blobfile

This is a standalone clone of TensorFlow's gfile, supporting both local paths and gs:// (Google Cloud Storage) paths.

The main function is BlobFile, a replacement for GFile. There are also a few additional functions, basename, dirname, and join, which mostly do the same thing as their os.path namesakes, only they also support gs:// paths.

Installation:

pip install blobfile

Usage:

import blobfile as bf

with bf.BlobFile("gs://my-bucket-name/cats", "wb") as w:
    w.write(b"meow!")

Here are the functions:

• BlobFile - like open() but works with gs:// paths too, data can be streamed to/from the remote file. It accepts the following arguments:
  • streaming:
    • The default for streaming is True when mode is in "r", "rb" and False when mode is in "w", "wb", "a", "ab".
    • streaming=True:
      • Reading is done without downloading the entire remote file.
      • Writing is done to the remote file directly, but only in chunks of a few MB in size. flush() will not cause an early write.
      • Appending is not implemented.
    • streaming=False:
      • Reading is done by downloading the remote file to a local file during the constructor.
      • Writing is done by uploading the file on close() or during destruction.
      • Appending is done by downloading the file during construction and uploading on close().
  • buffer_size: number of bytes to buffer, this can potentially make reading more efficient.
  • cache_dir: a directory in which to cache files for reading, only valid if streaming=False and mode is in "r", "rb". You are reponsible for cleaning up the cache directory.

Some are inspired by existing os.path and shutil functions:

• copy - copy a file from one path to another, will do a remote copy between two remote paths on the same blob storage service
• exists - returns True if the file or directory exists
• glob - return files matching a glob-style pattern as a generator. Globs can have surprising performance characteristics when used with blob storage. Character ranges are not supported in patterns.
• isdir - returns True if the path is a directory
• listdir - list contents of a directory as a generator
• makedirs - ensure that a directory and all parent directories exist
• remove - remove a file
• rmdir - remove an empty directory
• rmtree - remove a directory tree
• stat - get the size and modification time of a file
• walk - walk a directory tree with a generator that yields (dirpath, dirnames, filenames) tuples
• basename - get the final component of a path
• dirname - get the path except for the final component
• join - join 2 or more paths together, inserting directory separators between each component

There are a few bonus functions:

• get_url - returns a url for a path along with the expiration for that url (or None)
• md5 - get the md5 hash for a path, for GCS this is fast, but for other backends this may be slow
• set_log_callback - set a log callback function log(msg: string) to use instead of printing to stdout

Errors

• Error - base class for library-specific exceptions
• RequestFailure - a request has failed permanently, has message:str, request:Request, and response:urllib3.HTTPResponse attributes.
• The following generic exceptions are raised from some functions to make the behavior similar to the original versions: FileNotFoundError, FileExistsError, IsADirectoryError, NotADirectoryError, OSError, ValueError, io.UnsupportedOperation

Logging

In order to make diagnosing stalls easier, blobfile will log when retrying requests. blobfile will keep retrying transient errors until they succeed or a permanent error is encountered (which will raise an exception).

To route those log lines, use set_log_callback to set a callback function which will be called whenever a log line should be printed. The default callback prints to stdout.

While blobfile does not use the python logging module, it does use urllib3 which uses that module. So if you configure the python logging module, you may need to change the settings to adjust urllib3's logging. To only log errors from urllib3, you can do logging.getLogger("urllib3").setLevel(logging.ERROR).

Examples

Write and read a file:

import blobfile as bf

with bf.BlobFile("gs://my-bucket/file.name", "wb") as f:
    f.write(b"meow")

print("exists:", bf.exists("gs://my-bucket/file.name"))

print("contents:", bf.BlobFile("gs://my-bucket/file.name", "rb").read())

Parallel execution:

import blobfile as bf
import multiprocessing as mp
import tqdm

filenames = [f"{i}.ext" for i in range(1000)]

with mp.Pool() as pool:
    for filename, exists in tqdm.tqdm(zip(filenames, pool.imap(bf.exists, filenames)), total=len(filenames)):
        pass

Parallel download of a single file:

import blobfile as bf
import concurrent.futures
import time


def _download_chunk(path, start, size):
    with bf.BlobFile(path, "rb") as f:
        f.seek(start)
        return f.read(size)


def parallel_download(path, chunk_size=16 * 2**20):
    pieces = []
    stat = bf.stat(path)
    with concurrent.futures.ProcessPoolExecutor() as executor:
        start = 0
        futures = []
        while start < stat.size:
            future = executor.submit(_download_chunk, path, start, chunk_size)
            futures.append(future)
            start += chunk_size
        for future in futures:
            pieces.append(future.result())
    return b"".join(pieces)


def main():
    contents = parallel_download("<path to file>")


if __name__ == "__main__":
    main()

Parallel copytree:

import blobfile as bf
import concurrent.futures
import tqdm


def _perform_op(op_tuple):
    op, src, dst = op_tuple
    if op == "copy":
        bf.copy(src, dst, overwrite=True)
    elif op == "mkdir":
        bf.makedirs(dst)
    else:
        raise Exception(f"invalid op {op}")


def copytree(src, dst):
    """
    Copy a directory tree from one location to another
    """
    if not bf.isdir(src):
        raise NotADirectoryError(f"The directory name is invalid: '{src}'")
    assert not dst.startswith(src), "dst cannot be a subdir of src"
    if not src.endswith("/"):
        src += "/"
    bf.makedirs(dst)

    with tqdm.tqdm(desc="listing") as pbar:
        ops = []
        # walk with topdown=False should be faster for nested directory trees
        for src_root, dirnames, filenames in bf.walk(src, topdown=False):
            relpath = src_root[len(src):]
            dst_root = bf.join(dst, relpath)

            if len(filenames) == 0:
                # only make empty directories, other directories will be implicitly created by copy
                ops.append(("mkdir", src_root, dst_root))
                pbar.update(1)

            # on GCS we can have a directory name that has the same name as a file
            # if that's the case, skip it since that's too confusing
            skip_filenames = set(dirnames)
            for filename in filenames:
                if filename in skip_filenames:
                    continue
                src_path = bf.join(src_root, filename)
                dst_path = bf.join(dst_root, filename)
                ops.append(("copy", src_path, dst_path))
                pbar.update(1)

    with concurrent.futures.ProcessPoolExecutor() as executor:
        list(tqdm.tqdm(executor.map(_perform_op, ops), total=len(ops), desc="copying"))


def main():
    contents = copytree("<path to source>", "<path to destination>")


if __name__ == "__main__":
    main()

Changes

See CHANGES.md