session: Optional[snowpark.Session] = None) -> ProtocolInfo:
+        """Save the object to the destination directory."""
+        result_path = posixpath.join(dest_dir, self.DEFAULT_PATH)
+        with data_utils.open_stream(result_path, "wb", session=session) as f:
+            self._backend.dump(obj, f)
+        manifest: BinaryManifest = {"path": result_path}
+        return self.protocol_info.with_manifest(manifest)
+
+    def load(
+        self,
+        payload_info: ProtocolInfo,
+        session: Optional[snowpark.Session] = None,
+        path_transform: Optional[Callable[[str], str]] = None,
+    ) -> Any:
+        """Load the object from the source directory."""
+        if payload_info.name != self.protocol_info.name:
+            raise ProtocolMismatchError(
+                f"Invalid payload protocol: expected '{self.protocol_info.name}', got '{payload_info.name}'"
+            )
+
+        payload_manifest = cast(BinaryManifest, payload_info.manifest)
+        try:
+            if payload_bytes := payload_manifest.get("bytes"):
+                return self._backend.loads(payload_bytes)
+            if payload_b64 := payload_manifest.get("base64"):
+                return self._backend.loads(base64.b64decode(payload_b64))
+            result_path = path_transform(payload_manifest["path"]) if path_transform else payload_manifest["path"]
+            with data_utils.open_stream(result_path, "rb", session=session) as f:
+                return self._backend.load(f)
+        except (
+            pickle.UnpicklingError,
+            TypeError,
+            AttributeError,
+            MemoryError,
+        ) as pickle_error:
+            if error := self._get_compatibility_error(payload_info):
+                raise error from pickle_error
+            raise
+
+
+class ArrowTableProtocol(SerializationProtocol):
+    """
+    Arrow Table serialization protocol.
+    Uses ParquetManifest for manifest schema.
+    """
+
+    DEFAULT_PATH_PATTERN = "mljob_extra_{0}.parquet"
+
+    def __init__(self) -> None:
+        import pyarrow as pa
+        import pyarrow.parquet as pq
+
+        self._pa = pa
+        self._pq = pq
+
+    @property
+    def supported_types(self) -> Condition:
+        return cast(type, self._pa.Table)
+
+    @property
+    def protocol_info(self) -> ProtocolInfo:
+        return ProtocolInfo(
+            name="pyarrow",
+            version=self._pa.__version__,
+        )
+
+    def save(self, obj: Any, dest_dir: str, session: Optional[snowpark.Session] = None) -> ProtocolInfo:
+        """Save the object to the destination directory."""
+        if not isinstance(obj, self._pa.Table):
+            raise SerializationError(f"Expected {self._pa.Table.__name__} object, got {type(obj).__name__}")
+
+        # TODO: Support partitioned writes for large datasets
+        result_path = posixpath.join(dest_dir, self.DEFAULT_PATH_PATTERN.format(0))
+        with data_utils.open_stream(result_path, "wb", session=session) as stream:
+            self._pq.write_table(obj, stream)
+
+        manifest: ParquetManifest = {"paths": [result_path]}
+        return self.protocol_info.with_manifest(manifest)
+
+    def load(
+        self,
+        payload_info: ProtocolInfo,
+        session: Optional[snowpark.Session] = None,
+        path_transform: Optional[Callable[[str], str]] = None,
+    ) -> Any:
+        """Load the object from the source directory."""
+        if payload_info.name != self.protocol_info.name:
+            raise ProtocolMismatchError(
+                f"Invalid payload protocol: expected '{self.protocol_info.name}', got '{payload_info.name}'"
+            )
+
+        payload_manifest = cast(ParquetManifest, payload_info.manifest)
+        tables = []
+        for path in payload_manifest["paths"]:
+            transformed_path = path_transform(path) if path_transform else path
+            with data_utils.open_stream(transformed_path, "rb", session=session) as f:
+                table = self._pq.read_table(f)
+                tables.append(table)
+        return self._pa.concat_tables(tables) if len(tables) > 1 else tables[0]
+
+
+class PandasDataFrameProtocol(SerializationProtocol):
+    """
+    Pandas DataFrame serialization protocol.
+    Uses ParquetManifest for manifest schema.
