Usage¶

Search for historical mementos (archived copies) of a URL. Download metadata about the mementos and/or the memento content itself.

Tutorial¶

What is the earliest memento of nasa.gov?¶

Instantiate a WaybackClient.

In [1]: from wayback import WaybackClient

In [2]: client = WaybackClient()

Search for all Wayback’s records for nasa.gov.

In [3]: results = client.search('nasa.gov')

This statement should execute fairly quickly because it doesn’t actually do much work. The object we get back, results, is a generator, a “lazy” object from which we can pull results, one at a time. As we pull items out of it, it loads them as needed from the Wayback Machine in chronological order. We can see that results by itself is not informative:

In [4]: results
Out[4]: <generator object WaybackClient.search at 0x564952b78d60>

There are couple ways to pull items out of generator like results. One simple way is to use the built-in Python function next(), like so:

In [5]: record = next(results)

This takes a moment to run because, now that we’ve asked to see the first item in the generator, this lazy object goes to fetch a chunk of results from the Wayback Machine. Looking at the record in detail,

In [6]: record
Out[6]: CdxRecord(key='gov,nasa)/', timestamp=datetime.datetime(1996, 12, 31, 23, 58, 47, tzinfo=datetime.timezone.utc), url='http://www.nasa.gov/', mime_type='text/html', status_code=200, digest='MGIGF4GRGGF5GKV6VNCBAXOE3OR5BTZC', length=1811, raw_url='https://web.archive.org/web/19961231235847id_/http://www.nasa.gov/', view_url='https://web.archive.org/web/19961231235847/http://www.nasa.gov/')

we can find our answer: Wayback’s first memento of nasa.gov was in 1996. We can use dot access on record to access the timestamp specifically.

In [7]: record.timestamp
Out[7]: datetime.datetime(1996, 12, 31, 23, 58, 47, tzinfo=datetime.timezone.utc)

How many times does the word ‘mars’ appear on nasa.gov?¶

Above, we access the metadata for the oldest memento on nasa.gov, stored in the variable record. Starting from where we left off, we’ll access the content of the memento and do a very simple analysis.

The Wayback Machine provides multiple playback modes to view the data it has captured. The wayback.Mode.view mode is a copy edited for human viewers on the web, and the wayback.Mode.original mode is the original copy of what was captured when the page was scraped. For analysis purposes, we generally want original. (Check the documentation of wayback.Mode for a few other, less commonly used modes.)

Let’s download the original content using WaybackClient. (You could download the content directly with an HTTP library like requests, but WaybackClient adds extra tools for dealing with Wayback Machine servers.)

In [8]: from wayback import Mode

# `Mode.original` is the default and doesn't need to be explicitly set;
# we've set it here to show how you might choose other modes.
In [9]: response = client.get_memento(record, mode=Mode.original)

In [10]: content = response.content.decode()

We can use the built-in method count on strings to count the number of times that 'mars' appears in the content.

In [11]: content.count('mars')
Out[11]: 30

This is case-sensitive, so to be more accurate we should convert the content to lowercase first.

In [12]: content.lower().count('mars')
Out[12]: 39

We picked up a couple additional occurrences that the original count missed.

API Documentation¶

The Wayback Machine exposes its data through two different mechanisms, implementing two different standards for archival data, the CDX API and the Memento API. We implement a Python client that can speak both.

class wayback.WaybackClient(session=None)[source]¶

A client for retrieving data from the Internet Archive’s Wayback Machine.

You can use a WaybackClient as a context manager. When exiting, it will close the session it’s using (if you’ve passed in a custom session, make sure not to use the context manager functionality unless you want to live dangerously).

Parameters:

sessionWaybackSession, optional

search(url, *, match_type=None, limit=1000, offset=None, fast_latest=None, from_date=None, to_date=None, filter_field=None, collapse=None, resolve_revisits=True, skip_malformed_results=True, matchType=None, fastLatest=None, resolveRevisits=None)[source]¶

Search archive.org’s CDX API for all captures of a given URL. This returns an iterator of CdxRecord objects. The StopIteration value is the total count of found captures.

