backup/warcprox
1
0
Fork 0
mirror of https://github.com/internetarchive/warcprox.git synced 2025-01-18 13:22:09 +01:00
Code Issues Projects Releases Activity

Merge pull request #100 from nlevitt/karl-copy-edits

Browse Source 
Karl's copy edits
This commit is contained in:
Noah Levitt 2018-08-16 17:08:14 -07:00 committed by GitHub
commit 8be7ddee2b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 71 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

141
README.rst
View File

@ -3,22 +3,19 @@ Warcprox - WARC writing MITM HTTP/S proxy
.. image:: https://travis-ci.org/internetarchive/warcprox.svg?branch=master
:target: https://travis-ci.org/internetarchive/warcprox
Warcprox is a tool for archiving the web. It is an http proxy that stores its
traffic to disk in `WARC
<https://iipc.github.io/warc-specifications/specifications/warc-format/warc-1.1/>`_
format. Warcprox captures encrypted https traffic by using the
`"man-in-the-middle" <https://en.wikipedia.org/wiki/Man-in-the-middle_attack>`_
technique (see the `Man-in-the-middle`_ section for more info).
Warcprox is an HTTP proxy designed for web archiving applications. When used in
parallel with `brozzler <https://github.com/internetarchive/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
<https://iipc.github.io/warc-specifications/specifications/warc-format/warc-1.1/>`_,
which may then be accessed with web archival replay software like `OpenWayback
<https://github.com/iipc/openwayback>`_ and `pywb
<https://github.com/webrecorder/pywb>`_. It captures encrypted HTTPS traffic by
using the "man-in-the-middle" technique (see the `Man-in-the-middle`_ section
for more info).
The web pages that warcprox stores in WARC files can be played back using
software like `OpenWayback <https://github.com/iipc/openwayback>`_ or `pywb
<https://github.com/webrecorder/pywb>`_. Warcprox has been developed in
parallel with `brozzler <https://github.com/internetarchive/brozzler>`_ and
together they make a comprehensive modern distributed archival web crawling
system.
Warcprox was originally based on the excellent and simple pymiproxy by Nadeem
Douba. https://github.com/allfro/pymiproxy
Warcprox was originally based on `pymiproxy
<https://github.com/allfro/pymiproxy>`_ by Nadeem Douba.
.. contents::
@ -43,68 +40,72 @@ Try ``warcprox --help`` for documentation on command line options.
Man-in-the-middle
=================
Normally, http proxies can't read https traffic, because it's encrypted. 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
it's proxying. This nonsensical encrypted data would not be very useful to
archive.
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"
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. Then it makes a
separate, normal https connection to the remote site. It decrypts, archives,
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.
Although "man-in-the-middle" is often paired with "attack", there is nothing
malicious about what warcprox is doing. If you configure an instance of
warcprox as your browser's http proxy, you will see lots of certificate
warnings, since none of the certificates will be signed by trusted authorities.
To use warcprox effectively the client needs to disable certificate
verification, or add the CA cert generated by warcprox as a trusted authority.
(If you do this in your browser, make sure you undo it when you're done using
warcprox!)
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
===
For interacting with a running instance of warcprox.
The warcprox API may be used to retrieve information from and interact with a
running warcprox instance, including:
* ``/status`` url
* ``WARCPROX_WRITE_RECORD`` http method
* ``Warcprox-Meta`` http request header and response header
* Retrieving status information via ``/status`` URL
* Writing WARC records via ``WARCPROX_WRITE_RECORD`` HTTP method
* Controlling warcprox settings via the ``Warcprox-Meta`` HTTP header
See `<api.rst>`_.
For warcprox API documentation, see: `<api.rst>`_.
Deduplication
=============
Warcprox avoids archiving redundant content by "deduplicating" it. The process
for deduplication works similarly to heritrix and other web archiving tools.
for deduplication works similarly to deduplication by `Heritrix
<https://github.com/internetarchive/heritrix3>`_ and other web archiving tools:
1. while fetching url, calculate payload content digest (typically sha1)
2. look up digest in deduplication database (warcprox supports a few different
ones)
3. if found, write warc ``revisit`` record referencing the url and capture time
1. While fetching URL, calculate payload content digest (typically SHA1
checksum value)
2. Look up digest in deduplication database (warcprox currently supports
`sqlite <https://sqlite.org/>`_ by default, `rethinkdb
<https://github.com/rethinkdb/rethinkdb>`_ with two different schemas, and
`trough <https://github.com/internetarchive/trough>`_)
3. If found, write warc ``revisit`` record referencing the url and capture time
of the previous capture
4. else (if not found),
4. If not found,
a. write warc ``response`` record with full payload
b. store entry in deduplication database
a. Write ``response`` record with full payload
b. Store new entry in deduplication database
The dedup database is partitioned into different "buckets". Urls are
The deduplication database is partitioned into different "buckets". URLs are
deduplicated only against other captures in the same bucket. If specified, the
``dedup-bucket`` field of the ``Warcprox-Meta`` http request header determines
the bucket, otherwise the default bucket is used.
``dedup-bucket`` field of the `Warcprox-Meta HTTP request header
<api.rst#warcprox-meta-http-request-header>`_ determines the bucket. Otherwise,
the default bucket is used.
Deduplication can be disabled entirely by starting warcprox with the argument
``--dedup-db-file=/dev/null``.
Statistics
==========
Warcprox keeps some crawl statistics and stores them in sqlite or rethinkdb.
These are consulted for enforcing ``limits`` and ``soft-limits`` (see
`<api.rst#warcprox-meta-fields>`_), and can also be consulted by other
processes outside of warcprox, for reporting etc.
Warcprox stores some crawl statistics to sqlite or rethinkdb. These are
consulted for enforcing ``limits`` and ``soft-limits`` (see `Warcprox-Meta
fields <api.rst#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``
@ -113,21 +114,20 @@ request header. The fallback bucket in case none is specified is called
Within each bucket are three sub-buckets:
* ``new`` - tallies captures for which a complete record (usually a ``response``
record) was written to warc
* ``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
warc
* ``total`` - includes all urls processed, even those not written to warc (so the
numbers may be greater than new + revisit)
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 we keep two statistics:
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, including http headers,
from the remote server for each url
* ``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, in sqlite, the default store, statistics are kept as
json blobs::
For historical reasons, the default sqlite store keeps statistics as JSON blobs::
sqlite> select * from buckets_of_stats;
bucket stats
@ -139,14 +139,15 @@ 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 will be instantiated with one argument, a
``warcprox.Options``, which holds the values of all the command line arguments.
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.
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.
`A minimal example <https://github.com/internetarchive/warcprox/blob/318405e795ac0ab8760988a1a482cf0a17697148/warcprox/__init__.py#L165>`__
See a minimal example `here
<https://github.com/internetarchive/warcprox/blob/318405e795ac0ab8760988a1a482cf0a17697148/warcprox/__init__.py#L165>`__.
License
=======