b10-transfer 0.1.4__py3-none-any.whl → 0.1.6__py3-none-any.whl

b10_transfer/__init__.py

@@ -1,23 +1,24 @@
 """B10 Transfer - Lock-free PyTorch file transfer for Baseten."""
 
-from .core import load_compile_cache, save_compile_cache, clear_local_cache
+from .cache import load_compile_cache, save_compile_cache, clear_local_cache
+from .core import transfer
 from .utils import CacheError, CacheValidationError
 from .space_monitor import CacheOperationInterrupted
 from .info import get_cache_info, list_available_caches
-from .constants import SaveStatus, LoadStatus
+from .constants import OperationStatus
 
 # Version
-__version__ = "0.1.4"
+__version__ = "0.1.6"
 
 __all__ = [
     "CacheError",
     "CacheValidationError",
     "CacheOperationInterrupted",
-    "SaveStatus",
-    "LoadStatus",
+    "OperationStatus",
     "load_compile_cache",
     "save_compile_cache",
     "clear_local_cache",
+    "transfer",
     "get_cache_info",
     "list_available_caches",
 ]

b10_transfer/cache.py

@@ -0,0 +1,361 @@
+"""Cache operations for PyTorch compilation artifacts.
+
+This module provides functions for loading and saving PyTorch compilation cache
+to/from b10fs shared storage using atomic operations and space monitoring.
+"""
+
+import logging
+import tempfile
+from pathlib import Path
+
+from .environment import get_cache_filename
+from .cleanup import cooperative_cleanup_b10fs
+from .utils import (
+    timed_fn,
+    critical_section_b10fs_file_lock,
+    safe_execute,
+    temp_file_cleanup,
+    cache_operation,
+    safe_unlink,
+)
+from .space_monitor import (
+    CacheSpaceMonitor,
+    CacheOperationInterrupted,
+    run_monitored_process,
+    worker_process,
+)
+from .constants import (
+    TORCH_CACHE_DIR,
+    B10FS_CACHE_DIR,
+    LOCAL_WORK_DIR,
+    MAX_CACHE_SIZE_MB,
+    REQUIRED_B10FS_SPACE_MB,
+    MIN_LOCAL_SPACE_MB,
+    CACHE_FILE_EXTENSION,
+    CACHE_LATEST_SUFFIX,
+    CACHE_INCOMPLETE_SUFFIX,
+    OperationStatus,
+)
+from .core import transfer
+
+logger = logging.getLogger(__name__)
+
+
+"""
+FIXME(SRAY):
+What about the case in @b10-transfer/ where a single pod finishes an inference request,
+and then the client calls save_compile_cache. And while we are creating the local archive,
+another inference call on the same pod is kicked off, which then modifies the torch cache.
+How would this be handled? Maybe just accept that the cache will be recompiled/overwritten?
+Otherwise you'd need application level coordination to ensure that the cache is not modified
+while we are creating the archive, but this doesn't really seem like a good idea in terms of adoption.
+
+FIXME(SR):
+More things to consider:
+- [possible] What if b10fs dies *during* an op? right now we check for b10fs availability in the beginning of the op... Add some constants instead of just False for load().
+- [possible, and really bad if it happens] potential memory exhaustion during compression if the cache is super super large. very very edge case. higher compression levels also have high memory usage.
+"""
+
+
+def _setup_cache_paths():
+    """Common setup for cache operations - returns paths and performs cleanup."""
+    # Cooperative cleanup of stale shared resources
+    cooperative_cleanup_b10fs()
+
+    b10fs_dir = Path(B10FS_CACHE_DIR)
+    torch_dir = Path(TORCH_CACHE_DIR)
+    work_dir = Path(LOCAL_WORK_DIR)
+
+    return b10fs_dir, torch_dir, work_dir
+
+
+def _get_cache_file_paths(cache_filename: str, b10fs_dir: Path):
+    """Generate cache file paths for a given cache filename."""
+    final_file = (
+        b10fs_dir / f"{cache_filename}{CACHE_LATEST_SUFFIX}{CACHE_FILE_EXTENSION}"
+    )
+    temp_file = (
+        b10fs_dir / f"{cache_filename}{CACHE_INCOMPLETE_SUFFIX}{CACHE_FILE_EXTENSION}"
+    )
+    return final_file, temp_file
+
+
+def _run_with_space_monitoring(
+    space_threshold_mb: float,
+    monitor_dir: Path,
+    operation_name: str,
+    worker_func,
+    worker_args: tuple,
+    cleanup_func=None,
+):
+    """Helper to run an operation with space monitoring."""
+    space_monitor = CacheSpaceMonitor(space_threshold_mb, monitor_dir)
+    space_monitor.start()
+
+    try:
+        logger.info(
+            f"Starting {operation_name}: {' -> '.join(str(arg) for arg in worker_args[:2])}"
+        )
+        run_monitored_process(
+            worker_func,
+            worker_args,
+            space_monitor,
+            operation_name,
+            cleanup_func=cleanup_func,
+        )
+    finally:
+        space_monitor.stop()
+
+
+def _transfer_with_b10fs_lock(
+    source: str, dest: str, lock_type: str, cleanup_on_failure=True
+):
+    """Transfer a file with b10fs file locking and error handling."""
+
+    @critical_section_b10fs_file_lock(lock_type)
+    def _locked_transfer():
+        result = transfer(source, dest)
+        if result != OperationStatus.SUCCESS:
+            if cleanup_on_failure:
+                safe_unlink(
+                    Path(dest), f"Failed to cleanup after failed transfer {dest}"
+                )