Results include captures with similar, but not exactly matching URLs. They are matched by a SURT-formatted, canonicalized URL that:

Does not differentiate between HTTP and HTTPS,
Is not case-sensitive, and
Treats www. and www*. subdomains the same as no subdomain at all.

This will automatically page through all results for a given search. If you want fewer results, you can stop iterating early:

from itertools import islice
first10 = list(islice(client.search(...), 10))

Parameters:

urlstr

The URL to search for captures of.

Special patterns in url imply a value for the match_type parameter and match multiple URLs:

If the URL starts with *. (e.g. *.epa.gov) OR match_type='domain', the search will include all URLs at the given domain and its subdomains.
If the URL ends with /* (e.g. https://epa.gov/*) OR match_type='prefix', the search will include all URLs that start with the text up to the *.
Otherwise, this returns matches just for the requeted URL.

match_typestr, optional

Determines how to interpret the url parameter. It must be one of the following:

exact (default) returns results matching the requested URL (see notes about SURT above; this is not an exact string match of the URL you pass in).
prefix returns results that start with the requested URL.
host returns results from all URLs at the host in the requested URL.
domain returns results from all URLs at the domain or any subdomain of the requested URL.

The default value is calculated based on the format of url.

limitint, default: 1000

Maximum number of results per request to the API (not the maximum number of results this function yields).

Negative values return the most recent N results.

Positive values are complicated! The search server will only scan so much data on each query, and if it finds fewer than limit results before hitting its own internal limits, it will behave as if if there are no more results, even though there may be.

Unfortunately, ideal values for limit aren’t very predicatable because the search server combines data from different sources, and they do not all behave the same. Their parameters may also be changed over time.

In general…

The default value should work well in typical cases.
For frequently captured URLs, you may want to set a higher value (e.g. 12,000) for more efficient querying.
For infrequently captured URLs, you may want to set a lower value (e.g. 100 or even 10) to ensure that your query does not hit internal limits before returning.
For extremely infrequently captured URLs, you may simply want to call search() multiple times with different, close together from_date and to_date values.

offsetint, optional

Skip the first N results.

fast_latestbool, optional

Get faster results when using a negative value for limit. It may return a variable number of results that doesn’t match the value of limit. For example, search('http://epa.gov', limit=-10, fast_latest=True) may return any number of results between 1 and 10.

from_datedatetime or date, optional

Only include captures after this date. Equivalent to the from argument in the CDX API. If it does not have a time zone, it is assumed to be in UTC.

to_datedatetime or date, optional

Only include captures before this date. Equivalent to the to argument in the CDX API. If it does not have a time zone, it is assumed to be in UTC.

filter_fieldstr or list of str or tuple of str, optional

A filter or list of filters for any field in the results. Equivalent to the filter argument in the CDX HTTP API. Format: [!]field:regex or ~[!]field:substring, e.g. '!statuscode:200' to select only captures with a non-2xx status code, or '~urlkey:feature' to select only captures where the SURT-formatted URL key has “feature” somewhere in it.

To apply multiple filters, use a list or tuple of strings instead of a single string, e.g. ['statuscode:200', 'urlkey:.*feature.*'].

Regexes are matched against the entire field value. For example, 'statuscode:2' will never match, because all statuscode values are three characters. Instead, to match all status codes with a “2” in them, use 'statuscode:.*2.*'. Add a ! at before the field name to negate the match.

Valid field names are: urlkey, timestamp, original, mimetype, statuscode, digest, length.

collapsestr, optional

Collapse consecutive results that match on a given field. (format: fieldname or fieldname:N – N is the number of chars to match.)

resolve_revisitsbool, default: True

Attempt to resolve warc/revisit records to their actual content type and response code. Not supported on all CDX servers.

skip_malformed_resultsbool, default: True

If true, don’t yield records that look like they have no actual memento associated with them. Some crawlers will erroneously attempt to capture bad URLs like http://mailto:someone@domain.com or http://data:image/jpeg;base64,AF34... and so on. This is a filter performed client side and is not a CDX API argument.

