Utils for streaming large files (S3, HDFS, gzip, bz2...)
smart_open — utils for streaming large files in Python
What?
smart_open is a Python 3 library for efficient streaming of very large files from/to storages such as S3, GCS, Azure Blob Storage, HDFS, WebHDFS, HTTP, HTTPS, SFTP, or local filesystem. It supports transparent, on-the-fly (de-)compression for a variety of different formats.
smart_open is a drop-in replacement for Python's built-in open(): it can do anything open can (100% compatible, falls back to native open wherever possible), plus lots of nifty extra stuff on top.
Why?
Working with large remote files, for example using Amazon's boto3 Python library, is a pain. boto3's Object.uploadfileobj() and Object.downloadfileobj() methods require gotcha-prone boilerplate to use successfully, such as constructing file-like object wrappers. smart_open shields you from that. It builds on boto3 and other remote storage libraries, but offers a clean unified Pythonic API. The result is less code for you to write and fewer bugs to make.
How?
smart_open is well-tested, well-documented, and has a simple Pythonic API:
>>> from smart_open import open
>>>
>>> # stream lines from an S3 object
>>> for line in open('s3://commoncrawl/robots.txt'):
... print(repr(line))
... break
'User-Agent: *\n'
>>> # stream from/to compressed files, with transparent (de)compression: >>> for line in open('tests/test_data/1984.txt.gz', encoding='utf-8'): ... print(repr(line)) 'It was a bright cold day in April, and the clocks were striking thirteen.\n' 'Winston Smith, his chin nuzzled into his breast in an effort to escape the vile\n' 'wind, slipped quickly through the glass doors of Victory Mansions, though not\n' 'quickly enough to prevent a swirl of gritty dust from entering along with him.\n'
>>> # can use context managers too: >>> with open('tests/test_data/1984.txt.gz') as fin: ... with open('tests/test_data/1984.txt.bz2', 'w') as fout: ... for line in fin: ... fout.write(line) 74 80 78 79
>>> # can use any IOBase operations, like seek >>> with open('s3://commoncrawl/robots.txt', 'rb') as fin: ... for line in fin: ... print(repr(line.decode('utf-8'))) ... break ... offset = fin.seek(0) # seek to the beginning ... print(fin.read(4)) 'User-Agent: *\n' b'User'
>>> # stream from HTTP >>> for line in open('http://example.com'): ... print(repr(line[:15])) ... break '<!doctype html>'
For more examples of URIs that smartopen accepts, see help.txt or help('smartopen'). Some examples:
s3://bucket/key
s3://accesskeyid:secretaccesskey@bucket/key
gcs://bucket/blob
azure://bucket/blob
hdfs://host:port/path/file
./local/path/file.gz
file:///home/user/file.bz2
[ssh|scp|sftp]://username:password@host/path/fileDocumentation
The API reference can be viewed at help.txt or using help('smart_open').
Installation
smart_open supports a wide range of storage solutions. For all options, see the API reference. Each individual solution has its own dependencies. By default, smart_open does not install any dependencies in order to keep the installation size small. You can install one or more of these dependencies explicitly using optional dependencies defined in pyproject.toml:
pip install 'smart_open[s3,gcs,azure,http,webhdfs,ssh,zst,lz4]'Or, if you don't mind installing a large number of third party libraries, you can install all dependencies using:
pip install 'smart_open[all]'Built-in help
To view the API reference, use the help python builtin:
help("smart_open")or view help.txt in your browser.
More examples
For the sake of simplicity, the examples below assume you have all the dependencies installed, i.e. you have done:
pip install 'smart_open[all]'import os, boto3, botocore
from smart_open import open
stream content into S3 (write mode) using a custom client
this client is thread-safe ref https://github.com/boto/boto3/blob/1.38.41/docs/source/guide/clients.rst?plain=1#L111
config = botocore.client.Config(
maxpoolconnections=64,
tcp_keepalive=True,
retries={"max_attempts": 6, "mode": "adaptive"},
)
client = boto3.Session(
awsaccesskeyid=os.environ["AWSACCESSKEYID"],
awssecretaccesskey=os.environ["AWSSECRETACCESSKEY"],
).client("s3", config=config)
with open(
"s3://smart-open-py37-benchmark-results/test.txt", "wb", transport_params={"client": client}
) as fout:
bytes_written = fout.write(b"hello world!")