+    """
+
+    DEFAULT_PATH_PATTERN = "mljob_extra_{0}.parquet"
+
+    def __init__(self) -> None:
+        import pandas as pd
+
+        self._pd = pd
+
+    @property
+    def supported_types(self) -> Condition:
+        return cast(type, self._pd.DataFrame)
+
+    @property
+    def protocol_info(self) -> ProtocolInfo:
+        return ProtocolInfo(
+            name="pandas",
+            version=self._pd.__version__,
+        )
+
+    def save(self, obj: Any, dest_dir: str, session: Optional[snowpark.Session] = None) -> ProtocolInfo:
+        """Save the object to the destination directory."""
+        if not isinstance(obj, self._pd.DataFrame):
+            raise SerializationError(f"Expected {self._pd.DataFrame.__name__} object, got {type(obj).__name__}")
+
+        # TODO: Support partitioned writes for large datasets
+        result_path = posixpath.join(dest_dir, self.DEFAULT_PATH_PATTERN.format(0))
+        with data_utils.open_stream(result_path, "wb", session=session) as stream:
+            obj.to_parquet(stream)
+
+        manifest: ParquetManifest = {"paths": [result_path]}
+        return self.protocol_info.with_manifest(manifest)
+
+    def load(
+        self,
+        payload_info: ProtocolInfo,
+        session: Optional[snowpark.Session] = None,
+        path_transform: Optional[Callable[[str], str]] = None,
+    ) -> Any:
+        """Load the object from the source directory."""
+        if payload_info.name != self.protocol_info.name:
+            raise ProtocolMismatchError(
+                f"Invalid payload protocol: expected '{self.protocol_info.name}', got '{payload_info.name}'"
+            )
+
+        payload_manifest = cast(ParquetManifest, payload_info.manifest)
+        dfs = []
+        for path in payload_manifest["paths"]:
+            transformed_path = path_transform(path) if path_transform else path
+            with data_utils.open_stream(transformed_path, "rb", session=session) as f:
+                df = self._pd.read_parquet(f)
+                dfs.append(df)
+        return self._pd.concat(dfs) if len(dfs) > 1 else dfs[0]
+
+
+class NumpyArrayProtocol(SerializationProtocol):
+    """
+    Numpy Array serialization protocol.
+    Uses BinaryManifest for manifest schema.
+    """
+
+    DEFAULT_PATH_PATTERN = "mljob_extra.npy"
+
+    def __init__(self) -> None:
+        import numpy as np
+
+        self._np = np
+
+    @property
+    def supported_types(self) -> Condition:
+        return cast(type, self._np.ndarray)
+
+    @property
+    def protocol_info(self) -> ProtocolInfo:
+        return ProtocolInfo(
+            name="numpy",
+            version=self._np.__version__,
+        )
+
+    def save(self, obj: Any, dest_dir: str, session: Optional[snowpark.Session] = None) -> ProtocolInfo:
+        """Save the object to the destination directory."""
+        if not isinstance(obj, self._np.ndarray):
+            raise SerializationError(f"Expected {self._np.ndarray.__name__} object, got {type(obj).__name__}")
+        result_path = posixpath.join(dest_dir, self.DEFAULT_PATH_PATTERN)
+        with data_utils.open_stream(result_path, "wb", session=session) as stream:
+            self._np.save(stream, obj)
+
+        manifest: BinaryManifest = {"path": result_path}
+        return self.protocol_info.with_manifest(manifest)
+
+    def load(
+        self,
+        payload_info: ProtocolInfo,
+        session: Optional[snowpark.Session] = None,
+        path_transform: Optional[Callable[[str], str]] = None,
+    ) -> Any:
+        """Load the object from the source directory."""
+        if payload_info.name != self.protocol_info.name:
+            raise ProtocolMismatchError(
+                f"Invalid payload protocol: expected '{self.protocol_info.name}', got '{payload_info.name}'"
+            )
+
+        payload_manifest = cast(BinaryManifest, payload_info.manifest)
+        transformed_path = path_transform(payload_manifest["path"]) if path_transform else payload_manifest["path"]
+        with data_utils.open_stream(transformed_path, "rb", session=session) as f:
+            return self._np.load(f)
+
+
+class AutoProtocol(SerializationProtocol):
+    def __init__(self) -> None:
+        self._protocols: list[SerializationProtocol] = []
+        self._protocol_info = ProtocolInfo(
+            name="auto",
+            version=None,
+            metadata=None,
+        )
+
+    @property
+    def supported_types(self) -> Condition:
+        return None  # All types are supported
+
+    @property
+    def protocol_info(self) -> ProtocolInfo:
+        return self._protocol_info
+
+    def try_register_protocol(
+        self,
+        klass: type[SerializationProtocol],
+        *args: Any,
+        index: int = 0,
+        **kwargs: Any,
+    ) -> None:
+        """
+        Try to construct and register a protocol. If the protocol cannot be constructed,
+        log a warning and skip registration. By default (index=0), the most recently
+        registered protocol takes precedence.
