GCSFileSystem(*args, **kwargs)

Connect to Google Cloud Storage.

GCSFileSystem.cat(path[, recursive, on_error])

Fetch (potentially multiple) paths' contents

GCSFileSystem.du(path[, total, maxdepth, ...])

Space used by files and optionally directories within a path

GCSFileSystem.exists(path, **kwargs)

Is there a file at the given path

GCSFileSystem.get(rpath, lpath[, recursive, ...])

Copy file(s) to local.

GCSFileSystem.glob(path[, maxdepth])

Find files by glob-matching.

GCSFileSystem.info(path, **kwargs)

Give details of entry at path

GCSFileSystem.ls(path[, detail])

List objects at path.

GCSFileSystem.mkdir(path[, acl, ...])

New bucket

GCSFileSystem.mv(path1, path2[, recursive, ...])

Move file(s) from one location to another

GCSFileSystem.open(path[, mode, block_size, ...])

Return a file-like object from the filesystem

GCSFileSystem.put(lpath, rpath[, recursive, ...])

Copy file(s) from local.

GCSFileSystem.read_block(fn, offset, length)

Read a block of bytes from

GCSFileSystem.rm(path[, recursive, ...])

Delete files.

GCSFileSystem.tail(path[, size])

Get the last size bytes from file

GCSFileSystem.touch(path[, truncate])

Create empty file, or update timestamp

GCSFileSystem.get_mapper([root, check, ...])

Create key/value store based on this file-system

GCSFile(gcsfs, path[, mode, block_size, ...])


Close file


Write buffered data to backend store.


File information about this path


Return data from cache, or fetch pieces as necessary

GCSFile.seek(loc[, whence])

Set current file location


Current file location


Write data to buffer.

class gcsfs.core.GCSFileSystem(*args, **kwargs)[source]

Connect to Google Cloud Storage.

The following modes of authentication are supported:

  • token=None, GCSFS will attempt to guess your credentials in the following order: gcloud CLI default, gcsfs cached token, google compute metadata service, anonymous.

  • token='google_default', your default gcloud credentials will be used, which are typically established by doing gcloud login in a terminal.

  • token='cache', credentials from previously successful gcsfs authentication will be used (use this after “browser” auth succeeded)

  • token='anon', no authentication is performed, and you can only access data which is accessible to allUsers (in this case, the project and access level parameters are meaningless)

  • token='browser', you get an access code with which you can authenticate via a specially provided URL

  • if token='cloud', we assume we are running within google compute or google container engine, and query the internal metadata directly for a token.

  • you may supply a token generated by the [gcloud](https://cloud.google.com/sdk/docs/) utility; this is either a python dictionary, the name of a file containing the JSON returned by logging in with the gcloud CLI tool, or a Credentials object. gcloud typically stores its tokens in locations such as ~/.config/gcloud/application_default_credentials.json, ~/.config/gcloud/credentials, or ~\AppData\Roaming\gcloud\credentials, etc.

Specific methods, (eg. ls, info, …) may return object details from GCS. These detailed listings include the [object resource](https://cloud.google.com/storage/docs/json_api/v1/objects#resource)

GCS does not include “directory” objects but instead generates directories by splitting [object names](https://cloud.google.com/storage/docs/key-terms). This means that, for example, a directory does not need to exist for an object to be created within it. Creating an object implicitly creates it’s parent directories, and removing all objects from a directory implicitly deletes the empty directory.

GCSFileSystem generates listing entries for these implied directories in listing apis with the object properties:

  • “name”string

    The “{bucket}/{name}” path of the dir, used in calls to GCSFileSystem or GCSFile.

  • “bucket”string

    The name of the bucket containing this object.

  • “kind” : ‘storage#object’

  • “size” : 0

  • “storageClass” : ‘DIRECTORY’

  • type: ‘directory’ (fsspec compat)

GCSFileSystem maintains a per-implied-directory cache of object listings and fulfills all object information and listing requests from cache. This implied, for example, that objects created via other processes will not be visible to the GCSFileSystem until the cache refreshed. Calls to GCSFileSystem.open and calls to GCSFile are not effected by this cache.