print(bytes_written)
perform a single-part upload to S3 (saves billable API requests, and allows seek() before upload)
with open(
"s3://smart-open-py37-benchmark-results/test.txt", "wb", transportparams={"multipartupload": False}
) as fout:
bytes_written = fout.write(b"hello world!")
print(bytes_written)
now with tempfile.TemporaryFile instead of the default io.BytesIO (to reduce memory footprint)
import tempfile
with ( tempfile.TemporaryFile() as tmp, open( "s3://smart-open-py37-benchmark-results/test.txt", "wb", transportparams={"multipartupload": False, "writebuffer": tmp}, ) as fout, ): bytes_written = fout.write(b"hello world!") print(bytes_written)
stream from HDFS
for line in open("hdfs://host:port/user/hadoop/my_file.txt", encoding="utf8"):
print(line)
stream from WebHDFS
for line in open("webhdfs://host:port/user/hadoop/my_file.txt"):
print(line)
stream content into HDFS (write mode):
with open("hdfs://host:port/user/hadoop/my_file.txt", "wb") as fout:
fout.write(b"hello world")
stream content into WebHDFS (write mode):
with open("webhdfs://host:port/user/hadoop/my_file.txt", "wb") as fout:
fout.write(b"hello world")
stream from a completely custom s3 server, like s3proxy:
client = boto3.client(
"s3", endpointurl="http://host:port", awsaccesskeyid="user", awssecretaccess_key="secret"
)
for line in open("s3://mybucket/mykey.txt", transport_params={"client": client}):
print(line)
Stream to Digital Ocean Spaces bucket providing credentials from boto3 profile
session = boto3.Session(profile_name="digitalocean")
client = session.client("s3", endpoint_url="https://ams3.digitaloceanspaces.com")
transport_params = {"client": client}
with open("s3://bucket/key.txt", "wb", transportparams=transportparams) as fout:
fout.write(b"here we stand")
stream from GCS
for line in open("gcs://mybucket/myfile.txt"):
print(line)
stream content into GCS (write mode):
with open("gcs://mybucket/myfile.txt", "wb") as fout:
fout.write(b"hello world")
stream from Azure Blob Storage
connectstr = os.environ["AZURESTORAGECONNECTIONSTRING"]
transport_params = {
"client": azure.storage.blob.BlobServiceClient.fromconnectionstring(connect_str),
}
for line in open("azure://mycontainer/myfile.txt", transportparams=transportparams):
print(line)
stream content into Azure Blob Storage (write mode):
connectstr = os.environ["AZURESTORAGECONNECTIONSTRING"]
transport_params = {
"client": azure.storage.blob.BlobServiceClient.fromconnectionstring(connect_str),
}
with open("azure://mycontainer/myfile.txt", "wb", transportparams=transport_params) as fout:
fout.write(b"hello world")Compression Handling
The top-level compression parameter controls compression/decompression behavior when reading and writing. The supported values for this parameter are:
inferfromextension(default behavior)disable.bz2.gz.lz4.xz.zst
smart_open automatically (de)compresses the file if the filename ends with one of these extensions.
See also
smartopen.compression.getsupportedcompressiontypes and martopen.compression.registercompressor.
>>> from smart_open import open
>>> with open('tests/test_data/1984.txt.gz') as fin:
... print(fin.read(32))
It was a bright cold day in ApriYou can override this behavior to either disable compression, or explicitly specify the algorithm to use. To disable compression:
>>> from smart_open import open
>>> with open('tests/test_data/1984.txt.gz', 'rb', compression='disable') as fin:
... print(fin.read(32))
b'\x1f\x8b\x08\x08\x85F\x94\\\x00\x031984.txt\x005\x8f=r\xc3@\x08\x85{\x9d\xe2\x1d@'To specify the algorithm explicitly (e.g. for non-standard file extensions):
>>> from smart_open import open
>>> with open('tests/test_data/1984.txt.gzip', compression='.gz') as fin:
... print(fin.read(32))
It was a bright cold day in ApriTo forward per-call options to the compression library (e.g. lower gzip's default compresslevel of 9 for faster writes), pass compression_kwargs:
>>> import tempfile
>>> from smart_open import open
>>> with tempfile.NamedTemporaryFile(suffix='.gz') as tmp:
... with open(tmp.name, 'wb', compression_kwargs={'compresslevel': 6}) as fout:
... _ = fout.write(b'hello world')The dict is forwarded as-is; spell each option using the underlying library's own kwarg name (compresslevel for gzip/bz2, preset for xz, level for zstd, compression_level for lz4).