+            raise Exception(f"Failed to transfer {source} -> {dest}")
+
+    _locked_transfer()
+
+
+@timed_fn(logger=logger, name="Loading compile cache")
+@safe_execute("Load failed", False)
+def load_compile_cache() -> OperationStatus:
+    """Load PyTorch compilation cache from b10fs to local torch cache directory.
+
+    This function implements a lock-free pattern to safely load cached PyTorch
+    compilation artifacts from the b10fs shared filesystem to the local torch
+    cache directory. It validates b10fs availability, checks for existing cache,
+    and extracts the archive if needed.
+
+    The function monitors local disk space during both the copy from b10fs and
+    extraction phases, interrupting operations if space falls below MIN_LOCAL_SPACE_MB.
+
+    Returns:
+        OperationStatus:
+              OperationStatus.SUCCESS if cache was successfully loaded
+              OperationStatus.SKIPPED if already exists
+              OperationStatus.ERROR if b10fs is unavailable, local disk space is insufficient, or loading failed.
+              OperationStatus.DOES_NOT_EXIST if no cache file was found.
+
+    Raises:
+        CacheValidationError: If b10fs is not enabled (caught and returns OperationStatus.ERROR).
+        CacheOperationInterrupted: If operations interrupted due to insufficient
+                                  local disk space (caught and returns OperationStatus.ERROR).
+        Exception: Any other errors during loading (caught and returns OperationStatus.ERROR).
+    """
+    with cache_operation("Load"):
+        b10fs_dir, torch_dir, work_dir = _setup_cache_paths()
+
+        cache_filename = get_cache_filename()
+        final_file, _ = _get_cache_file_paths(cache_filename, b10fs_dir)
+        logger.debug(f"Looking for cache file: {final_file}")
+
+        if not final_file.exists():
+            logger.info("No cache file found in b10fs")
+            return OperationStatus.DOES_NOT_EXIST
+
+        # Skip if already loaded
+        if torch_dir.exists() and any(torch_dir.iterdir()):
+            logger.info("Torch cache already loaded, skipping extraction")
+            return OperationStatus.SKIPPED
+
+        # Create temp local copy
+        with tempfile.NamedTemporaryFile(
+            suffix=CACHE_FILE_EXTENSION, dir=work_dir, delete=False
+        ) as f:
+            temp_path = Path(f.name)
+        logger.debug(f"Created temporary file for cache: {temp_path}")
+
+        try:
+            with temp_file_cleanup(temp_path):
+                # Phase 1: Copy from b10fs to local temp file
+                _transfer_with_b10fs_lock(
+                    str(final_file),
+                    str(temp_path),
+                    "copy_out",
+                    cleanup_on_failure=False,
+                )
+
+                # Phase 2: Extract archive with space monitoring
+                _run_with_space_monitoring(
+                    MIN_LOCAL_SPACE_MB,
+                    work_dir,
+                    "archive extraction",
+                    _cache_extract_worker,
+                    (str(temp_path), str(torch_dir)),
+                    cleanup_func=lambda: _cleanup_torch_dir(torch_dir),
+                )
+
+            logger.info("Cache load complete")
+            return OperationStatus.SUCCESS
+
+        except CacheOperationInterrupted as e:
+            logger.warning(f"Cache load interrupted: {e}")
+            return OperationStatus.ERROR
+
+
+@timed_fn(logger=logger, name="Saving compile cache")
+@safe_execute("Save failed", False)
+def save_compile_cache() -> OperationStatus:
+    """Save local PyTorch compilation cache to b10fs using atomic journal pattern.
+
+    This function creates an archive of the local torch cache directory and
+    atomically saves it to b10fs using a journal pattern (write to temp file,
+    then rename). This ensures concurrent saves don't corrupt each other.
+
+    The function validates b10fs availability, checks if cache already exists
+    (early exit), performs initial space checks using pre-calculated requirements
+    for concurrent saves, starts background space monitoring, then runs compression
+    and copy operations in separate worker processes that can be terminated if disk
+    space becomes insufficient, finally performing an atomic rename to the final cache file.
+
+    Returns:
+        OperationStatus:
+              OperationStatus.SUCCESS if cache was successfully saved or already exists
+              OperationStatus.ERROR if b10fs is unavailable, insufficient disk space caused interruption,
+                no cache exists to save, or saving failed.
+              OperationStatus.SKIPPED if no cache exists to save or cache already exists in b10fs
+
+    Raises:
+        CacheValidationError: If b10fs is not enabled (caught and returns OperationStatus.ERROR).
+        CacheOperationInterrupted: If operations interrupted due to insufficient
+                                  disk space (caught and returns OperationStatus.ERROR).