Yields:

version: CdxRecord: A CdxRecord encapsulating one capture or revisit

Raises:

UnexpectedResponseFormat: If the CDX response was not parseable.

Notes

Several CDX API parameters are not relevant or handled automatically by this function. This does not support: output, fl, showDupeCount, showSkipCount, lastSkipTimestamp, showNumPages, showPagedIndex.

It also does not support page and pageSize for pagination because they work differently from the resumeKey method this uses, and results do not include recent captures when using them.

References

HTTP API Docs: https://github.com/internetarchive/wayback/tree/master/wayback-cdx-server
SURT formatting: http://crawler.archive.org/articles/user_manual/glossary.html#surt
SURT implementation: https://github.com/internetarchive/surt

get_memento(url, timestamp=None, mode=Mode.original, *, exact=True, exact_redirects=None, target_window=86400, follow_redirects=True, datetime=None)[source]¶

Fetch a memento (an archived HTTP response) from the Wayback Machine.

Not all mementos can be successfully fetched (or “played back” in Wayback terms). In this case, get_memento can load the next-closest-in-time memento or it will raise wayback.exceptions.MementoPlaybackError depending on the value of the exact and exact_redirects parameters (see more details below).

Parameters:

urlstr or CdxRecord

URL to retrieve a memento of. This can be any of:

A normal URL (e.g. http://www.noaa.gov/). When using this form, you must also specify timestamp.
A CdxRecord retrieved from wayback.WaybackClient.search().
A URL of the memento in Wayback, e.g. https://web.archive.org/web/20180816111911id_/http://www.noaa.gov/

timestampdatetime.datetime or datetime.date or str, optional

The time at which to retrieve a memento of url. If url is a wayback.CdxRecord or full memento URL, this parameter can be omitted.

modewayback.Mode or str, default: wayback.Mode.original

The playback mode of the memento. This determines whether the content of the returned memento is exactly as originally captured (the default) or modified in some way. See wayback.Mode for a description of possible values.

For more details, see: https://archive-access.sourceforge.net/projects/wayback/administrator_manual.html#Archival_URL_Replay_Mode

exactbool, default: True

If false and the requested memento either doesn’t exist or can’t be played back, this returns the closest-in-time memento to the requested one, so long as it is within target_window. If there was no memento in the target window or if exact=True, then this will raise wayback.exceptions.MementoPlaybackError.

exact_redirectsbool, optional

If false and the requested memento is a redirect whose target doesn’t exist or can’t be played back, this returns the closest-in-time memento to the intended target, so long as it is within target_window. If unset, this will be the same as exact.

target_windowint, default: 86400

If the memento is of a redirect, allow up to this many seconds between the capture of the redirect and the capture of the redirect’s target URL. This window also applies to the first memento if exact=False and the originally requested memento was not available. Defaults to 86,400 (24 hours).

follow_redirectsbool, default: True

If true (the default), get_memento will follow historical redirects to return the content that a web browser would have ultimately displayed at the requested URL and time, rather than the memento of an HTTP redirect response (i.e. a 3xx status code). That is, if http://example.com/a redirected to http://example.com/b, then this method returns the memento for /a when follow_redirects=False and the memento for /b when follow_redirects=True.

