Python docker image tend to become large after installing the application requirements (the slim base is ~150 MB uncompressed), therefore it's important to spend efforts to minimise the image size, even if it produces a slightly more complex multistage Dockerfile.
The implemented Dockerfile makes sure the production image will keep to a minimal size ("only" 360MB):
- 150MB base image
- 210MB python installed dependencies
If you look at the "dev" image is instead ~850MB, more than 400MB that would end up as a cost in traffic on each image pull.
flowchart TD
subgraph BASE
base["Base
====
Contains system runtime
libraries necessary to run
all the applications
(e.g. libmysql)"]
base_app["Base app
========
Copies shared application logic
independent from the used framework
(domains, storage, etc.)"]
base_builder["Base Builder
============
Contains system libraries
necessary to build python
dependencies
(e.g. gcc, library headers)"]
end
dev["Dev
===
Fat image containing everything
for local development"]
base-->base_app
base-->base_builder
base----->dev
subgraph HTTP
http_builder["HTTP builder
============
Installs requirements for
HTTP app in a virtualenv"]
http_app["HTTP app
========
Copies HTTP app,
shared logic
and requirements
from previous containers"]
http_builder-->http_app
end
subgraph Dramatiq
dramatiq_builder["Dramatiq builder
============
Installs requirements for
Dramatiq worker in a virtualenv"]
dramatiq_app["HTTP app
========
Copies Dramatiq worker app,
shared logic
and requirements
from previous containers"]
dramatiq_builder-->dramatiq_app
end
base_builder-->http_builder
base_builder-->dramatiq_builder
base_app-->http_app
base_app-->dramatiq_app