+        ArchiveError: If archive creation fails (caught and returns OperationStatus.ERROR).
+        Exception: Any other errors during saving (caught and returns OperationStatus.ERROR).
+    """
+    with cache_operation("Save"):
+        b10fs_dir, torch_dir, work_dir = _setup_cache_paths()
+
+        # Check if anything to save
+        if not torch_dir.exists() or not any(torch_dir.iterdir()):
+            logger.info("No torch cache to save")
+            return OperationStatus.SKIPPED
+
+        cache_filename = get_cache_filename()
+        final_file, temp_file = _get_cache_file_paths(cache_filename, b10fs_dir)
+
+        # Check for existing cache first (early exit)
+        if final_file.exists():
+            logger.info("Cache already exists in b10fs, skipping save")
+            return OperationStatus.SKIPPED
+
+        with tempfile.NamedTemporaryFile(
+            suffix=CACHE_FILE_EXTENSION, dir=work_dir, delete=False
+        ) as f:
+            local_temp = Path(f.name)
+        logger.debug(f"Created local temp file for archive: {local_temp}")
+
+        try:
+            with temp_file_cleanup(local_temp):
+                # Phase 1: Compression with space monitoring
+                _run_with_space_monitoring(
+                    REQUIRED_B10FS_SPACE_MB,
+                    b10fs_dir,
+                    "compression",
+                    _cache_compression_worker,
+                    (str(torch_dir), str(local_temp), MAX_CACHE_SIZE_MB),
+                )
+
+                # Phase 2: Copy to b10fs with locking
+                _transfer_with_b10fs_lock(
+                    str(local_temp), str(temp_file), "copy_in", cleanup_on_failure=True
+                )
+
+                # Phase 3: Atomic rename (fast, don't interrupt)
+                logger.info(
+                    f"Renaming temp file to final cache file: {temp_file} -> {final_file}"
+                )
+                temp_file.rename(final_file)
+
+            logger.info("Cache save complete")
+            return OperationStatus.SUCCESS
+
+        except CacheOperationInterrupted as e:
+            logger.warning(f"Cache save interrupted: {e}")
+            return OperationStatus.ERROR
+
+
+@safe_execute("Clear failed", False)
+def clear_local_cache() -> bool:
+    """Clear the local PyTorch compilation cache directory.
+
+    This function removes the entire local torch cache directory and all its
+    contents. This is useful for cleaning up disk space or forcing recompilation.
+
+    Returns:
+        bool: True if cache was successfully cleared or didn't exist, False if
+              clearing failed due to permissions or other filesystem errors.
+
+    Raises:
+        Exception: Any errors during directory removal (caught and returns False).
+    """
+    torch_dir = Path(TORCH_CACHE_DIR)
+    if not torch_dir.exists():
+        return True
+    import shutil
+
+    shutil.rmtree(torch_dir)
+    return True
+
+
+@worker_process("Compression was cancelled before starting")
+def _cache_compression_worker(
+    torch_dir_str: str, local_temp_str: str, max_size_mb: int
+) -> None:
+    """Worker process that handles cache compression.
+
+    This function runs in a separate process to compress the torch cache directory
+    into an archive. It can be terminated externally if disk space becomes insufficient.
+
+    Args:
+        torch_dir_str: String path to the torch cache directory to compress.
+        local_temp_str: String path where the compressed archive will be created.
+        max_size_mb: Maximum allowed archive size in megabytes.
+    """
+    torch_dir = Path(torch_dir_str)
+    local_temp = Path(local_temp_str)
+
+    # Import here to avoid issues with multiprocessing
+    from .archive import create_archive
+
+    create_archive(torch_dir, local_temp, max_size_mb)
+
+
+def _cleanup_torch_dir(torch_dir: Path) -> None:
+    """Helper function to safely cleanup torch directory during interrupted extraction."""
+    try:
+        if torch_dir.exists():
+            import shutil
+
+            shutil.rmtree(torch_dir)
+            logger.debug(f"Cleaned up torch directory: {torch_dir}")
+    except Exception as e:
+        logger.error(f"Failed to cleanup torch directory {torch_dir}: {e}")
+
+
+@worker_process("Extraction was cancelled before starting")
+def _cache_extract_worker(archive_path_str: str, dest_dir_str: str) -> None:
+    """Worker process that handles archive extraction.
+
+    This function runs in a separate process to extract the cache archive to
+    the torch cache directory. It can be terminated externally if local disk space becomes insufficient.
+
+    Args:
+        archive_path_str: String path to the archive file to extract.
+        dest_dir_str: String path to the directory where archive will be extracted.
+    """
+    archive_path = Path(archive_path_str)
+    dest_dir = Path(dest_dir_str)
+
+    # Import here to avoid issues with multiprocessing
+    from .archive import extract_archive
+
+    extract_archive(archive_path, dest_dir)

