WARCIO: WARC (and ARC) Streaming Library

Background

This library provides a fast, standalone way to read and write WARC Format commonly used in web archives. Python 3.7+ (minimally only needing six as an external dependency)

warcio supports reading and writing of WARC files compliant with both the WARC 1.0 and WARC 1.1 ISO standards.

Install with: pip install warcio (or pip install warcio[all] to get optional features)

This library is a spin-off of the WARC reading and writing component of the pywb high-fidelity replay library, a key component of Webrecorder

The library is designed for fast, low-level access to web archival content, oriented around a stream of WARC records rather than files.

Reading WARC Records

A key feature of the library is to be able to iterate over a stream of WARC records using the ArchiveIterator.

It includes the following features:

• Reading a WARC 1.0, WARC 1.1 or ARC stream
• On the fly ARC to WARC record conversion
• Decompressing and de-chunking HTTP payload content stored in WARC/ARC files.

For example, the following prints the the url for each WARC response record:

from warcio.archiveiterator import ArchiveIterator

with open('path/to/file', 'rb') as stream:
    for record in ArchiveIterator(stream):
        if record.rec_type == 'response':
            print(record.rec_headers.get_header('WARC-Target-URI'))

The stream object could be a file on disk or a remote network stream. The ArchiveIterator reads the WARC content in a single pass. The record is represented by an ArcWarcRecord object which contains the format (ARC or WARC), record type, the record headers, http headers (if any), and raw stream for reading the payload.

class ArcWarcRecord(object):
    def __init__(self, *args):
        (self.format, self.rec_type, self.rec_headers, self.raw_stream,
         self.http_headers, self.content_type, self.length) = args

Reading WARC Content

The raw_stream can be used to read the rest of the payload directly. A special ArcWarcRecord.content_stream() function provides a stream that automatically decompresses and de-chunks the HTTP payload, if it is compressed and/or transfer-encoding chunked.

ARC Files

The library provides support for reading (but not writing ARC) files. The ARC format is legacy but is important to support in a consistent matter. The ArchiveIterator can equally iterate over ARC and WARC files to emit ArcWarcRecord objects. The special arc2warc option converts ARC records to WARCs on the fly, allowing for them to be accessed using the same API.

(Special WARCIterator and ARCIterator subclasses of ArchiveIterator are also available to read only WARC or only ARC files).

WARC and ARC Streaming

For example, here is a snippet for reading an ARC and a WARC using the same API.

The example streams a WARC and ARC file over HTTP using requests, printing the warcinfo record (or ARC header) and any response records (or all ARC records) that contain HTML:

import requests
from warcio.archiveiterator import ArchiveIterator

def print_records(url):
    resp = requests.get(url, stream=True)

    for record in ArchiveIterator(resp.raw, arc2warc=True):
        if record.rec_type == 'warcinfo':
            print(record.raw_stream.read())

        elif record.rec_type == 'response':
            if record.http_headers.get_header('Content-Type') == 'text/html':
                print(record.rec_headers.get_header('WARC-Target-URI'))
                print(record.content_stream().read())
                print('')

# WARC
print_records('https://archive.org/download/ExampleArcAndWarcFiles/IAH-20080430204825-00000-blackbook.warc.gz')


# ARC with arc2warc
print_records('https://archive.org/download/ExampleArcAndWarcFiles/IAH-20080430204825-00000-blackbook.arc.gz')

Writing WARC Records

Starting with 1.6, warcio introduces a way to capture HTTP/S traffic directly to a WARC file, by monkey-patching Python's http.client library.

This approach works well with the popular requests library often used to fetch HTTP/S content. Note that requests must be imported after the capture_http module.

Quick Start to Writing a WARC

Fetching the url https://example.com/ while capturing the response and request into a gzip compressed WARC file named example.warc.gz can be done with the following four lines:

from warcio.capture_http import capture_http
import requests  # requests must be imported after capture_http

with capture_http('example.warc.gz'):
    requests.get('https://example.com/')

The WARC example.warc.gz will contain two records (the response is written first, then the request).

To write to a default in-memory buffer (BufferWARCWriter), don't specify a filename, using with capture_http() as writer:.

Additional requests in the capture_http context and will be appended to the WARC as expected.

The WARC-IP-Address header will also be added for each record if the IP address is available.

The following example (similar to a unit test from the test suite) demonstrates the resulting records created with capture_http:

with capture_http() as writer:
    requests.get('http://example.com/')
    requests.get('https://google.com/')

expected = [('http://example.com/', 'response', True),
            ('http://example.com/', 'request', True),
            ('https://google.com/', 'response', True),
            ('https://google.com/', 'request', True),
            ('https://www.google.com/', 'response', True),
            ('https://www.google.com/', 'request', True)
           ]

 actual = [
            (record.rec_headers['WARC-Target-URI'],
             record.rec_type,
             'WARC-IP-Address' in record.rec_headers)

            for record in ArchiveIterator(writer.get_stream())
          ]

 assert actual == expected

Customizing WARC Writing

The library provides a simple and extensible interface for writing standards-compliant WARC files.

The library comes with a basic WARCWriter class for writing to a single WARC file and BufferWARCWriter for writing to an in-memory buffer. The BaseWARCWriter can be extended to support more complex operations.

