Introduction | Requirements | Word Count | Motivation | Usage | Installation | Design Principles | Scripts | Command-line Interface | Contributing | Credits | More Info | Project Structure | License
riko is a pure Python library for analyzing and processing streams of
structured data. riko has synchronous and asynchronous APIs, supports parallel
execution, and is well suited for processing RSS feeds [1]. riko also supplies
a command-line interface for executing flows, i.e., stream processors aka workflows.
With riko, you can
- Read csv/xml/json/html files
- Create text and data based
flowsvia modular pipes - Parse, extract, and process RSS/Atom feeds
- Create awesome mashups [2], APIs, and maps
- Perform parallel processing via cpus/processors or threads
- and much more...
| [1] | Really Simple Syndication |
| [2] | Mashup (web application hybrid) |
riko has been tested and is known to work on Python 3.7, 3.8, and 3.9; and PyPy3.7.
| Feature | Dependency | Installation |
|---|---|---|
| Async API | Twisted | pip install riko[async] |
| Accelerated xml parsing | lxml [3] | pip install riko[xml] |
| Accelerated feed parsing | speedparser [4] | pip install riko[xml] |
| [3] | If lxml isn't present, riko will default to the builtin Python xml parser |
| [4] | If speedparser isn't present, riko will default to feedparser |
In this example, we use several pipes to count the words on a webpage.
>>> ### Create a SyncPipe flow ###
>>> #
>>> # `SyncPipe` is a convenience class that creates chainable flows
>>> # and allows for parallel processing.
>>> from riko.collections import SyncPipe
>>>
>>> ### Set the pipe configurations ###
>>> #
>>> # Notes:
>>> # 1. the `detag` option will strip all html tags from the result
>>> # 2. fetch the text contained inside the 'body' tag of the hackernews
>>> # homepage
>>> # 3. replace newlines with spaces and assign the result to 'content'
>>> # 4. tokenize the resulting text using whitespace as the delimeter
>>> # 5. count the number of times each token appears
>>> # 6. obtain the raw stream
>>> # 7. extract the first word and its count
>>> # 8. extract the second word and its count
>>> # 9. extract the third word and its count
>>> url = 'https://news.ycombinator.com/'
>>> fetch_conf = {
... 'url': url, 'start': '<body>', 'end': '</body>', 'detag': True} # 1
>>>
>>> replace_conf = {
... 'rule': [
... {'find': '\r\n', 'replace': ' '},
... {'find': '\n', 'replace': ' '}]}
>>>
>>> flow = (
... SyncPipe('fetchpage', conf=fetch_conf) # 2
... .strreplace(conf=replace_conf, assign='content') # 3
... .tokenizer(conf={'delimiter': ' '}, emit=True) # 4
... .count(conf={'count_key': 'content'})) # 5
>>>
>>> stream = flow.output # 6
>>> next(stream) # 7
{"'sad": 1}
>>> next(stream) # 8
{'(': 28}
>>> next(stream) # 9
{'(1999)': 1}Yahoo! Pipes [5] was a user friendly web application used to
aggregate, manipulate, and mashup content from around the web
Wanting to create custom pipes, I came across pipe2py which translated a
Yahoo! Pipe into python code. pipe2py suited my needs at the time
but was unmaintained and lacked asynchronous or parallel processing.
riko addresses the shortcomings of pipe2py but removed support for
importing Yahoo! Pipes json workflows. riko contains ~ 40 built-in
modules, aka pipes, that allow you to programatically perform most of the
tasks Yahoo! Pipes allowed.
riko provides a number of benefits / differences from other stream processing
applications such as Huginn, Flink, Spark, and Storm [6]. Namely:
- a small footprint (CPU and memory usage)
- native RSS/Atom support
- simple installation and usage
- a pure python library with pypy support
- builtin modular
pipesto filter, sort, and modifystreams
The subsequent tradeoffs riko makes are:
- not distributed (able to run on a cluster of servers)
- no GUI for creating
flows - doesn't continually monitor
streamsfor new data - can't react to specific events
- iterator (pull) based so streams only support a single consumer [7]
The following table summarizes these observations:
| library | Stream Type | Footprint | RSS | simple [8] | async | parallel | CEP [9] | distributed |
|---|---|---|---|---|---|---|---|---|
| riko | pull | small | √ | √ | √ | √ | ||
| pipe2py | pull | small | √ | √ | ||||
| Huginn | push | med | √ | [10] | √ | √ | ||
| Others | push | large | [11] | [12] | [13] | √ | √ | √ |
For more detailed information, please check-out the FAQ.