b10_transfer/constants.py

@@ -114,25 +114,10 @@ class WorkerStatus(Enum):
     CANCELLED = auto()
 
 
-class LoadStatus(Enum):
-    """Status values for cache loading operations."""
-
-    SUCCESS = auto()
-    ERROR = auto()
-    DOES_NOT_EXIST = auto()
-    SKIPPED = auto()
-
-
-class SaveStatus(Enum):
-    """Status values for cache saving operations."""
-
-    SUCCESS = auto()
-    ERROR = auto()
-    SKIPPED = auto()
-
-
-class TransferStatus(Enum):
-    """Status values for file transfer operations."""
+class OperationStatus(Enum):
+    """Status values for all b10-transfer operations (load, save, transfer)."""
 
     SUCCESS = auto()
     ERROR = auto()
+    DOES_NOT_EXIST = auto()  # Used by load operations when cache file not found
+    SKIPPED = auto()  # Used by load/save operations when operation not needed

b10_transfer/core.py

@@ -1,19 +1,16 @@
-import os
+"""Core file transfer operations for b10-transfer.
+
+This module provides generic file transfer functionality with space monitoring
+and error handling for b10fs operations.
+"""
+
 import logging
-import tempfile
 import shutil
 from pathlib import Path
 
-import time
-
-from .environment import get_cache_filename
-from .cleanup import cooperative_cleanup_b10fs
 from .utils import (
     timed_fn,
-    critical_section_b10fs_file_lock,
     safe_execute,
-    temp_file_cleanup,
-    cache_operation,
     safe_unlink,
 )
 from .space_monitor import (
@@ -24,241 +21,18 @@ from .space_monitor import (
     worker_process,
 )
 from .constants import (
-    TORCH_CACHE_DIR,
     B10FS_CACHE_DIR,
-    LOCAL_WORK_DIR,
-    MAX_CACHE_SIZE_MB,
     REQUIRED_B10FS_SPACE_MB,
     MIN_LOCAL_SPACE_MB,
-    CACHE_FILE_EXTENSION,
-    CACHE_LATEST_SUFFIX,
-    CACHE_INCOMPLETE_SUFFIX,
-    LoadStatus,
-    SaveStatus,
-    TransferStatus,
+    OperationStatus,
 )
 
 logger = logging.getLogger(__name__)
 
