Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Cooperative Multitasking with Twisted: Getting ...
Search
David Reid
February 01, 2012
Programming
1
800
Cooperative Multitasking with Twisted: Getting Things Done Concurrently
David Reid
February 01, 2012
Tweet
Share
Other Decks in Programming
See All in Programming
AIによるイベントストーミング図からのコード生成 / AI-powered code generation from Event Storming diagrams
nrslib
2
1.3k
.NET Conf 2025 の興味のあるセッ ションを復習した / dotnet conf 2025 quick recap for backend engineer
tomohisa
0
110
これならできる!個人開発のすゝめ
tinykitten
PRO
0
150
16年目のピクシブ百科事典を支える最新の技術基盤 / The Modern Tech Stack Powering Pixiv Encyclopedia in its 16th Year
ahuglajbclajep
4
730
生成AIを利用するだけでなく、投資できる組織へ
pospome
2
440
TerraformとStrands AgentsでAmazon Bedrock AgentCoreのSSO認証付きエージェントを量産しよう!
neruneruo
4
2.4k
Python札幌 LT資料
t3tra
7
1.1k
Implementation Patterns
denyspoltorak
0
150
MDN Web Docs に日本語翻訳でコントリビュート
ohmori_yusuke
0
410
The Past, Present, and Future of Enterprise Java
ivargrimstad
0
700
まだ間に合う!Claude Code元年をふりかえる
nogu66
5
930
Go コードベースの構成と AI コンテキスト定義
andpad
0
160
Featured
See All Featured
Game over? The fight for quality and originality in the time of robots
wayneb77
1
78
How to make the Groovebox
asonas
2
1.9k
First, design no harm
axbom
PRO
2
1.1k
What the history of the web can teach us about the future of AI
inesmontani
PRO
0
400
KATA
mclloyd
PRO
33
15k
YesSQL, Process and Tooling at Scale
rocio
174
15k
The #1 spot is gone: here's how to win anyway
tamaranovitovic
1
890
The untapped power of vector embeddings
frankvandijk
1
1.5k
brightonSEO & MeasureFest 2025 - Christian Goodrich - Winning strategies for Black Friday CRO & PPC
cargoodrich
2
79
State of Search Keynote: SEO is Dead Long Live SEO
ryanjones
0
83
A designer walks into a library…
pauljervisheath
210
24k
The AI Search Optimization Roadmap by Aleyda Solis
aleyda
1
5.1k
Transcript
Cooperative Multitasking with Twisted “Getting Things Done Concurrently” David Reid
<
[email protected]
>
Alternate Title
Alternate Title How I learned to stop worrying (about threads)
and use timed events to yield to the reactor so that synchronous tasks can cooperate with each other in a single thread.
Cooperative Multitasking
Cooperative Multitasking
Cooperative Multitasking
Cooperative Multitasking
Cooperative Multitasking
Cooperative Multitasking
Things will block
Fine in moderation
So what blocks?
Non-twisted networking
Non-twisted networking urllib & httplib smtplib python-memcached
Non-twisted networking twisted.web.client twisted.mail.smtp twisted.protocols.memcache
C Extensions
C Extensions DB-APIs: psycopg2, sqlite Hard math: PyCrypto socket.gethostbyname
C Extensions twisted.enterprise deferToThread deferToThread/twisted.names
File I/O
File I/O Yeah ... pretty much all of it.
The for loop.
def do_stuff(data): for x in data: do_work(x)
def do_stuff(data): for x in data: do_work(x)
Next: Introducing Twisted in 3 slides.
while true: stuff = wait_for_stuff(timeout) if stuff: check_reads_and_writes() run_timed_events() The
Reactor: Approximately
while true: stuff = wait_for_stuff(timeout) if stuff: check_reads_and_writes() run_timed_events() The
Reactor: Approximately
while true: stuff = wait_for_stuff(timeout) if stuff: check_reads_and_writes() run_timed_events() The
Reactor: Approximately
while true: stuff = wait_for_stuff(timeout) if stuff: check_reads_and_writes() run_timed_events() The
Reactor: Approximately
Deferreds Data Source result or failure Deferred
Deferreds Data Source result or failure Deferred res
Deferreds Data Source result or failure Deferred err
Deferreds Data Source result or failure Deferred
Deferreds Data Source result or failure Deferred err
Deferreds Data Source result or failure Deferred res
reactor.callLater
reactor.callLater reactor.callLater(10, do_stuff)
reactor.callLater reactor.callLater(0.001, do_stuff)
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
def do_stuff(data): d = Deferred() def do_later(): do_work(data.pop()) if not
data: d.callback(True) else: reactor.callLater(0.1, do_later) reactor.callLater(0.1, do_later) return d
twisted.internet.task.Cooperator
twisted.internet.task.Cooperator twisted.internet.task.coiterate
twisted.internet.task.Cooperator twisted.internet.task.coiterate twisted.internet.task.cooperate
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return coiterate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return coiterate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return coiterate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return coiterate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return coiterate(doterator())
• Similar to coiterate • Returns a CooperativeTask instead of
a Deferred cooperate
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator())
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
def do_stuff(data): def doterator(): for x in data: do_work(x) yield
None return cooperate(doterator()) task = do_stuff(someData) task_d = task.whenDone() task.pause() task.resume() task.stop()
• Your code is going to block. • Iterating over
stuff is a big place where synchronous code blocks for unacceptable periods of time. • Iterating is easily broken up by timed events. • Twisted comes with APIs for making this easy. • These APIs are the building blocks for your ideal task system.
Questions (Mine, not yours.)
Why?
Why? Being able to do more stuff is sometimes more
important than doing stuff quickly. * * Making a single user’s request 10ms faster is great, but not if you can only handle 1 request at a time.
Why call later instead of next iteration?
Why call later instead of next iteration? Timed events are
more flexible and you need them anyway.
Can’t I just use deferToThread for python code?
Can’t I just use deferToThread for python code? Yes, but
the GIL hates you.
Can’t I just use deferToThread for python code? Yes, but
the GIL hates you. It’s OK, it hates me too.
Can’t I just run code in a separate processes?
Can’t I just run code in a separate processes? Yes,
look at Ampoule.
Can’t I just run code in a separate processes? Yes,
look at Ampoule. But you probably want those processes to do more than one thing too.
This isn’t appropriate for my situation at all.
This isn’t appropriate for my situation at all. So don’t
do it.
This isn’t appropriate for my situation at all. So don’t
do it. Seriously.
SERIOUSLY.
• Twisted (http://twistedmatrix.com) • Ampoule (https://launchpad.net/ampoule) • txRDQ (https://launchpad.net/txrdq) •
Mochi Media (WE’RE HIRING) • http://mochimedia.com • http://bit.ly/mochi-jobs References