In the default case the cache is never expired. This may be controlled via the cache_timeout GCSFileSystem parameter or via explicit calls to GCSFileSystem.invalidate_cache.

  • project (string) – project_id to work under. Note that this is not the same as, but often very similar to, the project name. This is required in order to list all the buckets you have access to within a project and to create/delete buckets, or update their access policies. If token='google_default', the value is overridden by the default, if token='anon', the value is ignored.

  • access (one of {'read_only', 'read_write', 'full_control'}) – Full control implies read/write as well as modifying metadata, e.g., access control.

  • token (None, dict or string) – (see description of authentication methods, above)

  • consistency ('none', 'size', 'md5') – Check method when writing files. Can be overridden in open().

  • cache_timeout (float, seconds) – Cache expiration time in seconds for object metadata cache. Set cache_timeout <= 0 for no caching, None for no cache expiration.

  • secure_serialize (bool (deprecated)) –

  • requester_pays (bool, or str default False) – Whether to use requester-pays requests. This will include your project ID project in requests as the userProject, and you’ll be billed for accessing data from requester-pays buckets. Optionally, pass a project-id here as a string to use that as the userProject.

  • session_kwargs (dict) – passed on to aiohttp.ClientSession; can contain, for example, proxy settings.

  • endpoint_url (str) – If given, use this URL (format protocol://host:port , without any path part) for communication. If not given, defaults to the value of environment variable “STORAGE_EMULATOR_HOST”; if that is not set either, will use the standard Google endpoint.

  • default_location (str) – Default location where buckets are created, like ‘US’ or ‘EUROPE-WEST3’. You can find a list of all available locations here: https://cloud.google.com/storage/docs/locations#available-locations

  • version_aware (bool) – Whether to support object versioning. If enabled this will require the user to have the necessary permissions for dealing with versioned objects.

property buckets

Return list of available project buckets.

cat(path, recursive=False, on_error='raise', **kwargs)

Fetch (potentially multiple) paths’ contents

  • recursive (bool) – If True, assume the path(s) are directories, and get all the contained files

  • on_error ("raise", "omit", "return") – If raise, an underlying exception will be raised (converted to KeyError if the type is in self.missing_exceptions); if omit, keys with exception will simply not be included in the output; if “return”, all keys are included in the output, but the value will be bytes or an exception instance.

  • kwargs (passed to cat_file) –


  • dict of {path (contents} if there are multiple paths)

  • or the path has been otherwise expanded

cat_file(path, start=None, end=None, **kwargs)

Get the content of a file

  • path (URL of file on this filesystems) –

  • start (int) – Bytes limits of the read. If negative, backwards from end, like usual python slices. Either can be None for start or end of file, respectively

  • end (int) – Bytes limits of the read. If negative, backwards from end, like usual python slices. Either can be None for start or end of file, respectively

  • kwargs (passed to open().) –

cat_ranges(paths, starts, ends, max_gap=None, on_error='return', **kwargs)

Get the contents of byte ranges from one or more files

  • paths (list) – A list of of filepaths on this filesystems

  • starts (int or list) – Bytes limits of the read. If using a single int, the same value will be used to read all the specified files.

  • ends (int or list) – Bytes limits of the read. If using a single int, the same value will be used to read all the specified files.


Unique value for current version of file

If the checksum is the same from one moment to another, the contents are guaranteed to be the same. If the checksum changes, the contents might have changed.

This should normally be overridden; default will probably capture creation/modification timestamp (which would be good) or maybe access timestamp (which would be bad)

classmethod clear_instance_cache()

Clear the cache of filesystem instances.


Unless overridden by setting the cachable class attribute to False, the filesystem class stores a reference to newly created instances. This prevents Python’s normal rules around garbage collection from working, since the instances refcount will not drop to zero until clear_instance_cache is called.

copy(path1, path2, recursive=False, maxdepth=None, on_error=None, **kwargs)

Copy within two locations in the filesystem

on_error“raise”, “ignore”

If raise, any not-found exceptions will be raised; if ignore any not-found exceptions will cause the path to be skipped; defaults to raise unless recursive is true, where the default is ignore

cp(path1, path2, **kwargs)

Alias of AbstractFileSystem.copy.