 
-@timed_fn(logger=logger, name="Loading compile cache")
-@safe_execute("Load failed", False)
-def load_compile_cache() -> LoadStatus:
-    """Load PyTorch compilation cache from b10fs to local torch cache directory.
-
-    This function implements a lock-free pattern to safely load cached PyTorch
-    compilation artifacts from the b10fs shared filesystem to the local torch
-    cache directory. It validates b10fs availability, checks for existing cache,
-    and extracts the archive if needed.
-
-    The function monitors local disk space during both the copy from b10fs and
-    extraction phases, interrupting operations if space falls below MIN_LOCAL_SPACE_MB.
-
-    Returns:
-        LoadStatus:
-              LoadStatus.SUCCESS if cache was successfully loaded
-              LoadStatus.SKIPPED if already exists
-              LoadStatus.ERROR if b10fs is unavailable, local disk space is insufficient, or loading failed.
-              LoadStatus.DOES_NOT_EXIST if no cache file was found.
-
-    Raises:
-        CacheValidationError: If b10fs is not enabled (caught and returns LoadStatus.ERROR).
-        CacheOperationInterrupted: If operations interrupted due to insufficient
-                                  local disk space (caught and returns LoadStatus.ERROR).
-        Exception: Any other errors during loading (caught and returns LoadStatus.ERROR).
-    """
-    with cache_operation("Load"):
-        # Cooperative cleanup of stale shared resources
-        cooperative_cleanup_b10fs()
-
-        b10fs_dir = Path(B10FS_CACHE_DIR)
-        torch_dir = Path(TORCH_CACHE_DIR)
-        work_dir = Path(LOCAL_WORK_DIR)
-
-        cache_filename = get_cache_filename()
-        cache_file = (
-            b10fs_dir / f"{cache_filename}{CACHE_LATEST_SUFFIX}{CACHE_FILE_EXTENSION}"
-        )
-        logger.debug(f"Looking for cache file: {cache_file}")
-
-        if not cache_file.exists():
-            logger.info("No cache file found in b10fs")
-            return LoadStatus.DOES_NOT_EXIST
-
-        # Skip if already loaded
-        if torch_dir.exists() and any(torch_dir.iterdir()):
-            logger.info("Torch cache already loaded, skipping extraction")
-            return LoadStatus.SKIPPED
-
-        # Create temp local copy
-        with tempfile.NamedTemporaryFile(
-            suffix=CACHE_FILE_EXTENSION, dir=work_dir, delete=False
-        ) as f:
-            temp_path = Path(f.name)
-        logger.debug(f"Created temporary file for cache: {temp_path}")
-
-        try:
-            with temp_file_cleanup(temp_path):
-                # Phase 1: Copy from b10fs to local temp file using transfer()
-                @critical_section_b10fs_file_lock("copy_out")
-                def _monitored_copy_from_b10fs():
-                    result = transfer(str(cache_file), str(temp_path))
-                    if result != TransferStatus.SUCCESS:
-                        raise Exception("Failed to copy cache file from b10fs")
-
-                _monitored_copy_from_b10fs()
-
-                # Phase 2: Extract archive in separate process with space monitoring
-                space_monitor = CacheSpaceMonitor(MIN_LOCAL_SPACE_MB, work_dir)
-                space_monitor.start()
-
-                try:
-                    logger.info(f"Starting extraction: {temp_path} -> {torch_dir}")
-                    run_monitored_process(
-                        _cache_extract_worker,
-                        (str(temp_path), str(torch_dir)),
-                        space_monitor,
-                        "archive extraction",
-                        cleanup_func=lambda: _cleanup_torch_dir(torch_dir),
-                    )
-                finally:
-                    space_monitor.stop()
-
-            logger.info("Cache load complete")
-            return LoadStatus.SUCCESS
-
-        except CacheOperationInterrupted as e:
-            logger.warning(f"Cache load interrupted: {e}")
-            return LoadStatus.ERROR
-
-
-"""
-FIXME(SRAY):
-What about the case in @b10-transfer/ where a single pod finishes an inference request,
-and then the client calls save_compile_cache. And while we are creating the local archive,
-another inference call on the same pod is kicked off, which then modifies the torch cache.
-How would this be handled? Maybe just accept that the cache will be recompiled/overwritten?
-Otherwise you'd need application level coordination to ensure that the cache is not modified
-while we are creating the archive, but this doesn't really seem like a good idea in terms of adoption.
-
-FIXME(SR):
-More things to consider:
-- [possible] What if b10fs dies *during* an op? right now we check for b10fs availability in the beginning of the op... Add some constants instead of just False for load().
-- [possible, and really bad if it happens] potential memory exhaustion during compression if the cache is super super large. very very edge case. higher compression levels also have high memory usage.
-"""
-
-
-@timed_fn(logger=logger, name="Saving compile cache")
-@safe_execute("Save failed", False)