| [5] | Yahoo discontinued Yahoo! Pipes in 2015, but you can view what remains |
| [6] | Huginn, Flink, Spark, and Storm |
| [7] | You can mitigate this via the split module |
| [8] | Doesn't depend on outside services like MySQL, Kafka, YARN, ZooKeeper, or Mesos |
| [9] | Complex Event Processing |
| [10] | Huginn doesn't appear to make async web requests |
| [11] | Many libraries can't parse RSS streams without the use of 3rd party libraries |
| [12] | While most libraries offer a local mode, many require integrating with a data ingestor (e.g., Flume/Kafka) to do anything useful |
| [13] | I can't find evidence that these libraries offer an async APIs (and apparently Spark doesn't) |
riko is intended to be used directly as a Python library.
riko can fetch rss feeds from both local and remote filepaths via "source"
pipes. Each "source" pipe returns a stream, i.e., an iterator of
dictionaries, aka items.
>>> from riko.modules import fetch, fetchsitefeed
>>>
>>> ### Fetch an RSS feed ###
>>> stream = fetch.pipe(conf={'url': 'https://news.ycombinator.com/rss'})
>>>
>>> ### Fetch the first RSS feed found ###
>>> stream = fetchsitefeed.pipe(conf={'url': 'http://arstechnica.com/rss-feeds/'})
>>>
>>> ### View the fetched RSS feed(s) ###
>>> #
>>> # Note: regardless of how you fetch an RSS feed, it will have the same
>>> # structure
>>> item = next(stream)
>>> item.keys()
dict_keys(['title_detail', 'author.uri', 'tags', 'summary_detail', 'author_detail',
'author.name', 'y:published', 'y:title', 'content', 'title', 'pubDate',
'guidislink', 'id', 'summary', 'dc:creator', 'authors', 'published_parsed',
'links', 'y:id', 'author', 'link', 'published'])
>>> item['title'], item['author'], item['id']
('Gravity doesn’t care about quantum spin',
'Chris Lee',
'http://arstechnica.com/?p=924009')Please see the FAQ for a complete list of supported file types and protocols. Please see Fetching data and feeds for more examples.
riko can modify streams via the 40 built-in pipes
>>> from riko.collections import SyncPipe
>>>
>>> ### Set the pipe configurations ###
>>> fetch_conf = {'url': 'https://news.ycombinator.com/rss'}
>>> filter_rule = {'field': 'link', 'op': 'contains', 'value': '.com'}
>>> xpath = '/html/body/center/table/tr[3]/td/table[2]/tr[1]/td/table/tr/td[3]/span/span'
>>> xpath_conf = {'url': {'subkey': 'comments'}, 'xpath': xpath}
>>>
>>> ### Create a SyncPipe flow ###
>>> #
>>> # `SyncPipe` is a convenience class that creates chainable flows
>>> # and allows for parallel processing.
>>> #
>>> # The following flow will:
>>> # 1. fetch the hackernews RSS feed
>>> # 2. filter for items with '.com' in the link
>>> # 3. sort the items ascending by title
>>> # 4. fetch the first comment from each item
>>> # 5. flatten the result into one raw stream
>>> # 6. extract the first item's content
>>> #
>>> # Note: sorting is not lazy so take caution when using this pipe
>>>
>>> flow = (
... SyncPipe('fetch', conf=fetch_conf) # 1
... .filter(conf={'rule': filter_rule}) # 2
... .sort(conf={'rule': {'sort_key': 'title'}}) # 3
... .xpathfetchpage(conf=xpath_conf)) # 4
>>>
>>> stream = flow.output # 5
>>> next(stream)['content'] # 6
'Open Artificial Pancreas home:'Please see alternate workflow creation for an alternative (function based) method for
creating a stream. Please see pipes for a complete list of available pipes.
An example using riko's parallel API to spawn a ThreadPool [14]
>>> from riko.collections import SyncPipe
>>>
>>> ### Set the pipe configurations ###
>>> fetch_conf = {'url': 'https://news.ycombinator.com/rss'}
>>> filter_rule = {'field': 'link', 'op': 'contains', 'value': '.com'}
>>> xpath = '/html/body/center/table/tr[3]/td/table[2]/tr[1]/td/table/tr/td[3]/span/span'
>>> xpath_conf = {'url': {'subkey': 'comments'}, 'xpath': xpath}
>>>
>>> ### Create a parallel SyncPipe flow ###
>>> #
>>> # The following flow will:
>>> # 1. fetch the hackernews RSS feed
>>> # 2. filter for items with '.com' in the article link
>>> # 3. fetch the first comment from all items in parallel (using 4 workers)
>>> # 4. flatten the result into one raw stream
>>> # 5. extract the first item's content
>>> #
>>> # Note: no point in sorting after the filter since parallel fetching doesn't guarantee
>>> # order
>>> flow = (
... SyncPipe('fetch', conf=fetch_conf, parallel=True, workers=4) # 1
... .filter(conf={'rule': filter_rule}) # 2
... .xpathfetchpage(conf=xpath_conf)) # 3
>>>
>>> stream = flow.output # 4
>>> next(stream)['content'] # 5
'He uses the following example for when to throw your own errors:'To enable asynchronous processing, you must install the async module.
pip install riko[async]An example using riko's asynchronous API.