Returns:

Memento: A Memento object with information about the archived HTTP response.

class wayback.CdxRecord(key, timestamp, url, mime_type, status_code, digest, length, raw_url, view_url)¶

Item from iterable of results returned by WaybackClient.search()

These attributes contain information provided directly by CDX.

digest¶: Content hashed as a base 32 encoded SHA-1.

key¶: SURT-formatted URL

length¶: Size of captured content in bytes, such as 2767. This may be inaccurate, and may even be None instead of an integer. If the record is a “revisit record”, indicated by MIME type 'warc/revisit', the length seems to be the length of the reference, not the length of the content itself. In other cases, the record has no length information at all, and this attribute will be None instead of a number.

mime_type¶: MIME type of record, such as 'text/html', 'warc/revisit' or 'unk' (“unknown”) if this information was not captured.

status_code¶: Status code returned by the server when the record was captured, such as 200. This is may be None if the record is a revisit record.

timestamp¶: The capture time represented as a datetime.datetime, such as datetime.datetime(1996, 12, 31, 23, 58, 47, tzinfo=timezone.utc).

url¶: The URL that was captured by this record, such as 'http://www.nasa.gov/'.

And these attributes are synthesized from the information provided by CDX.

raw_url¶: The URL to the raw captured content, such as 'https://web.archive.org/web/19961231235847id_/http://www.nasa.gov/'.

view_url¶: The URL to the public view on Wayback Machine. In this view, the links and some subresources in the document are rewritten to point to Wayback URLs. There is also a navigation panel around the content. Example URL: 'https://web.archive.org/web/19961231235847/http://www.nasa.gov/'.

class wayback.Memento(*, url, timestamp, mode, memento_url, status_code, headers, encoding, raw, raw_headers, links, history, debug_history)[source]¶

Represents a memento (an archived HTTP response). This object is similar to a response object from the popular “Requests” package, although it has some differences designed to differentiate historical information vs. current metadata about the stored memento (for example, the headers attribute lists the headers recorded in the memento, and does not include additional headers that provide metadata about the Wayback Machine).

Note that, like an HTTP response, this object represents a potentially open network connection to the Wayback Machine. Reading the content or text attributes will read all the data being received and close the connection automatically, but if you do not read those properties, you must make sure to call close() to close to connection. Alternatively, you can use a Memento as a context manager. The connection will be closed for you when the context ends:

>>> with a_memento:
>>>     do_something()
>>> # Connection is automatically closed here.

Fields

encoding: str¶: The text encoding of the response, e.g. 'utf-8'.

headers: dict¶

A dict representing the headers of the archived HTTP response. The keys are case-insensitive. If you iterate over it, you will receive the header names as they were originally sent. However, you can look them up via strings that vary in upper/lower-case. For example:

list(memento.headers) == ['Content-Type', 'Date']
memento.headers['Content-Type'] == memento.headers['content-type']

history: tuple[wayback.Memento]¶: A list of wayback.Memento objects that were redirects and were followed to produce this memento.

debug_history: tuple[str]¶: List of all URLs redirects followed in order to produce this memento. These are “memento URLs” – that is, they are absolute URLs to the Wayback machine like https://web.archive.org/web/20180816111911id_/http://www.noaa.gov/, rather than URLs of captured redirects, like http://www.noaa.gov. Many of the URLs in this list do not represent actual mementos.

status_code: int¶: The HTTP status code of the archived HTTP response.

mode: str¶: The playback mode used to produce the Memento.

timestamp: datetime.datetime¶: The time the memento was originally captured. This includes tzinfo, and will always be in UTC.

url: str¶: The URL that the memento represents, e.g. http://www.noaa.gov.

memento_url: str¶: The URL at which the memento was fetched from the Wayback Machine, e.g. https://web.archive.org/web/20180816111911id_/http://www.noaa.gov/.

ok: bool¶: Whether the response had an non-error status (i.e. < 400).

is_redirect: bool¶: Whether the response was a redirect (i.e. had a 3xx status).

content: bytes¶: The body of the archived HTTP response in bytes.

text: str¶: The body of the archived HTTP response decoded as a string.

links: dict of (str, dict of (str, str))¶

Related links to this Memento (e.g. the previous and/or next Memento in time). The keys are the relationship (e.g. 'prev memento') as a string and the values are dicts where the keys and values are strings.