Return the created timestamp of a file as a datetime.datetime

classmethod current()

Return the most recently instantiated FileSystem

If no instance has been created, then create one with defaults

delete(path, recursive=False, maxdepth=None)

Alias of AbstractFileSystem.rm.

disk_usage(path, total=True, maxdepth=None, **kwargs)

Alias of AbstractFileSystem.du.

download(rpath, lpath, recursive=False, **kwargs)

Alias of AbstractFileSystem.get.

du(path, total=True, maxdepth=None, withdirs=False, **kwargs)

Space used by files and optionally directories within a path

Directory size does not include the size of its contents.

  • path (str) –

  • total (bool) – Whether to sum all the file sizes

  • maxdepth (int or None) – Maximum number of directory levels to descend, None for unlimited.

  • withdirs (bool) – Whether to include directory paths in the output.

  • kwargs (passed to find) –


  • Dict of {path (size} if total=False, or int otherwise, where numbers)

  • refer to bytes used.


Finish write transaction, non-context version

exists(path, **kwargs)

Is there a file at the given path

expand_path(path, recursive=False, maxdepth=None, **kwargs)

Turn one or more globs or directories into a list of all matching paths to files or directories.

kwargs are passed to glob or find, which may in turn call ls

find(path, maxdepth=None, withdirs=False, detail=False, **kwargs)

List all files below path.

Like posix find command without conditions

  • path (str) –

  • maxdepth (int or None) – If not None, the maximum number of levels to descend

  • withdirs (bool) – Whether to include directory paths in the output. This is True when used by glob, but users usually only want files.

  • ls. (kwargs are passed to) –

static from_json(blob)

Recreate a filesystem instance from JSON representation

See .to_json() for the expected structure of the input


blob (str) –

Return type

file system instance, not necessarily of this particular class.

property fsid

Persistent filesystem id that can be used to compare filesystems across sessions.

get(rpath, lpath, recursive=False, callback=<fsspec.callbacks.NoOpCallback object>, maxdepth=None, **kwargs)

Copy file(s) to local.

Copies a specific file or tree of files (if recursive=True). If lpath ends with a “/”, it will be assumed to be a directory, and target files will go within. Can submit a list of paths, which may be glob-patterns and will be expanded.

Calls get_file for each source.

get_file(rpath, lpath, callback=<fsspec.callbacks.NoOpCallback object>, outfile=None, **kwargs)

Copy single remote file to local

get_mapper(root='', check=False, create=False, missing_exceptions=None)

Create key/value store based on this file-system

Makes a MutableMapping interface to the FS at the given root path. See fsspec.mapping.FSMap for further details.

getxattr(path, attr)

Get user-defined metadata attribute

glob(path, maxdepth=None, **kwargs)

Find files by glob-matching.

If the path ends with ‘/’, only folders are returned.

We support "**", "?" and "[..]". We do not support ^ for pattern negation.

The maxdepth option is applied on the first ** found in the path.

kwargs are passed to ls.

head(path, size=1024)

Get the first size bytes from file

info(path, **kwargs)

Give details of entry at path

Returns a single dictionary, with exactly the same information as ls would with detail=True.

The default implementation should calls ls and could be overridden by a shortcut. kwargs are passed on to `ls().

Some file systems might not be able to measure the file’s size, in which case, the returned dict will include 'size': None.


  • dict with keys (name (full path in the FS), size (in bytes), type (file,)

  • directory, or something else) and other FS-specific keys.


Invalidate listing cache for given path, it is reloaded on next use.


path (string or None) – If None, clear all listings cached else listings at or under given path.


Is this entry directory-like?


Is this entry file-like?

lexists(path, **kwargs)

If there is a file at the given path (including broken links)

listdir(path, detail=True, **kwargs)

Alias of AbstractFileSystem.ls.

ls(path, detail=True, **kwargs)

List objects at path.

