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
Services, Architecture and Channels
Search
Andrew Godwin
April 03, 2017
Programming
4
690
Services, Architecture and Channels
A talk I gave at DjangoCon Europe 2017 in Florence, Italy
Andrew Godwin
April 03, 2017
Tweet
Share
More Decks by Andrew Godwin
See All by Andrew Godwin
Reconciling Everything
andrewgodwin
1
330
Django Through The Years
andrewgodwin
0
220
Writing Maintainable Software At Scale
andrewgodwin
0
450
A Newcomer's Guide To Airflow's Architecture
andrewgodwin
0
370
Async, Python, and the Future
andrewgodwin
2
680
How To Break Django: With Async
andrewgodwin
1
740
Taking Django's ORM Async
andrewgodwin
0
740
The Long Road To Asynchrony
andrewgodwin
0
680
The Scientist & The Engineer
andrewgodwin
1
780
Other Decks in Programming
See All in Programming
Serena MCPのすすめ
wadakatu
4
900
CSC305 Lecture 04
javiergs
PRO
0
250
CSC305 Lecture 01
javiergs
PRO
1
400
ててべんす独演会〜Flowの全てを語ります〜
tbsten
1
220
CSC305 Lecture 02
javiergs
PRO
1
260
2分台で1500examples完走!爆速CIを支える環境構築術 - Kaigi on Rails 2025
falcon8823
3
3.3k
Pythonスレッドとは結局何なのか? CPython実装から見るNoGIL時代の変化
curekoshimizu
4
1.4k
タスクの特性や不確実性に応じた最適な作業スタイルの選択(ペアプロ・モブプロ・ソロプロ)と実践 / Optimal Work Style Selection: Pair, Mob, or Solo Programming.
honyanya
3
140
Swift Concurrency - 状態監視の罠
objectiveaudio
2
460
Your Perfect Project Setup for Angular @BASTA! 2025 in Mainz
manfredsteyer
PRO
0
130
そのpreloadは必要?見過ごされたpreloadが技術的負債として爆発した日
mugitti9
2
3k
Breaking Up with Big ViewModels — Without Breaking Your Architecture (droidcon Berlin 2025)
steliosf
PRO
1
330
Featured
See All Featured
Writing Fast Ruby
sferik
629
62k
What’s in a name? Adding method to the madness
productmarketing
PRO
23
3.7k
The Myth of the Modular Monolith - Day 2 Keynote - Rails World 2024
eileencodes
26
3.1k
Improving Core Web Vitals using Speculation Rules API
sergeychernyshev
19
1.2k
Building a Scalable Design System with Sketch
lauravandoore
462
33k
The World Runs on Bad Software
bkeepers
PRO
71
11k
実際に使うSQLの書き方 徹底解説 / pgcon21j-tutorial
soudai
PRO
188
55k
BBQ
matthewcrist
89
9.8k
A designer walks into a library…
pauljervisheath
209
24k
Scaling GitHub
holman
463
140k
We Have a Design System, Now What?
morganepeng
53
7.8k
Fashionably flexible responsive web design (full day workshop)
malarkey
407
66k
Transcript
Andrew Godwin @andrewgodwin
Andrew Godwin Hi, I'm Django core developer Senior Software Engineer
at Apparently now does software architecture
The Monolith
The Monolith Only one version of any dependency Everything can
import everything Deployed all at once No separation Side-effects from other code
None
Services Code split up by purpose/team Defined cross-boundary API Deployed
separately Can use different versions of dependencies Isolated from each other
Why services?
Easier to manage Smaller, self-contained problems
Independent Scaling And easier performance analysis
Faster Individual Deployment Less to deploy at once
Complex Interdependencies Harder to deploy & track bugs
Requires great communication Teams need calling contracts and APIs
More points of failure Not just one set of homogenous
servers
No more quick hacks Separation forces a level of code
design
Switching To Services Or: How I Learned To Stop Worrying
And Love The Monolith
Identify the "cut points" You might need to make some
Allocate inventory Calculate Price & Charge card Finalise order Make
Order Row &
Make Order Row Allocate inventory Calculate Price Charge card Finalise
inventory Finalise order
Make Order Row Allocate inventory Calculate Price Charge card Finalise
inventory Finalise order
Define APIs between services Behave like all other teams are
third-party
Separate Datastores & Servers Make them as separate as possible
Communication & Transport
Service 2 Service 3 Service 1
Service 2 Service 3 Service 1 Direct Communication (20 services?
190 connections!)
Service 2 Service 3 Service 1 Direct with discovery Orchestrator
Service 2 Service 3 Service 1 Centralised Routing Router
Service 2 Service 3 Service 1 Message Bus
Centralised Comms Tradeoffs Distributed Comms Single point of failure Nasty
partial failures
At-least-once delivery Tradeoffs At-most-once delivery Some messages duplicated Some messages
lost
First-In-First-Out Tradeoffs First-In-Last-Out Easily backlogged Wide range of latencies
Channels & ASGI
Channel Layer Interface Server Worker Server Process 1 ASGI ASGI
Asynchronous socket handling Synchronous Django project Interface Server Worker Server ASGI ASGI Worker Server ASGI Process 2 Process 3 Process 4
Service 2 Service 3 Service 1 Channel Layer
Service Client inventory.request response.aF53Vds21
At-most-once delivery ASGI's Tradeoffs You have to design for potential
loss Low-latency but non-persistent Good for protocols, bad for important task queues Capacity, Backpressure and FIFO Informs producers quickly about pileups in the queue
Top Service-Oriented Architecture Tips
Per-request "correlation IDs" Track a set of service calls through
the stack
Feature Flag message headers Bundle them in, don't have every
service query them
Source Of Truth Each data model has a service that
owns (& caches) it
Metrics. Metrics everywhere. Both performance and network health
Design for failure Don't assume two things will both succeed
DO NOT START OFF WITH SERVICES Write separate Python libraries
instead
Thanks. Andrew Godwin @andrewgodwin