In each entry, the 'url' key is the URL of the related link, the 'rel' key is the relationship (the same as the key in the top-level dict), and the rest of the keys will be any other attributes that are relevant for that link (e.g. 'datetime' or 'type').

For example:

{
    'original': {
        'url': 'https://www.fws.gov/birds/',
        'rel': 'original'
    },
    'first memento': {
        'url': 'https://web.archive.org/web/20050323155300/http://www.fws.gov:80/birds',
        'rel': 'first memento',
        'datetime': 'Wed, 23 Mar 2005 15:53:00 GMT'
    },
    'prev memento': {
        'url': 'https://web.archive.org/web/20210125125216/https://www.fws.gov/birds/',
        'rel': 'prev memento',
        'datetime': 'Mon, 25 Jan 2021 12:52:16 GMT'
    },
    'next memento': {
        'url': 'https://web.archive.org/web/20210321180831/https://www.fws.gov/birds',
        'rel': 'next memento',
        'datetime': 'Sun, 21 Mar 2021 18:08:31 GMT'
    },
    'last memento': {
        'url': 'https://web.archive.org/web/20221006031005/https://fws.gov/birds',
        'rel': 'last memento',
        'datetime': 'Thu, 06 Oct 2022 03:10:05 GMT'
    }
}

Links to other mementos use the same mode as the memento object this links attribute belongs to. For example:

raw_memento = client.get_memento('https://fws.gov/birds', '20210318004901')
raw_memento.links['next memento']['url'] == 'https://web.archive.org/web/20210321180831id_/https://fws.gov/birds'
# The "id_" after the timestamp means "original" mode ---------------------------------^^^

view_memento = client.get_memento('https://fws.gov/birds', '20210318004901', mode=Mode.view)
view_memento.links['next memento']['url'] == 'https://web.archive.org/web/20210321180831/https://fws.gov/birds'
# Nothing after the timestamp for "view" mode -----------------------------------------^

close()[source]¶: Close the HTTP response for this Memento. This happens automatically if you read content or text, and if you use the memento as a context manager. This method is always safe to call – it does nothing if the response has already been closed.

classmethod parse_memento_headers(raw_headers, url='https://web.archive.org/')[source]¶

Extract historical headers from the Memento HTTP response’s headers.

Parameters:

raw_headersdict: A dict of HTTP headers from the Memento’s HTTP response.
urlstr, optional: The URL of the resource the headers are being parsed for. It’s used when header data contains relative/incomplete URL information.

Returns:

dict

class wayback.WaybackSession(retries=6, backoff=2, timeout=60, user_agent=None, search_calls_per_second=1, memento_calls_per_second=30)[source]¶

A custom session object that pools network connections and resources for requests to the Wayback Machine.

Parameters:

retriesint, default: 6

The maximum number of retries for requests.

backoffint or float, default: 2

Number of seconds from which to calculate how long to back off and wait when retrying requests. The first retry is always immediate, but subsequent retries increase by powers of 2:

seconds = backoff * 2 ^ (retry number - 1)

So if this was 4, retries would happen after the following delays: 0 seconds, 4 seconds, 8 seconds, 16 seconds, …

timeoutint or float or tuple of (int or float, int or float), default: 60

A timeout to use for all requests. See the Requests docs for more: https://docs.python-requests.org/en/master/user/advanced/#timeouts

user_agentstr, optional

A custom user-agent string to use in all requests. Defaults to: wayback/{version} (+https://github.com/edgi-govdata-archiving/wayback)

search_calls_per_secondint or float, default: 1

The maximum number of calls made to the search API per second. To disable the rate limit, set this to 0.

memento_calls_per_secondint or float, default: 30

The maximum number of calls made to the memento API per second. To disable the rate limit, set this to 0.

reset()[source]¶: Reset any network connections the session is using.

Utilities¶

wayback.memento_url_data(memento_url)[source]¶

Get the original URL, time, and mode that a memento URL represents a capture of.