This should include subdirectories and files at that location. The difference between a file and a directory must be clear when details are requested.

The specific keys, or perhaps a FileInfo class, or similar, is TBD, but must be consistent across implementations. Must include:

  • full path to the entry (without protocol)

  • size of the entry, in bytes. If the value cannot be determined, will be None.

  • type of entry, “file”, “directory” or other

Additional information may be present, appropriate to the file-system, e.g., generation, checksum, etc.

May use refresh=True|False to allow use of self._ls_from_cache to check for a saved listing and avoid calling the backend. This would be common where listing may be expensive.

  • path (str) –

  • detail (bool) – if True, gives a list of dictionaries, where each is the same as the result of info(path). If False, gives a list of paths (str).

  • kwargs (may have additional backend-specific options, such as version) – information


  • List of strings if detail is False, or list of directory information

  • dicts if detail is True.

makedir(path, create_parents=True, **kwargs)

Alias of AbstractFileSystem.mkdir.

makedirs(path, exist_ok=False)

Recursively make directories

Creates directory at path and any intervening required directories. Raises exception if, for instance, the path already exists but is a file.

  • path (str) – leaf directory name

  • exist_ok (bool (False)) – If False, will error if the target already exists

merge(path, paths, acl=None)

Concatenate objects within a single bucket

mkdir(path, acl='projectPrivate', default_acl='bucketOwnerFullControl', location=None, create_parents=True, enable_versioning=False, **kwargs)

New bucket

If path is more than just a bucket, will create bucket if create_parents=True; otherwise is a noop. If create_parents is False and bucket does not exist, will produce FileNotFFoundError.

  • path (str) – bucket name. If contains ‘/’ (i.e., looks like subdir), will have no effect because GCS doesn’t have real directories.

  • acl (string, one of bACLs) – access for the bucket itself

  • default_acl (str, one of ACLs) – default ACL for objects created in this bucket

  • location (Optional[str]) – Location where buckets are created, like ‘US’ or ‘EUROPE-WEST3’. If not provided, defaults to self.default_location. You can find a list of all available locations here: https://cloud.google.com/storage/docs/locations#available-locations

  • create_parents (bool) – If True, creates the bucket in question, if it doesn’t already exist

  • enable_versioning (bool) – If True, creates the bucket in question with object versioning enabled.

mkdirs(path, exist_ok=False)

Alias of AbstractFileSystem.makedirs.


Return the modified timestamp of a file as a datetime.datetime

move(path1, path2, **kwargs)

Alias of AbstractFileSystem.mv.

mv(path1, path2, recursive=False, maxdepth=None, **kwargs)

Move file(s) from one location to another

open(path, mode='rb', block_size=None, cache_options=None, compression=None, **kwargs)

Return a file-like object from the filesystem

The resultant instance must function correctly in a context with block.

  • path (str) – Target file

  • mode (str like 'rb', 'w') – See builtin open()

  • block_size (int) – Some indication of buffering - this is a value in bytes

  • cache_options (dict, optional) – Extra arguments to pass through to the cache.

  • compression (string or None) – If given, open file using compression codec. Can either be a compression name (a key in fsspec.compression.compr) or “infer” to guess the compression from the filename suffix.

  • encoding (passed on to TextIOWrapper for text mode) –

  • errors (passed on to TextIOWrapper for text mode) –

  • newline (passed on to TextIOWrapper for text mode) –

pipe(path, value=None, **kwargs)

Put value into path

(counterpart to cat)

  • path (string or dict(str, bytes)) – If a string, a single remote location to put value bytes; if a dict, a mapping of {path: bytesvalue}.

  • value (bytes, optional) – If using a single path, these are the bytes to put there. Ignored if path is a dict

pipe_file(path, value, **kwargs)

Set the bytes of given file

put(lpath, rpath, recursive=False, callback=<fsspec.callbacks.NoOpCallback object>, maxdepth=None, **kwargs)

Copy file(s) from local.