-def save_compile_cache() -> SaveStatus:
-    """Save local PyTorch compilation cache to b10fs using atomic journal pattern.
-
-    This function creates an archive of the local torch cache directory and
-    atomically saves it to b10fs using a journal pattern (write to temp file,
-    then rename). This ensures concurrent saves don't corrupt each other.
-
-    The function validates b10fs availability, checks if cache already exists
-    (early exit), performs initial space checks using pre-calculated requirements
-    for concurrent saves, starts background space monitoring, then runs compression
-    and copy operations in separate worker processes that can be terminated if disk
-    space becomes insufficient, finally performing an atomic rename to the final cache file.
-
-    Returns:
-        SaveStatus:
-              SaveStatus.SUCCESS if cache was successfully saved or already exists
-              SaveStatus.ERROR if b10fs is unavailable, insufficient disk space caused interruption,
-                no cache exists to save, or saving failed.
-              SaveStatus.SKIPPED if no cache exists to save or cache already exists in b10fs
-
-    Raises:
-        CacheValidationError: If b10fs is not enabled (caught and returns SaveStatus.ERROR).
-        CacheOperationInterrupted: If operations interrupted due to insufficient
-                                  disk space (caught and returns SaveStatus.ERROR).
-        ArchiveError: If archive creation fails (caught and returns SaveStatus.ERROR).
-        Exception: Any other errors during saving (caught and returns SaveStatus.ERROR).
-    """
-    with cache_operation("Save"):
-        # Cooperative cleanup of stale shared resources
-        cooperative_cleanup_b10fs()
-
-        b10fs_dir = Path(B10FS_CACHE_DIR)
-        torch_dir = Path(TORCH_CACHE_DIR)
-        work_dir = Path(LOCAL_WORK_DIR)
-
-        # Check if anything to save
-        if not torch_dir.exists() or not any(torch_dir.iterdir()):
-            logger.info("No torch cache to save")
-            return SaveStatus.SKIPPED
-
-        cache_filename = get_cache_filename()
-        final_file = (
-            b10fs_dir / f"{cache_filename}{CACHE_LATEST_SUFFIX}{CACHE_FILE_EXTENSION}"
-        )
-
-        # Check for existing cache first (early exit)
-        if final_file.exists():
-            logger.info("Cache already exists in b10fs, skipping save")
-            return SaveStatus.SKIPPED
-
-        temp_file = (
-            b10fs_dir
-            / f"{cache_filename}{CACHE_INCOMPLETE_SUFFIX}{CACHE_FILE_EXTENSION}"
-        )
-
-        with tempfile.NamedTemporaryFile(
-            suffix=CACHE_FILE_EXTENSION, dir=work_dir, delete=False
-        ) as f:
-            local_temp = Path(f.name)
-        logger.debug(f"Created local temp file for archive: {local_temp}")
-
-        try:
-            with temp_file_cleanup(local_temp):
-                # Phase 1: Compression in separate process with space monitoring
-                space_monitor = CacheSpaceMonitor(REQUIRED_B10FS_SPACE_MB, b10fs_dir)
-                space_monitor.start()
-
-                try:
-                    logger.info(f"Starting compression: {torch_dir} -> {local_temp}")
-                    run_monitored_process(
-                        _cache_compression_worker,
-                        (str(torch_dir), str(local_temp), MAX_CACHE_SIZE_MB),
-                        space_monitor,
-                        "compression",
-                    )
-                finally:
-                    space_monitor.stop()
-
-                # Phase 2: Copy to b10fs using transfer()
-                @critical_section_b10fs_file_lock("copy_in")
-                def _monitored_copy_to_b10fs():
-                    result = transfer(str(local_temp), str(temp_file))
-                    if result != TransferStatus.SUCCESS:
-                        # Clean up the temp file if transfer failed
-                        safe_unlink(
-                            temp_file,
-                            f"Failed to cleanup after failed copy {temp_file}",
-                        )
-                        raise Exception("Failed to copy cache file to b10fs")
-
-                _monitored_copy_to_b10fs()
-
-                # Phase 3: Atomic rename (fast, don't interrupt)
-                logger.info(
-                    f"Renaming temp file to final cache file: {temp_file} -> {final_file}"
-                )
-                temp_file.rename(final_file)
-
-            logger.info("Cache save complete")
-            return SaveStatus.SUCCESS
-
-        except CacheOperationInterrupted as e:
-            logger.warning(f"Cache save interrupted: {e}")
-            return SaveStatus.ERROR
-
-
 @timed_fn(logger=logger, name="Transferring file")
-@safe_execute("Transfer failed", TransferStatus.ERROR)
-def transfer(source: str, dest: str) -> TransferStatus:
+@safe_execute("Transfer failed", OperationStatus.ERROR)
+def transfer(source: str, dest: str) -> OperationStatus:
     """Transfer a file from source to destination with space monitoring.
 