>>> from riko.bado import coroutine, react
>>> from riko.collections import AsyncPipe
>>>
>>> ### Set the pipe configurations ###
>>> fetch_conf = {'url': 'https://news.ycombinator.com/rss'}
>>> filter_rule = {'field': 'link', 'op': 'contains', 'value': '.com'}
>>> xpath = '/html/body/center/table/tr[3]/td/table[2]/tr[1]/td/table/tr/td[3]/span/span'
>>> xpath_conf = {'url': {'subkey': 'comments'}, 'xpath': xpath}
>>>
>>> ### Create an AsyncPipe flow ###
>>> #
>>> # The following flow will:
>>> # 1. fetch the hackernews RSS feed
>>> # 2. filter for items with '.com' in the article link
>>> # 3. asynchronously fetch the first comment from each item (using 4 connections)
>>> # 4. flatten the result into one raw stream
>>> # 5. extract the first item's content
>>> #
>>> # Note: no point in sorting after the filter since async fetching doesn't guarantee
>>> # order
>>> @coroutine
... def run(reactor):
... stream = yield (
... AsyncPipe('fetch', conf=fetch_conf, connections=4) # 1
... .filter(conf={'rule': filter_rule}) # 2
... .xpathfetchpage(conf=xpath_conf) # 3
... .output) # 4
...
... print(next(stream)['content']) # 5
>>>
>>> try:
... react(run)
... except SystemExit:
... pass
Here's how iteration works ():Please see the cookbook or ipython notebook for more examples.
| [14] | You can instead enable a ProcessPool by additionally passing threads=False to SyncPipe, i.e., SyncPipe('fetch', conf={'url': url}, parallel=True, threads=False). |
(You are using a virtualenv, right?)
At the command line, install riko using either pip (recommended)
pip install rikoor easy_install
easy_install rikoPlease see the installation doc for more details.
The primary data structures in riko are the item and stream. An item
is just a python dictionary, and a stream is an iterator of items. You can
create a stream manually with something as simple as
[{'content': 'hello world'}]. You manipulate streams in
riko via pipes. A pipe is simply a function that accepts either a
stream or item, and returns a stream. pipes are composable: you
can use the output of one pipe as the input to another pipe.
riko pipes come in two flavors; operators and processors.
operators operate on an entire stream at once and are unable to handle
individual items. Example operators include count, pipefilter,
and reverse.
>>> from riko.modules.reverse import pipe
>>>
>>> stream = [{'title': 'riko pt. 1'}, {'title': 'riko pt. 2'}]
>>> next(pipe(stream))
{'title': 'riko pt. 2'}processors process individual items and can be parallelized across
threads or processes. Example processors include fetchsitefeed,
hash, pipeitembuilder, and piperegex.
>>> from riko.modules.hash import pipe
>>>
>>> item = {'title': 'riko pt. 1'}
>>> stream = pipe(item, field='title')
>>> next(stream)
{'title': 'riko pt. 1', 'hash': 2853617420}Some processors, e.g., pipetokenizer, return multiple results.
>>> from riko.modules.tokenizer import pipe
>>>
>>> item = {'title': 'riko pt. 1'}
>>> tokenizer_conf = {'delimiter': ' '}
>>> stream = pipe(item, conf=tokenizer_conf, field='title')
>>> next(stream)
{'tokenizer': [{'content': 'riko'},
{'content': 'pt.'},
{'content': '1'}],
'title': 'riko pt. 1'}
>>> # In this case, if we just want the result, we can `emit` it instead
>>> stream = pipe(item, conf=tokenizer_conf, field='title', emit=True)
>>> next(stream)
{'content': 'riko'}operators are split into sub-types of aggregators
and composers. aggregators, e.g., count, combine
all items of an input stream into a new stream with a single item;
while composers, e.g., filter, create a new stream containing
some or all items of an input stream.
>>> from riko.modules.count import pipe
>>>
>>> stream = [{'title': 'riko pt. 1'}, {'title': 'riko pt. 2'}]
>>> next(pipe(stream))
{'count': 2}In case you are confused from the "Word Count" example up top, count can return
multiple items if you pass in the count_key config option.