+
+        Args:
+            klass: The class of the protocol to register.
+            args: The positional arguments to pass to the protocol constructor.
+            index: The index to register the protocol at. If -1, the protocol is registered at the end of the list.
+            kwargs: The keyword arguments to pass to the protocol constructor.
+        """
+        try:
+            protocol = klass(*args, **kwargs)
+            self.register_protocol(protocol, index=index)
+        except Exception as e:
+            logger.warning(f"Failed to register protocol {klass}: {e}")
+
+    def register_protocol(
+        self,
+        protocol: SerializationProtocol,
+        index: int = 0,
+    ) -> None:
+        """
+        Register a protocol with a condition. By default (index=0), the most recently
+        registered protocol takes precedence.
+
+        Args:
+            protocol: The protocol to register.
+            index: The index to register the protocol at. If -1, the protocol is registered at the end of the list.
+
+        Raises:
+            ValueError: If the condition is invalid.
+            ValueError: If the index is invalid.
+        """
+        # Validate condition
+        # TODO: Build lookup table of supported types to protocols (in priority order)
+        # for faster lookup at save/load time (instead of iterating over all protocols)
+        if not isinstance(protocol, SerializationProtocol):
+            raise ValueError(f"Invalid protocol type: {type(protocol)}. Expected SerializationProtocol.")
+        if index == -1:
+            self._protocols.append(protocol)
+        elif index < 0:
+            raise ValueError(f"Invalid index: {index}. Expected -1 or >= 0.")
+        else:
+            self._protocols.insert(index, protocol)
+
+    def save(self, obj: Any, dest_dir: str, session: Optional[snowpark.Session] = None) -> ProtocolInfo:
+        """Save the object to the destination directory."""
+        last_protocol_error = None
+        for protocol in self._protocols:
+            try:
+                if self._is_supported_type(obj, protocol):
+                    logger.debug(f"Dumping object of type {type(obj)} with protocol {protocol}")
+                    return protocol.save(obj, dest_dir, session)
+            except Exception as e:
+                logger.warning(f"Error dumping object {obj} with protocol {protocol}: {repr(e)}")
+                last_protocol_error = (protocol.protocol_info, e)
+        last_error_str = (
+            f", most recent error ({last_protocol_error[0]}): {repr(last_protocol_error[1])}"
+            if last_protocol_error
+            else ""
+        )
+        raise ProtocolNotFoundError(
+            f"No suitable protocol found for type {type(obj).__name__}"
+            f" (available: {', '.join(str(p.protocol_info) for p in self._protocols)}){last_error_str}"
+        )
+
+    def load(
+        self,
+        payload_info: ProtocolInfo,
+        session: Optional[snowpark.Session] = None,
+        path_transform: Optional[Callable[[str], str]] = None,
+    ) -> Any:
+        """Load the object from the source directory."""
+        last_error = None
+        for protocol in self._protocols:
+            if protocol.protocol_info.name == payload_info.name:
+                try:
+                    return protocol.load(payload_info, session, path_transform)
+                except Exception as e:
+                    logger.warning(f"Error loading object with protocol {protocol}: {repr(e)}")
+                    last_error = e
+        if last_error:
+            raise last_error
+        raise ProtocolNotFoundError(
+            f"No protocol matching {payload_info} available"
+            f" (available: {', '.join(str(p.protocol_info) for p in self._protocols)})"
+            ", possibly due to snowflake-ml-python package version mismatch"
+        )
+
+    def _is_supported_type(self, obj: Any, protocol: SerializationProtocol) -> bool:
+        if protocol.supported_types is None:
+            return True  # None means all types are supported
+        elif isinstance(protocol.supported_types, (type, tuple)):
+            return isinstance(obj, protocol.supported_types)
+        elif callable(protocol.supported_types):
+            return protocol.supported_types(obj) is True
+        raise ValueError(f"Invalid supported types: {protocol.supported_types} for protocol {protocol}")

snowflake/ml/jobs/_interop/results.py

@@ -0,0 +1,51 @@
+from dataclasses import dataclass
+from typing import Any, Optional
+
+
+@dataclass(frozen=True)
+class ExecutionResult:
+    """
+    A result of a job execution.
