Warcprox - WARC writing MITM HTTP/S proxy

Warcprox is an HTTP proxy designed for web archiving applications. When used in parallel with brozzler it supports a comprehensive, modern, and distributed archival web capture system. Warcprox stores its traffic to disk in the Web ARChive (WARC) file format, which may then be accessed with web archival replay software like OpenWayback and pywb. It captures encrypted HTTPS traffic by using the "man-in-the-middle" technique (see the Man-in-the-middle section for more info).

Warcprox was originally based on pymiproxy by Nadeem Douba.

Contents

• Getting started
• Man-in-the-middle
• API
• Deduplication
• Statistics
• Plugins
• Architecture
• License

Getting started

Warcprox runs on python 3.4+.

To install latest release run:

# apt-get install libffi-dev libssl-dev
pip install warcprox

You can also install the latest bleeding edge code:

pip install git+https://github.com/internetarchive/warcprox.git

To start warcprox run:

warcprox

Try warcprox --help for documentation on command line options.

Man-in-the-middle

Normally, HTTP proxies can't read encrypted HTTPS traffic. The browser uses the HTTP CONNECT method to establish a tunnel through the proxy, and the proxy merely routes raw bytes between the client and server. Since the bytes are encrypted, the proxy can't make sense of the information that it proxies. This nonsensical encrypted data is not typically useful for web archiving purposes.

In order to capture HTTPS traffic, warcprox acts as a "man-in-the-middle" (MITM). When it receives a CONNECT directive from a client, it generates a public key certificate for the requested site, presents to the client, and proceeds to establish an encrypted connection with the client. It then makes a separate, normal HTTPS connection to the remote site. It decrypts, archives, and re-encrypts traffic in both directions.

Configuring a warcprox instance as a browser’s HTTP proxy will result in security certificate warnings because none of the certificates will be signed by trusted authorities. However, there is nothing malicious about warcprox functions. To use warcprox effectively, the client needs to disable certificate verification or add the CA certificate generated by warcprox as a trusted authority. When using the latter, remember to undo this change when finished using warcprox.

API

The warcprox API may be used to retrieve information from and interact with a running warcprox instance, including:

• Retrieving status information via /status URL
• Writing WARC records via WARCPROX_WRITE_RECORD HTTP method
• Controlling warcprox settings via the Warcprox-Meta HTTP header

For warcprox API documentation, see: api.rst.

Deduplication

Warcprox avoids archiving redundant content by "deduplicating" it. The process for deduplication works similarly to deduplication by Heritrix and other web archiving tools:

1. While fetching URL, calculate payload content digest (typically SHA1 checksum value)
2. Look up digest in deduplication database (warcprox currently supports sqlite by default, rethinkdb with two different schemas, and trough)
3. If found, write warc revisit record referencing the url and capture time of the previous capture
4. If not found,
   1. Write response record with full payload
   2. Store new entry in deduplication database (can be disabled, see Warcprox-Meta HTTP request header)

The deduplication database is partitioned into different "buckets". URLs are deduplicated only against other captures in the same bucket. If specified, the dedup-buckets field of the Warcprox-Meta HTTP request header determines the bucket(s). Otherwise, the default bucket is used.

Deduplication can be disabled entirely by starting warcprox with the argument --dedup-db-file=/dev/null.

Statistics

Warcprox stores some crawl statistics to sqlite or rethinkdb. These are consulted for enforcing limits and soft-limits (see Warcprox-Meta fields), and can also be consulted by other processes outside of warcprox, such as for crawl job reporting.

Statistics are grouped by "bucket". Every capture is counted as part of the __all__ bucket. Other buckets can be specified in the Warcprox-Meta request header. The fallback bucket in case none is specified is called __unspecified__.

Within each bucket are three sub-buckets:

• new - tallies captures for which a complete record (usually a response record) was written to a WARC file
• revisit - tallies captures for which a revisit record was written to a WARC file
• total - includes all URLs processed, even those not written to a WARC file, and so may be greater than the sum of new and revisit records

Within each of these sub-buckets, warcprox generates two kinds of statistics:

• urls - simple count of URLs
• wire_bytes - sum of bytes received over the wire from the remote server for each URL, including HTTP headers

For historical reasons, the default sqlite store keeps statistics as JSON blobs:

sqlite> select * from buckets_of_stats;
bucket           stats
---------------  ---------------------------------------------------------------------------------------------
__unspecified__  {"bucket":"__unspecified__","total":{"urls":37,"wire_bytes":1502781},"new":{"urls":15,"wire_bytes":1179906},"revisit":{"urls":22,"wire_bytes":322875}}
__all__          {"bucket":"__all__","total":{"urls":37,"wire_bytes":1502781},"new":{"urls":15,"wire_bytes":1179906},"revisit":{"urls":22,"wire_bytes":322875}}

Plugins

Warcprox supports a limited notion of plugins by way of the --plugin command line argument. Plugin classes are loaded from the regular python module search path. They are instantiated with one argument that contains the values of all command line arguments, warcprox.Options. Legacy plugins with constructors that take no arguments are also supported. Plugins should either have a method notify(self, recorded_url, records) or should subclass warcprox.BasePostfetchProcessor. More than one plugin can be configured by specifying --plugin multiples times.

See a minimal example here.

Architecture

Warcprox is multithreaded. It has pool of http proxy threads (100 by default). When handling a request, a proxy thread records data from the remote server to an in-memory buffer that spills over to disk if necessary (after 512k by default), while it streams the data to the proxy client. Once the HTTP transaction is complete, it puts the recorded URL in a thread-safe queue, to be picked up by the first processor in the postfetch chain.

The postfetch chain normally includes processors for loading deduplication information, writing records to the WARC, saving deduplication information, and updating statistics. The exact set of processors in the chain depends on command line arguments; for example, plugins specified with --plugin are processors in the postfetch chain. Each postfetch processor has its own thread or threads. Thus the processors are able to run in parallel, independent of one another. This design also enables them to process URLs in batch. For example, the statistics processor gathers statistics for up to 10 seconds or 500 URLs, whichever comes first, then updates the statistics database with just a few queries.

License

Warcprox is a derivative work of pymiproxy, which is GPL. Thus warcprox is also GPL.

• Copyright (C) 2012 Cygnos Corporation
• Copyright (C) 2013-2018 Internet Archive

This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.