>>> counted = pipe(stream, conf={'count_key': 'title'})
>>> next(counted)
{'riko pt. 1': 1}
>>> next(counted)
{'riko pt. 2': 1}processors are split into sub-types of source and transformer.
sources, e.g., itembuilder, can create a stream while
transformers, e.g. hash can only transform items in a stream.
>>> from riko.modules.itembuilder import pipe
>>>
>>> attrs = {'key': 'title', 'value': 'riko pt. 1'}
>>> next(pipe(conf={'attrs': attrs}))
{'title': 'riko pt. 1'}The following table summaries these observations:
| type | sub-type | input | output | parallelizable? | creates streams? |
| operator | aggregator | stream | stream [15] | ||
| composer | stream | stream | |||
| processor | source | item | stream | √ | √ |
| transformer | item | stream | √ |
If you are unsure of the type of pipe you have, check its metadata.
>>> from riko.modules import fetchpage, count
>>>
>>> fetchpage.async_pipe.__dict__
{'type': 'processor', 'name': 'fetchpage', 'sub_type': 'source'}
>>> count.pipe.__dict__
{'type': 'operator', 'name': 'count', 'sub_type': 'aggregator'}The SyncPipe and AsyncPipe classes (among other things) perform this
check for you to allow for convenient method chaining and transparent
parallelization.
>>> from riko.collections import SyncPipe
>>>
>>> attrs = [
... {'key': 'title', 'value': 'riko pt. 1'},
... {'key': 'content', 'value': "Let's talk about riko!"}]
>>> flow = SyncPipe('itembuilder', conf={'attrs': attrs}).hash()
>>> flow.list[0]
{'title': 'riko pt. 1',
'content': "Let's talk about riko!",
'hash': 1346301218}Please see the cookbook for advanced examples including how to wire in vales from other pipes or accept user input.
| [15] | the output stream of an aggregator is an iterator of only 1 item. |
riko provides a command, runpipe, to execute workflows. A
workflow is simply a file containing a function named pipe that creates
a flow and processes the resulting stream.
usage: runpipe [pipeid]
description: Runs a riko pipe
- positional arguments:
- pipeid The pipe to run (default: reads from stdin).
- optional arguments:
-h, --help show this help message and exit -a, --async Load async pipe. -t, --test Run in test mode (uses default inputs).
flow.py
from __future__ import print_function
from riko.collections import SyncPipe
conf1 = {'attrs': [{'value': 'https://google.com', 'key': 'content'}]}
conf2 = {'rule': [{'find': 'com', 'replace': 'co.uk'}]}
def pipe(test=False):
kwargs = {'conf': conf1, 'test': test}
flow = SyncPipe('itembuilder', **kwargs).strreplace(conf=conf2)
stream = flow.output
for i in stream:
print(i)Now to execute flow.py, type the command runpipe flow. You should
then see the following output in your terminal:
https://google.co.ukrunpipe will also search the examples directory for workflows. Type
runpipe demo and you should see the following output:
Deadline to clear up health law eligibility near 682riko comes with a built in task manager manage.
pip install riko[develop]Run python linter and nose tests
manage lint
manage testPlease mimic the coding style/conventions used in this repo. If you add new classes or functions, please add the appropriate doc blocks with examples. Also, make sure the python linter and nose tests pass.
Please see the contributing doc for more details.
Shoutout to pipe2py for heavily inspiring riko. riko started out as a fork
of pipe2py, but has since diverged so much that little (if any) of the original
code-base remains.
┌── benchmarks
│ ├── __init__.py
│ └── parallel.py
├── bin
│ └── run
├── data/*
├── docs
│ ├── AUTHORS.rst
│ ├── CHANGES.rst
│ ├── COOKBOOK.rst
│ ├── FAQ.rst
│ ├── INSTALLATION.rst
│ └── TODO.rst
├── examples/*
├── helpers/*
├── riko
│ ├── __init__.py
│ ├── lib
│ │ ├── __init__.py
│ │ ├── autorss.py
│ │ ├── collections.py
│ │ ├── dotdict.py
│ │ ├── log.py
│ │ ├── tags.py
│ │ └── py
│ ├── modules/*
│ └── twisted
│ ├── __init__.py
│ ├── collections.py
│ └── py
├── tests
│ ├── __init__.py
│ ├── standard.rc
│ └── test_examples.py
├── CONTRIBUTING.rst
├── dev-requirements.txt
├── LICENSE
├── Makefile
├── manage.py
├── MANIFEST.in
├── optional-requirements.txt
├── py2-requirements.txt
├── README.rst
├── requirements.txt
├── setup.cfg
├── setup.py
└── tox.iniriko is distributed under the MIT License.