Copies a specific file or tree of files (if recursive=True). If rpath ends with a “/”, it will be assumed to be a directory, and target files will go within.

Calls put_file for each source.

put_file(lpath, rpath, callback=<fsspec.callbacks.NoOpCallback object>, **kwargs)

Copy single file to remote

read_block(fn, offset, length, delimiter=None)

Read a block of bytes from

Starting at offset of the file, read length bytes. If delimiter is set then we ensure that the read starts and stops at delimiter boundaries that follow the locations offset and offset + length. If offset is zero then we start at zero. The bytestring returned WILL include the end delimiter string.

If offset+length is beyond the eof, reads to eof.

  • fn (string) – Path to filename

  • offset (int) – Byte offset to start read

  • length (int) – Number of bytes to read. If None, read to end.

  • delimiter (bytes (optional)) – Ensure reading starts and stops at delimiter bytestring


>>> fs.read_block('data/file.csv', 0, 13)  
b'Alice, 100\nBo'
>>> fs.read_block('data/file.csv', 0, 13, delimiter=b'\n')  
b'Alice, 100\nBob, 200\n'

Use length=None to read to the end of the file. >>> fs.read_block(‘data/file.csv’, 0, None, delimiter=b’n’) # doctest: +SKIP b’Alice, 100nBob, 200nCharlie, 300’

See also


read_bytes(path, start=None, end=None, **kwargs)

Alias of AbstractFileSystem.cat_file.

read_text(path, encoding=None, errors=None, newline=None, **kwargs)

Get the contents of the file as a string.

  • path (str) – URL of file on this filesystems

  • encoding (same as open.) –

  • errors (same as open.) –

  • newline (same as open.) –

rename(path1, path2, **kwargs)

Alias of AbstractFileSystem.mv.

rm(path, recursive=False, maxdepth=None, batchsize=20)

Delete files.

  • path (str or list of str) – File(s) to delete.

  • recursive (bool) – If file(s) are directories, recursively delete contents and then also remove the directory

  • maxdepth (int or None) – Depth to pass to walk for finding files to delete, if recursive. If None, there will be no limit and infinite recursion may be possible.


Delete a file


Delete an empty bucket


bucket (str) – bucket name. If contains ‘/’ (i.e., looks like subdir), will have no effect because GCS doesn’t have real directories.

setxattrs(path, content_type=None, content_encoding=None, fixed_key_metadata=None, **kwargs)

Set/delete/add writable metadata attributes

Note: uses PATCH method (update), leaving unedited keys alone. fake-gcs-server:latest does not seem to support this.

  • content_type (str) – If not None, set the content-type to this value

  • content_encoding (str) – This parameter is deprecated, you may use fixed_key_metadata instead. If not None, set the content-encoding. See https://cloud.google.com/storage/docs/transcoding

  • fixed_key_metadata (dict) –

    Google metadata, in key/value pairs, supported keys:
    • cache_control

    • content_disposition

    • content_encoding

    • content_language

    • custom_time

    More info: https://cloud.google.com/storage/docs/metadata#mutable

  • kw_args (key-value pairs like field="value" or field=None) – value must be string to add or modify, or None to delete

Return type

Entire metadata after update (even if only path is passed)

sign(path, expiration=100, **kwargs)[source]

Create a signed URL representing the given path.

  • path (str) – The path on the filesystem

  • expiration (int) – Number of seconds to enable the URL for


URL – The signed URL

Return type



Size in bytes of file


Size in bytes of each file in a list of paths


Begin write transaction for deferring files, non-context version

stat(path, **kwargs)

Alias of AbstractFileSystem.info.

tail(path, size=1024)

Get the last size bytes from file


JSON representation of this filesystem instance


str – protocol (text name of this class’s protocol, first one in case of multiple), args (positional args, usually empty), and all other kwargs as their own keys.