(There is no support for writing legacy ARC files)

For more flexibility, such as to use a custom WARCWriter class, the above example can be written as:

from warcio.capture_http import capture_http
from warcio import WARCWriter
import requests  # requests *must* be imported after capture_http

with open('example.warc.gz', 'wb') as fh:
    warc_writer = WARCWriter(fh)
    with capture_http(warc_writer):
        requests.get('https://example.com/')

WARC/1.1 Support

By default, warcio creates WARC 1.0 records for maximum compatibility with existing tools. To create WARC/1.1 records, simply specify the warc version as follows:

with capture_http('example.warc.gz', warc_version='1.1'):
    ...

WARCWriter(fh, warc_version='1.1)
...

When using WARC 1.1, the main difference is that the WARC-Date timestamp header will be written with microsecond precision, while WARC 1.0 only supports second precision.

WARC 1.0:

WARC/1.0
...
WARC-Date: 2018-12-26T10:11:12Z

WARC 1.1:

WARC/1.1
...
WARC-Date: 2018-12-26T10:11:12.456789Z

Filtering HTTP Capture

When capturing via HTTP, it is possible to provide a custom filter function, which can be used to determine if a particular request and response records should be written to the WARC file or skipped.

The filter function is called with the request and response record before they are written, and can be used to substitute a different record (for example, a revisit instead of a response), or to skip writing altogether by returning nothing, as shown below:

def filter_records(request, response, request_recorder):
    # return None, None to indicate records should be skipped
    if response.http_headers.get_statuscode() != '200':
        return None, None

    # the response record can be replaced with a revisit record
    elif check_for_dedup():
        response = create_revisit_record(...)

    return request, response

with capture_http('example.warc.gz', filter_records):
     requests.get('https://example.com/')

Please refer to test/test_capture_http.py for additional examples of capturing requests traffic to WARC.

Manual/Advanced WARC Writing

Before 1.6, this was the primary method for fetching a url and then writing to a WARC. This process is a bit more verbose, but provides for full control of WARC creation and avoid monkey-patching.

The following example loads http://example.com/, creates a WARC response record, and writes it, gzip compressed, to example.warc.gz The block and payload digests are computed automatically.

from warcio.warcwriter import WARCWriter
from warcio.statusandheaders import StatusAndHeaders

import requests

with open('example.warc.gz', 'wb') as output:
    writer = WARCWriter(output, gzip=True)

    resp = requests.get('http://example.com/',
                        headers={'Accept-Encoding': 'identity'},
                        stream=True)

    # get raw headers from urllib3
    headers_list = resp.raw.headers.items()

    http_headers = StatusAndHeaders('200 OK', headers_list, protocol='HTTP/1.0')

    record = writer.create_warc_record('http://example.com/', 'response',
                                        payload=resp.raw,
                                        http_headers=http_headers)

    writer.write_record(record)

The library also includes additional semantics for:

• Creating warcinfo and revisit records
• Writing response and request records together
• Writing custom WARC records
• Reading a full WARC record from a stream

Please refer to warcwriter.py and test/test_writer.py for additional examples.

WARCIO CLI: Indexing and Recompression

The library currently ships with a few simple command line tools.

Index

The warcio index cmd will print a simple index of the records in the warc file as newline delimited JSON lines (NDJSON).

WARC header fields to include in the index can be specified via the -f flag, and are included in the JSON block (in order, for convenience).

warcio index ./test/data/example-iana.org-chunked.warc -f warc-type,warc-target-uri,content-length
{"warc-type": "warcinfo", "content-length": "137"}
{"warc-type": "response", "warc-target-uri": "http://www.iana.org/", "content-length": "7566"}
{"warc-type": "request", "warc-target-uri": "http://www.iana.org/", "content-length": "76"}

HTTP header fields can be included by prefixing them with the prefix http:. The special field offset refers to the record offset within the warc file.

warcio index ./test/data/example-iana.org-chunked.warc -f offset,content-type,http:content-type,warc-target-uri
{"offset": "0", "content-type": "application/warc-fields"}
{"offset": "405", "content-type": "application/http;msgtype=response", "http:content-type": "text/html; charset=UTF-8", "warc-target-uri": "http://www.iana.org/"}
{"offset": "8379", "content-type": "application/http;msgtype=request", "warc-target-uri": "http://www.iana.org/"}

(Note: this library does not produce CDX or CDXJ format indexes often associated with web archives. To create these indexes, please see the cdxj-indexer tool which extends warcio indexing to provide this functionality)

Check

The warcio check command will check the payload and block digests of WARC records, if possible. An exit value of 1 indicates a failure. warcio check -v will print verbose output for each record in the WARC file.

Recompress

The recompress command allows for re-compressing or normalizing WARC (or ARC) files to a record-compressed, gzipped WARC file.

Each WARC record is compressed individually and concatenated. This is the 'canonical' WARC storage format used by Webrecorder and other web archiving institutions, and usually stored with a .warc.gz extension.

It can be used to: - Compress an uncompressed WARC - Convert any ARC file to a compressed WARC - Fix an improperly compressed WARC file (eg. a WARC compressed entirely instead of by record)

warcio recompress ./input.arc.gz ./output.warc.gz

Extract

The extract command provides a way to extract either the WARC and HTTP headers and/or payload of a WARC record to stdout. Given a WARC filename and an offset, extract will print the (decompressed) record at that offset in the file to stdout

Specifying --payload or --headers will output only the payload or only the WARC + HTTP headers (if any), respectively.

warcio extract [--payload | --headers] filename offset

License

warcio is licensed under the Apache 2.0 License and is part of the Webrecorder project.

See NOTICE and LICENSE for details.