     This function copies a file from source to destination using the same
@@ -270,15 +44,15 @@ def transfer(source: str, dest: str) -> TransferStatus:
         dest: Path to the destination where the file will be copied.
 
     Returns:
-        TransferStatus:
-              TransferStatus.SUCCESS if transfer was successful
-              TransferStatus.ERROR if transfer failed due to insufficient disk space,
+        OperationStatus:
+              OperationStatus.SUCCESS if transfer was successful
+              OperationStatus.ERROR if transfer failed due to insufficient disk space,
                 file not found, or other errors.
 
     Raises:
         CacheOperationInterrupted: If transfer interrupted due to insufficient
-                                  disk space (caught and returns TransferStatus.ERROR).
-        Exception: Any other errors during transfer (caught and returns TransferStatus.ERROR).
+                                  disk space (caught and returns OperationStatus.ERROR).
+        Exception: Any other errors during transfer (caught and returns OperationStatus.ERROR).
     """
     source_path = Path(source)
     dest_path = Path(dest)
@@ -286,7 +60,7 @@ def transfer(source: str, dest: str) -> TransferStatus:
     # Validate source file exists
     if not source_path.exists():
         logger.error(f"Source file does not exist: {source}")
-        return TransferStatus.ERROR
+        return OperationStatus.ERROR
 
     # Create destination directory if it doesn't exist
     dest_path.parent.mkdir(parents=True, exist_ok=True)
@@ -330,66 +104,22 @@ def transfer(source: str, dest: str) -> TransferStatus:
         )
 
         logger.info("File transfer complete")
-        return TransferStatus.SUCCESS
+        return OperationStatus.SUCCESS
 
     except CacheOperationInterrupted as e:
         logger.warning(f"File transfer interrupted: {e}")
-        return TransferStatus.ERROR
+        return OperationStatus.ERROR
 
     finally:
         space_monitor.stop()
 
 
-@safe_execute("Clear failed", False)
-def clear_local_cache() -> bool:
-    """Clear the local PyTorch compilation cache directory.
-
-    This function removes the entire local torch cache directory and all its
-    contents. This is useful for cleaning up disk space or forcing recompilation.
-
-    Returns:
-        bool: True if cache was successfully cleared or didn't exist, False if
-              clearing failed due to permissions or other filesystem errors.
-
-    Raises:
-        Exception: Any errors during directory removal (caught and returns False).
-    """
-    torch_dir = Path(TORCH_CACHE_DIR)
-    if not torch_dir.exists():
-        return True
-    shutil.rmtree(torch_dir)
-    return True
-
-
-@worker_process("Compression was cancelled before starting")
-def _cache_compression_worker(
-    torch_dir_str: str, local_temp_str: str, max_size_mb: int
-) -> None:
-    """Worker process that handles cache compression.
-
-    This function runs in a separate process to compress the torch cache directory
-    into an archive. It can be terminated externally if disk space becomes insufficient.
-
-    Args:
-        torch_dir_str: String path to the torch cache directory to compress.
-        local_temp_str: String path where the compressed archive will be created.
-        max_size_mb: Maximum allowed archive size in megabytes.
-    """
-    torch_dir = Path(torch_dir_str)
-    local_temp = Path(local_temp_str)
-
-    # Import here to avoid issues with multiprocessing
-    from .archive import create_archive
-
-    create_archive(torch_dir, local_temp, max_size_mb)
-
-
 @worker_process("Copy was cancelled before starting")
 def _cache_copy_worker(source_path_str: str, dest_path_str: str) -> None:
-    """Worker process that handles file copy to b10fs.
+    """Worker process that handles file copy operations.
 
-    This function runs in a separate process to copy the compressed cache file
-    to the b10fs filesystem. It can be terminated externally if disk space becomes insufficient.
+    This function runs in a separate process to copy files between locations.
+    It can be terminated externally if disk space becomes insufficient.
 
     Args:
         source_path_str: String path to the source file to copy.
@@ -399,33 +129,3 @@ def _cache_copy_worker(source_path_str: str, dest_path_str: str) -> None:
     dest_path = Path(dest_path_str)
 
     shutil.copy2(source_path, dest_path)
-
-
-def _cleanup_torch_dir(torch_dir: Path) -> None:
-    """Helper function to safely cleanup torch directory during interrupted extraction."""
-    try:
-        if torch_dir.exists():
-            shutil.rmtree(torch_dir)
-            logger.debug(f"Cleaned up torch directory: {torch_dir}")
-    except Exception as e:
-        logger.error(f"Failed to cleanup torch directory {torch_dir}: {e}")
-
-
-@worker_process("Extraction was cancelled before starting")
-def _cache_extract_worker(archive_path_str: str, dest_dir_str: str) -> None:
-    """Worker process that handles archive extraction.
-
-    This function runs in a separate process to extract the cache archive to
-    the torch cache directory. It can be terminated externally if local disk space becomes insufficient.
-
-    Args:
-        archive_path_str: String path to the archive file to extract.
-        dest_dir_str: String path to the directory where archive will be extracted.
-    """
-    archive_path = Path(archive_path_str)
-    dest_dir = Path(dest_dir_str)
-
-    # Import here to avoid issues with multiprocessing
-    from .archive import extract_archive
-
-    extract_archive(archive_path, dest_dir)