You can also easily add support for other file extensions and compression formats. For example, to open xz-compressed files:
>>> import lzma, os
>>> from smartopen import open, registercompressor
>>> def handlexz(file_obj, mode, **kwargs): ... return lzma.open(filename=file_obj, mode=mode, **kwargs)
>>> registercompressor('.xz', handle_xz)
>>> with open('tests/test_data/1984.txt.xz') as fin: ... print(fin.read(32)) It was a bright cold day in Apri
This is just an example: lzma is in the standard library and is registered by default.
Transport-specific Options
smart_open supports a wide range of transport options out of the box. For the full list of supported URI schemes, see help.txt or help('smart_open'). Some examples:
- AWS S3 (and any S3-Compatible)
- HTTP, HTTPS (read-only)
- SSH, SCP and SFTP
- HDFS / WebHDFS
- Google Cloud Storage
- Azure Blob Storage
smartopen's open function accepts a keyword argument transportparams which accepts additional parameters for the transport layer.
Here are some examples of using this parameter:
>>> import boto3
>>> fin = open('s3://commoncrawl/robots.txt', transport_params=dict(client=boto3.client('s3')))
>>> fin = open('s3://commoncrawl/robots.txt', transportparams=dict(buffersize=1024))For the full list of keyword arguments supported by each transport option, see help.txt or help('smart_open').
help("smart_open.open")S3 Credentials
smart_open uses the boto3 library to talk to S3. boto3 has several mechanisms for determining the credentials to use. By default, smart_open will defer to boto3 and let the latter take care of the credentials. There are several ways to override this behavior.
The first is to pass a boto3.Client object as a transport parameter to the open function. You can customize the credentials when constructing the session for the client. smart_open will then use the session when talking to S3.
session = boto3.Session(
awsaccesskeyid=ACCESSKEY,
awssecretaccesskey=SECRETKEY,
awssessiontoken=SESSION_TOKEN,
)
client = session.client("s3", endpoint_url=..., config=...)
fin = open("s3://bucket/key", transport_params={"client": client})Your second option is to specify the credentials within the S3 URL itself:
fin = open("s3://awsaccesskeyid:awssecretaccesskey@bucket/key", ...)Important: The two methods above are mutually exclusive. If you pass an AWS client and the URL contains credentials, smart_open will ignore the latter.
Important: smart_open ignores configuration files from the older boto library. Port your old boto settings to boto3 in order to use them with smart_open.
S3 Advanced Usage
Additional keyword arguments can be propagated to the boto3 methods that are used by smartopen under the hood using the clientkwargs transport parameter.
For instance, to upload a blob with Metadata, ACL, StorageClass, these keyword arguments can be passed to createmultipartupload (docs).
kwargs = {"Metadata": {"version": 2}, "ACL": "authenticated-read", "StorageClass": "STANDARD_IA"}
fout = open(
"s3://bucket/key", "wb", transportparams={"clientkwargs": {"S3.Client.createmultipartupload": kwargs}}
)Iterating Over an S3 Bucket's Contents
Since going over all (or select) keys in an S3 bucket is a very common operation, there's also an extra function smartopen.s3.iterbucket() that does this efficiently, processing the bucket keys in parallel (using multithreading):
>>> from smart_open import s3
>>> # we use workers=1 for reproducibility; you should use as many workers as you have cores
>>> bucket = 'silo-open-data'
>>> prefix = 'Official/annual/monthly_rain/'
>>> for key, content in s3.iterbucket(bucket, prefix=prefix, acceptkey=lambda key: '/201' in key, workers=1, key_limit=3):
... print(key, round(len(content) / 2**20))
Official/annual/monthlyrain/2010.monthlyrain.nc 13
Official/annual/monthlyrain/2011.monthlyrain.nc 13
Official/annual/monthlyrain/2012.monthlyrain.nc 13GCS Credentials
smart_open uses the google-cloud-storage library to talk to GCS. google-cloud-storage uses the google-cloud package under the hood to handle authentication. There are several options to provide credentials. By default, smart_open will defer to google-cloud-storage and let it take care of the credentials.