+
+    Args:
+        success: Whether the execution was successful.
+        value: The value of the execution.
+    """
+
+    success: bool
+    value: Any
+
+    def get_value(self, wrap_exceptions: bool = True) -> Any:
+        if not self.success:
+            assert isinstance(self.value, BaseException), "Unexpected non-exception value for failed result"
+            self._raise_exception(self.value, wrap_exceptions)
+        return self.value
+
+    def _raise_exception(self, exception: BaseException, wrap_exceptions: bool) -> None:
+        if wrap_exceptions:
+            raise RuntimeError(f"Job execution failed with error: {exception!r}") from exception
+        else:
+            raise exception
+
+
+@dataclass(frozen=True)
+class LoadedExecutionResult(ExecutionResult):
+    """
+    A result of a job execution that has been loaded from a file.
+    """
+
+    load_error: Optional[Exception] = None
+    result_metadata: Optional[dict[str, Any]] = None
+
+    def get_value(self, wrap_exceptions: bool = True) -> Any:
+        if not self.success:
+            # Raise the original exception if available, otherwise raise the load error
+            ex = self.value
+            if not isinstance(ex, BaseException):
+                ex = RuntimeError(f"Unknown error {ex or ''}")
+                ex.__cause__ = self.load_error
+            self._raise_exception(ex, wrap_exceptions)
+        else:
+            if self.load_error:
+                raise ValueError("Job execution succeeded but result retrieval failed") from self.load_error
+            return self.value

snowflake/ml/jobs/_interop/utils.py

@@ -0,0 +1,144 @@
+import logging
+import os
+import traceback
+from pathlib import PurePath
+from typing import Any, Callable, Optional
+
+import pydantic
+
+from snowflake import snowpark
+from snowflake.ml.jobs._interop import data_utils, exception_utils, legacy, protocols
+from snowflake.ml.jobs._interop.dto_schema import (
+    ExceptionMetadata,
+    ResultDTO,
+    ResultMetadata,
+)
+from snowflake.ml.jobs._interop.results import ExecutionResult, LoadedExecutionResult
+from snowflake.snowpark import exceptions as sp_exceptions
+
+DEFAULT_CODEC = data_utils.JsonDtoCodec
+DEFAULT_PROTOCOL = protocols.AutoProtocol()
+DEFAULT_PROTOCOL.try_register_protocol(protocols.CloudPickleProtocol)
+DEFAULT_PROTOCOL.try_register_protocol(protocols.ArrowTableProtocol)
+DEFAULT_PROTOCOL.try_register_protocol(protocols.PandasDataFrameProtocol)
+DEFAULT_PROTOCOL.try_register_protocol(protocols.NumpyArrayProtocol)
+
+
+logger = logging.getLogger(__name__)
+
+
+def save_result(result: ExecutionResult, path: str, session: Optional[snowpark.Session] = None) -> None:
+    """
+    Save the result to a file.
+    """
+    result_dto = ResultDTO(
+        success=result.success,
+        value=result.value,
+    )
+
+    try:
+        # Try to encode result directly
+        payload = DEFAULT_CODEC.encode(result_dto)
+    except TypeError:
+        result_dto.value = None  # Remove raw value to avoid serialization error
+        result_dto.metadata = _get_metadata(result.value)  # Add metadata for client fallback on protocol mismatch
+        try:
+            path_dir = PurePath(path).parent.as_posix()
+            protocol_info = DEFAULT_PROTOCOL.save(result.value, path_dir, session=session)
+            result_dto.protocol = protocol_info
+
+        except Exception as e:
+            logger.warning(f"Error dumping result value: {repr(e)}")
+            result_dto.serialize_error = repr(e)
+
+        # Encode the modified result DTO
+        payload = DEFAULT_CODEC.encode(result_dto)
+
+    with data_utils.open_stream(path, "wb", session=session) as stream:
+        stream.write(payload)
+
+
+def load_result(
+    path: str, session: Optional[snowpark.Session] = None, path_transform: Optional[Callable[[str], str]] = None
+) -> ExecutionResult:
+    """Load the result from a file on a Snowflake stage."""