Return type

JSON structure with keys cls (the python location of this class),

touch(path, truncate=True, **kwargs)

Create empty file, or update timestamp

  • path (str) – file location

  • truncate (bool) – If True, always set file size to 0; if False, update timestamp and leave file unchanged, if backend allows this

property transaction

A context within which files are committed together upon exit

Requires the file class to implement .commit() and .discard() for the normal and exception cases.


alias of Transaction


Hash of file properties, to tell if it has changed

unstrip_protocol(name: str) str

Format FS-specific path to generic, including protocol

upload(lpath, rpath, recursive=False, **kwargs)

Alias of AbstractFileSystem.put.


Get HTTP URL of the given path

walk(path, maxdepth=None, topdown=True, on_error='omit', **kwargs)

Return all files belows path

List all files, recursing into subdirectories; output is iterator-style, like os.walk(). For a simple list of files, find() is available.

When topdown is True, the caller can modify the dirnames list in-place (perhaps using del or slice assignment), and walk() will only recurse into the subdirectories whose names remain in dirnames; this can be used to prune the search, impose a specific order of visiting, or even to inform walk() about directories the caller creates or renames before it resumes walk() again. Modifying dirnames when topdown is False has no effect. (see os.walk)

Note that the “files” outputted will include anything that is not a directory, such as links.

  • path (str) – Root to recurse into

  • maxdepth (int) – Maximum recursion depth. None means limitless, but not recommended on link-based file-systems.

  • topdown (bool (True)) – Whether to walk the directory tree from the top downwards or from the bottom upwards.

  • on_error ("omit", "raise", a collable) – if omit (default), path with exception will simply be empty; If raise, an underlying exception will be raised; if callable, it will be called with a single OSError instance as argument

  • kwargs (passed to ls) –

write_bytes(path, value, **kwargs)

Alias of AbstractFileSystem.pipe_file.

write_text(path, value, encoding=None, errors=None, newline=None, **kwargs)

Write the text to the given file.

An existing file will be overwritten.

  • path (str) – URL of file on this filesystems

  • value (str) – Text to write.

  • encoding (same as open.) –

  • errors (same as open.) –

  • newline (same as open.) –

class gcsfs.core.GCSFile(gcsfs, path, mode='rb', block_size=5242880, autocommit=True, cache_type='readahead', cache_options=None, acl=None, consistency='md5', metadata=None, content_type=None, timeout=None, fixed_key_metadata=None, generation=None, **kwargs)[source]

Close file

Finalizes writes, discards cache


If not auto-committing, finalize file


Cancel in-progress multi-upload

Should only happen during discarding this write-mode file


Returns underlying file descriptor if one exists.

OSError is raised if the IO object does not use a file descriptor.


Write buffered data to backend store.

Writes the current buffer, if it is larger than the block-size, or if the file is being closed.


force (bool) – When closing, write the last block even if it is smaller than blocks are allowed to be. Disallows further writing to this file.


File information about this path


Return whether this is an ‘interactive’ stream.

Return False if it can’t be determined.

read(length=- 1)

Return data from cache, or fetch pieces as necessary


length (int (-1)) – Number of bytes to read; if <0, all remaining bytes.


Whether opened for reading


mirrors builtin file’s readinto method



Read until first occurrence of newline character

Note that, because of character encoding, this is not necessarily a true line ending.


Return all data, split by the newline character

readuntil(char=b'\n', blocks=None)

Return data between current position and first occurrence of char

char is included in the output, except if the end of the tile is encountered first.

  • char (bytes) – Thing to find

  • blocks (None or int) – How much to read in each go. Defaults to file blocksize - which may mean a new read on every call.

seek(loc, whence=0)

Set current file location

  • loc (int) – byte location

  • whence ({0, 1, 2}) – from start of file, current location or end of file, resp.


Whether is seekable (only in read mode)


Current file location


Truncate file to size bytes.

File pointer is left unchanged. Size defaults to the current IO position as reported by tell(). Returns the new size.


HTTP link to this file’s data


Whether opened for writing


Write data to buffer.

Buffer only sent on flush() or if buffer is greater than or equal to blocksize.


data (bytes) – Set of bytes to be written.

writelines(lines, /)

Write a list of lines to stream.

Line separators are not added, so it is usual for each of the lines provided to have a line separator at the end.