To override this behavior, pass a google.cloud.storage.Client object as a transport parameter to the open function. You can customize the credentials when constructing the client. smart_open will then use the client when talking to GCS. To follow allow with the example below, refer to Google's guide to setting up GCS authentication with a service account.
import os
from google.cloud.storage import Client
serviceaccountpath = os.environ["GOOGLEAPPLICATIONCREDENTIALS"] client = Client.fromserviceaccountjson(serviceaccount_path) fin = open("gcs://gcp-public-data-landsat/index.csv.gz", transport_params=dict(client=client))
If you need more credential options, you can create an explicit google.auth.credentials.Credentials object and pass it to the Client. To create an API token for use in the example below, refer to the GCS authentication guide.
import os
from google.auth.credentials import Credentials
from google.cloud.storage import Client
token = os.environ["GOOGLEAPITOKEN"] credentials = Credentials(token=token) client = Client(credentials=credentials) fin = open("gcs://gcp-public-data-landsat/index.csv.gz", transport_params={"client": client})
GCS Advanced Usage
Additional keyword arguments can be propagated to the GCS open method (docs), which is used by smartopen under the hood, using the blobopen_kwargs transport parameter.
Additionally keyword arguments can be propagated to the GCS getblob method (docs) when in a read-mode, using the getblobkwargs transport parameter.
Additional blob properties (docs) can be set before an upload, as long as they are not read-only, using the blobproperties transport parameter.
openkwargs = {"predefinedacl": "authenticated-read"}
properties = {"metadata": {"version": 2}, "storage_class": "COLDLINE"}
fout = open(
"gcs://bucket/key",
"wb",
transportparams={"blobopenkwargs": openkwargs, "blob_properties": properties},
)Azure Credentials
smart_open uses the azure-storage-blob library to talk to Azure Blob Storage. By default, smart_open will defer to azure-storage-blob and let it take care of the credentials.
Azure Blob Storage does not have any ways of inferring credentials therefore, passing a azure.storage.blob.BlobServiceClient object as a transport parameter to the open function is required. You can customize the credentials when constructing the client. smart_open will then use the client when talking to. To follow allow with the example below, refer to Azure's guide to setting up authentication.
import os
from azure.storage.blob import BlobServiceClient
azurestorageconnectionstring = os.environ["AZURESTORAGECONNECTIONSTRING"] client = BlobServiceClient.fromconnectionstring(azurestorageconnection_string) fin = open("azure://mycontainer/myblob.txt", transport_params={"client": client})
If you need more credential options, refer to the Azure Storage authentication guide.
Azure Advanced Usage
Additional keyword arguments can be propagated to the commitblocklist method (docs), which is used by smartopen under the hood for uploads, using the blob_kwargs transport parameter.
kwargs = {"metadata": {"version": 2}}
fout = open("azure://container/key", "wb", transportparams={"blobkwargs": kwargs})For Azure, smartopen also supports append mode (mode="ab") by creating new (or appending to existing) Append Blobs (max 195 GiB).
Drop-in replacement of pathlib.Path.open
smart_open.open can also be used with Path objects. The built-in Path.open() is not able to read text from compressed files, so use patchpathlib to replace it with smartopen.open() instead. This can be helpful when e.g. working with compressed files.
>>> from pathlib import Path
>>> from smartopen.smartopenlib import patchpathlib
>>>
>>> = patchpathlib() # replace Path.open with smart_open.open
>>>
>>> path = Path("tests/test_data/crime-and-punishment.txt.gz")
>>>
>>> with path.open("r") as infile:
... print(infile.readline()[:41])
В начале июля, в чрезвычайно жаркое времяHow do I ...?
See HOWTO.md.
Extending smart_open
See EXTENDING.md.
Testing smart_open
smart_open comes with a comprehensive suite of unit tests. Before you can run the test suite, install the test dependencies:
pip install -e .[dev]Now, you can run the unit tests:
pytest testsThe tests are also run automatically with GitHub Actions on every commit push & pull request.
Comments, bug reports
smartopen lives on Github. You can file issues or pull requests there. Suggestions, pull requests and improvements welcome!
smartopen is open source software released under the MIT license. Copyright (c) 2015-now Radim Řehůřek.