+    try:
+        with data_utils.open_stream(path, "r", session=session) as stream:
+            # Load the DTO as a dict for easy fallback to legacy loading if necessary
+            dto_dict = DEFAULT_CODEC.decode(stream, as_dict=True)
+    except UnicodeDecodeError:
+        # Path may be a legacy result file (cloudpickle)
+        # TODO: Re-use the stream
+        assert session is not None
+        return legacy.load_legacy_result(session, path)
+
+    try:
+        dto = ResultDTO.model_validate(dto_dict)
+    except pydantic.ValidationError as e:
+        if "success" in dto_dict:
+            assert session is not None
+            if path.endswith(".json"):
+                path = os.path.splitext(path)[0] + ".pkl"
+            return legacy.load_legacy_result(session, path, result_json=dto_dict)
+        raise ValueError("Invalid result schema") from e
+
+    # Try loading data from file using the protocol info
+    result_value = None
+    data_load_error = None
+    if dto.protocol is not None:
+        try:
+            logger.debug(f"Loading result value with protocol {dto.protocol}")
+            result_value = DEFAULT_PROTOCOL.load(dto.protocol, session=session, path_transform=path_transform)
+        except sp_exceptions.SnowparkSQLException:
+            raise  # Data retrieval errors should be bubbled up
+        except Exception as e:
+            logger.debug(f"Error loading result value with protocol {dto.protocol}: {repr(e)}")
+            data_load_error = e
+
+    # Wrap serialize_error in a TypeError
+    if dto.serialize_error:
+        serialize_error = TypeError("Original result serialization failed with error: " + dto.serialize_error)
+        if data_load_error:
+            data_load_error.__context__ = serialize_error
+        else:
+            data_load_error = serialize_error
+
+    # Prepare to assemble the final result
+    result_value = result_value if result_value is not None else dto.value
+    if not dto.success and result_value is None:
+        # Try to reconstruct exception from metadata if available
+        if isinstance(dto.metadata, ExceptionMetadata):
+            logger.debug(f"Reconstructing exception from metadata {dto.metadata}")
+            result_value = exception_utils.build_exception(
+                type_str=dto.metadata.type,
+                message=dto.metadata.message,
+                traceback=dto.metadata.traceback,
+                original_repr=dto.metadata.repr,
+            )
+
+        # Generate a generic error if we still don't have a value,
+        # attaching the data load error if any
+        if result_value is None:
+            result_value = exception_utils.RemoteError("Unknown remote error")
+            result_value.__cause__ = data_load_error
+
+    return LoadedExecutionResult(
+        success=dto.success,
+        value=result_value,
+        load_error=data_load_error,
+    )
+
+
+def _get_metadata(value: Any) -> ResultMetadata:
+    type_name = f"{type(value).__module__}.{type(value).__name__}"
+    if isinstance(value, BaseException):
+        return ExceptionMetadata(
+            type=type_name,
+            repr=repr(value),
+            message=str(value),
+            traceback="".join(traceback.format_tb(value.__traceback__)),
+        )
+    return ResultMetadata(
+        type=type_name,
+        repr=repr(value),
+    )

snowflake/ml/jobs/_utils/constants.py

@@ -12,6 +12,9 @@ PAYLOAD_DIR_ENV_VAR = "MLRS_PAYLOAD_DIR"
 RESULT_PATH_ENV_VAR = "MLRS_RESULT_PATH"
 MIN_INSTANCES_ENV_VAR = "MLRS_MIN_INSTANCES"
 TARGET_INSTANCES_ENV_VAR = "SNOWFLAKE_JOBS_COUNT"
+INSTANCES_MIN_WAIT_ENV_VAR = "MLRS_INSTANCES_MIN_WAIT"
+INSTANCES_TIMEOUT_ENV_VAR = "MLRS_INSTANCES_TIMEOUT"
+INSTANCES_CHECK_INTERVAL_ENV_VAR = "MLRS_INSTANCES_CHECK_INTERVAL"
 RUNTIME_IMAGE_TAG_ENV_VAR = "MLRS_CONTAINER_IMAGE_TAG"
 
 # Stage mount paths
@@ -19,7 +22,7 @@ STAGE_VOLUME_MOUNT_PATH = "/mnt/job_stage"
 APP_STAGE_SUBPATH = "app"
 SYSTEM_STAGE_SUBPATH = "system"
 OUTPUT_STAGE_SUBPATH = "output"
-RESULT_PATH_DEFAULT_VALUE = f"{OUTPUT_STAGE_SUBPATH}/mljob_result.pkl"
+RESULT_PATH_DEFAULT_VALUE = f"{OUTPUT_STAGE_SUBPATH}/mljob_result"
 
 # Default container image information
 DEFAULT_IMAGE_REPO = "/snowflake/images/snowflake_images"