{b10_transfer-0.1.4.dist-info → b10_transfer-0.1.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: b10-transfer
-Version: 0.1.4
+Version: 0.1.6
 Summary: Distributed PyTorch file transfer for Baseten - Environment-aware, lock-free file transfer management
 License: MIT
 Keywords: pytorch,file-transfer,cache,machine-learning,inference

b10_transfer-0.1.6.dist-info/RECORD

@@ -0,0 +1,13 @@
+b10_transfer/__init__.py,sha256=1oxaP7np1iu1GbODcGmujd4K4T1bBZjOiVq5e1GW9JM,665
+b10_transfer/archive.py,sha256=GKb0mi0-YeM7ch4FLAoOLHXw0T6LkRerYad2N2y9TYM,6400
+b10_transfer/cache.py,sha256=B5fNCJkMIpUBwZuKMoQVbn0NeEuIrcAtYMk0gXkkOAM,13768
+b10_transfer/cleanup.py,sha256=3RnqWNGMCcko5GQdq1Gr9VPpGzAF5J6x7xjIH9SNZ78,6226
+b10_transfer/constants.py,sha256=iuLShDW6hInhyz2YTQ8CzBanqW4chCkQOAzPZkCtOoA,4322
+b10_transfer/core.py,sha256=vsOcH0ve2GP-YBgHU58WgCEbx0h7dXn2R5sJErnQt8k,4437
+b10_transfer/environment.py,sha256=aC0biEMQrtHk0ke_3epdcq1X9J5fPmPpBVt0fH7XF2Y,5625
+b10_transfer/info.py,sha256=I3iOuImZ5r6DMJTDeBtVvzlSn6IuyPJbLJYUO_OF0ks,6299
+b10_transfer/space_monitor.py,sha256=C_CKDH43bNsWdq60WStSZ3c_nQkWvScQmqU_SYHesew,10531
+b10_transfer/utils.py,sha256=Stee0DFK-8MRRYNIocqaK64cJvfs4jPW3Mpx7zkWV6Y,11932
+b10_transfer-0.1.6.dist-info/METADATA,sha256=vwu77uY1CnrqTEdLYChlk91M6odOmstA_4a8AzeaH5M,4108
+b10_transfer-0.1.6.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+b10_transfer-0.1.6.dist-info/RECORD,,

b10_transfer-0.1.4.dist-info/RECORD

@@ -1,12 +0,0 @@
-b10_transfer/__init__.py,sha256=LKMroIusY1itfMVrJT07xLS1XVehwr54Wk5dhEl8MzY,641
-b10_transfer/archive.py,sha256=GKb0mi0-YeM7ch4FLAoOLHXw0T6LkRerYad2N2y9TYM,6400
-b10_transfer/cleanup.py,sha256=3RnqWNGMCcko5GQdq1Gr9VPpGzAF5J6x7xjIH9SNZ78,6226
-b10_transfer/constants.py,sha256=qCViKTyfHTLpiFVF2SwsbHp2IMz3kg3syxJfgRAq2dc,4446
-b10_transfer/core.py,sha256=XWLuwjHXuhh-6abZMAl2yuLB7R2deyUc6gGPn6-Yfkc,17006
-b10_transfer/environment.py,sha256=aC0biEMQrtHk0ke_3epdcq1X9J5fPmPpBVt0fH7XF2Y,5625
-b10_transfer/info.py,sha256=I3iOuImZ5r6DMJTDeBtVvzlSn6IuyPJbLJYUO_OF0ks,6299
-b10_transfer/space_monitor.py,sha256=C_CKDH43bNsWdq60WStSZ3c_nQkWvScQmqU_SYHesew,10531
-b10_transfer/utils.py,sha256=Stee0DFK-8MRRYNIocqaK64cJvfs4jPW3Mpx7zkWV6Y,11932
-b10_transfer-0.1.4.dist-info/METADATA,sha256=69s3ACBUFzGB7J97eVt4aCGSXrIpld1oV0Wj8Z0HLZ8,4108
-b10_transfer-0.1.4.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-b10_transfer-0.1.4.dist-info/RECORD,,