Returns:

urlstr: The URL that the memento is a capture of.
timedatetime.datetime: The time the memento was captured in the UTC timezone.
modestr: The playback mode.

Examples

Extract original URL, time and mode.

>>> url = ('https://web.archive.org/web/20170813195036id_/'
...        'https://arpa-e.energy.gov/?q=engage/events-workshops')
>>> memento_url_data(url)
('https://arpa-e.energy.gov/?q=engage/events-workshops',
 datetime.datetime(2017, 8, 13, 19, 50, 36, tzinfo=timezone.utc),
 'id_')

class wayback.Mode(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)[source]¶

An enum describing the playback mode of a memento. When requesting a memento (e.g. with wayback.WaybackClient.get_memento()), you can use these values to determine how the response body should be formatted.

For more details, see: https://archive-access.sourceforge.net/projects/wayback/administrator_manual.html#Archival_URL_Replay_Mode

Examples

>>> waybackClient.get_memento('https://noaa.gov/',
>>>                           timestamp=datetime.datetime(2018, 1, 2),
>>>                           mode=wayback.Mode.view)

Values

original¶: Returns the HTTP response body as originally captured.

view¶: Formats the response body so it can be viewed with a web browser. URLs for links and subresources like scripts, stylesheets, images, etc. will be modified to point to the equivalent memento in the Wayback Machine so that the resulting page looks as similar as possible to how it would have appeared when originally captured. It’s mainly meant for use with HTML pages. This is the playback mode you typically use when browsing the Wayback Machine with a web browser.

javascript¶: Formats the response body by updating URLs, similar to Mode.view, but designed for JavaScript instead of HTML.

css¶: Formats the response body by updating URLs, similar to Mode.view, but designed for CSS instead of HTML.

image¶: formats the response body similar to Mode.view, but designed for image files instead of HTML.

Exception Classes¶

class wayback.exceptions.WaybackException[source]¶: Base exception class for all Wayback-specific errors.

class wayback.exceptions.UnexpectedResponseFormat[source]¶: Raised when data returned by the Wayback Machine is formatted in an unexpected or unparseable way.

class wayback.exceptions.BlockedByRobotsError[source]¶: Raised when a URL can’t be queried in Wayback because it was blocked by a site’s robots.txt file.

class wayback.exceptions.BlockedSiteError[source]¶: Raised when a URL has been blocked from access or querying in Wayback. This is often because of a takedown request. (URLs that are blocked because of robots.txt get a BlockedByRobotsError instead.)

class wayback.exceptions.MementoPlaybackError[source]¶: Raised when a Memento can’t be ‘played back’ (loaded) by the Wayback Machine for some reason. This is a server-side issue, not a problem in parsing data from Wayback.

class wayback.exceptions.NoMementoError[source]¶

Raised when there was no memento available for a given URL. This might mean the given URL has no mementos at all or that none that are available for playback.

This also means you should not try to request a memento of the same URL in a different timeframe. If there may be other mementos of the URL available, you’ll get a different error.

class wayback.exceptions.RateLimitError(response, retry_after)[source]¶

Raised when the Wayback Machine responds with a 429 (too many requests) status code. In general, this package’s built-in limits should help you avoid ever hitting this, but if you are running multiple processes in parallel, you could go overboard.

Attributes:

retry_afterint, optional: Recommended number of seconds to wait before retrying. If the Wayback Machine does not include it in the HTTP response, it will be set to None.

class wayback.exceptions.WaybackRetryError(retries, total_time, causal_error)[source]¶

Raised when a request to the Wayback Machine has been retried and failed too many times. The number of tries before this exception is raised generally depends on your WaybackSession settings.

Attributes:

retriesint: The number of retries that were attempted.
causeException: The actual, underlying error that would have caused a retry.
timeint: The total time spent across all retried requests, in seconds.

class wayback.exceptions.SessionClosedError[source]¶: Raised when a Wayback session is used to make a request after it has been closed and disabled.