Lock in $30 Savings on PRO—Offer Ends Soon! ⏳
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Containerless Django
Search
Peter Baumgartner
October 16, 2018
Programming
5
1.8k
Containerless Django
Deploying Django without Docker
Peter Baumgartner
October 16, 2018
Tweet
Share
More Decks by Peter Baumgartner
See All by Peter Baumgartner
High Performance Django at Ten
ipmb
0
41
Just Enough Ops for Developers
ipmb
0
270
Prepping Your Project for Production
ipmb
2
890
Django Deployments Done Right
ipmb
6
1.5k
High Performance Django: From runserver to Reddit hugs
ipmb
1
1.4k
Getting Started with Salt (PyCon 2014)
ipmb
6
1.3k
Monitoring Infrastructure with SaltStack
ipmb
16
10k
Getting Started with Salt
ipmb
11
1.7k
Other Decks in Programming
See All in Programming
堅牢なフロントエンドテスト基盤を構築するために行った取り組み
shogo4131
6
1.9k
新卒エンジニアのプルリクエスト with AI駆動
fukunaga2025
0
140
[堅牢.py #1] テストを書かない研究者に送る、最初にテストを書く実験コード入門 / Let's start your ML project by writing tests
shunk031
11
6.9k
Reactive Thinking with Signals and the new Resource API
manfredsteyer
PRO
0
160
CSC305 Lecture 17
javiergs
PRO
0
270
目的で駆動する、AI時代のアーキテクチャ設計 / purpose-driven-architecture
minodriven
11
3.9k
251126 TestState APIってなんだっけ?Step Functionsテストどう変わる?
east_takumi
0
300
WebRTC と Rust と8K 60fps
tnoho
2
1.9k
Media Capture and Streams: W3C仕様と現場での知見
nowaki28
0
130
関数実行の裏側では何が起きているのか?
minop1205
1
560
CloudNative Days Winter 2025: 一週間で作る低レイヤコンテナランタイム
ternbusty
7
1.9k
「文字列→日付」の落とし穴 〜Ruby Date.parseの意外な挙動〜
sg4k0
0
360
Featured
See All Featured
No one is an island. Learnings from fostering a developers community.
thoeni
21
3.5k
Building Better People: How to give real-time feedback that sticks.
wjessup
370
20k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
25
1.6k
Fireside Chat
paigeccino
41
3.7k
A designer walks into a library…
pauljervisheath
210
24k
The Web Performance Landscape in 2024 [PerfNow 2024]
tammyeverts
12
960
Java REST API Framework Comparison - PWX 2021
mraible
34
9k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
285
14k
Large-scale JavaScript Application Architecture
addyosmani
514
110k
Done Done
chrislema
186
16k
We Have a Design System, Now What?
morganepeng
54
7.9k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
52
5.7k
Transcript
DjangoCon US—San Diego Oct 2018 Deploying without Docker Containerless Django
Peter Baumgartner
Founder at Lincoln Loop—lincolnloop.com Former SysAdmin, DevOps for 8 years
Author of High Performance Django About Me
Docker is cool!
@ipmb | #djangocon Docker is cool! The “pipeline” Security Isolation
Dev/prod parity
Just bundle the entire OS
Some philosophy
“ —Mike Perham https://www.mikeperham.com/2016/02/09/kill-your-dependencies/ No code runs faster than no
code. No code has fewer bugs than no code. No code uses less memory than no code. No code is easier to understand than no code. “
None
@ipmb | #djangocon Docker Drawbacks Slow Extra abstractions More software,
more problems
None
How did we get here?
@ipmb | #djangocon Deployments sucked Dependencies would shift underneath you
Build tools and dev packages needed to be installed Multiple languages, multiple builds (Python & Node)
@ipmb | #djangocon The ideal deployment Download a binary Create
a configuration file Run it
@ipmb | #djangocon The ideal deployment /usr/local/bin/traefik \ --configFile=/etc/traefik/traefik.toml /usr/local/bin/telegraf
\ --config=/etc/telegraf/telegraf.conf /usr/sbin/nginx -c /etc/nginx/nginx.conf
@ipmb | #djangocon Python isn’t C or Go Requires a
VM Dynamic linking Packaging isn’t straightforward
None
Can we do better?
@ipmb | #djangocon We already are! Lock files via pipenv
or poetry Pre-compiled wheels (Pillow, psycopg2-binary, etc.) Still lots of holes - Assembling virtualenvs - Static files - Production webserver
@ipmb | #djangocon Prior art Private PyPI virtualenv-clone Platter dh-virtualenv
Pex
@ipmb | #djangocon ZIP applications? Part of Python since 2.6
PEP-441 improves support in 3.5 Create a ZIP archive of your project. Run it with Python. …but no mechanism for handling dependencies
None
@ipmb | #djangocon Enter shiv! A project from LinkedIn Zipapps
with dependencies A single artifact you can build → test → deploy ./myproject.pyz runserver
Django as a zipapp
Package your project with setup.py
@ipmb | #djangocon Include templates & static files Create a
MANIFEST.in graft your_project/collected_static graft your_project/templates
@ipmb | #djangocon Production webserver gunicorn + whitenoise ⭐ https://pypi.org/project/django-pyuwsgi/
@ipmb | #djangocon Build your zipapp
@ipmb | #djangocon Run your zipapp ./yourproject.pyz pyuwsgi --http=:8000
@ipmb | #djangocon Configuration Same zipapp, but different settings per
environment Options: - Multiple settings files and DJANGO_SETTINGS_MODULE - Environment variables - ⭐ https://pypi.org/project/goodconf/
@ipmb | #djangocon The zipapp pipeline Use CI (Travis, CircleCI,
Bitbucket, etc.) to: - Build - Test - Push Deploy = Download and run
None
What about security?
Systemd's got your back
@ipmb | #djangocon Systemd is awesome ProtectSystem=strict ProtectHome=true DynamicUser=true CapabilityBoundingSet=~CAP_SYS_ADMIN
AppArmorProfile=srv.yourproject.pyz ProtectKernelTunables=true ProtectControlGroups=true ProtectKernelModules=true PrivateDevices=true PrivateTmp=true SystemCallArchitectures=native
What about isolation?
@ipmb | #djangocon Isolation You still need Python installed globally
Easy to install multiple Pythons on one server Docker has better isolation, but do you need it?
@ipmb | #djangocon What about parity? Zipapp is the same
from CI to all deployed environments Use Docker to mimic deployment envionrment locally (or don't)
@ipmb | #djangocon Pros Simpler. No Docker on the server.
No registry. ~1M fewer lines of code to depend on. Smaller artifacts Faster deployments It's just Python
@ipmb | #djangocon Cons Not as isolated as true containers
Requires Python runtime on the server Python-specific Not cross-platform compatible (if you have packages with C extensions)
@ipmb | #djangocon Sweet spot for zipapps You are deploying
primarily Python services You have outgrown PaaS (Heroku, PythonAnywhere, Divio, etc.) You have fewer than 50 services to maintain
None
None
Thanks! Peter Baumgartner
[email protected]
@ipmb