Docker & deployment — 60-second sheets¶

~7 min total · 6 concepts · Last-minute revision layer — full treatment: W2

Docker: EXPOSE vs -p¶

EXPOSE 8501        # documentation: "this app listens on 8501" — a LABEL

docker run -p 8501:8501 app   # THIS publishes the port to the host

The one-liner

EXPOSE is a label; -p is the bridge. If a container runs but localhost:8501 refuses to connect, you forgot -p.

Practice: Container runs, localhost:8501 doesn't connect. What's missing?

The -p 8501:8501 flag on docker run. → T1-AN Q28

Docker: the missing install¶

A container that can't find the requests library is missing:

RUN pip install -r requirements.txt

Containers are isolated — nothing from your laptop leaks in, including installed packages.

Practice: Why does the app work locally but crash in Docker?

The Dockerfile doesn't install dependencies. Containers share nothing with your laptop. → T1-FN Q11

Docker: environment parity¶

An Ubuntu container built anywhere runs identically on Windows, macOS, Linux — the OS is inside the image. "Works on my machine" dies here.

Practice: You build on Linux, push to Docker Hub, colleague pulls on Windows. What happens?

It runs successfully. Docker packages the entire runtime environment. → T1-FN Q5

Docker: layer caching¶

Docker rebuilds a layer only when its inputs change — and every layer after a changed layer rebuilds too.

COPY requirements.txt .              # ← dependency list FIRST
RUN pip install -r requirements.txt # ← install (stays cached)
COPY . .                            # ← code LAST

Trap — the build-speed killer

Copy the code before pip install and every code edit re-downloads every dependency. Copy requirements.txt first so the install layer stays cached.

Practice: Why does COPY . . before RUN pip install ruin builds?

Every code edit invalidates the COPY layer → the install layer after it rebuilds → dependencies re-download on every minor change. → T1-AN Q24 (short answer)

Serverless limits — RAM + time as a pair¶

Limit	The story	The fix
Time	Edge function needs 40s, dies at 10s	Long jobs → containers/queues
RAM	2GB file works on laptop, dies on function (128–512 MB)	Stream/chunk, don't load it all

The numbers

Your laptop: 8–32 GB RAM. Free-tier function: 128–512 MB. Edge timeout: ~10s.

Practice: A 2GB file crashes a serverless function but works locally. Why?

Free-tier functions enforce strict RAM limits (128–512 MB vs your laptop's 8–32 GB). → T1-FN Q10

GitHub Pages — static only¶

Serves HTML, CSS, images, client-side JS. No server-side runtime, no database, no FastAPI backend. Form submissions have nothing to receive them.

Practice: Can you host a FastAPI backend on GitHub Pages?

No — Pages serves only static files. The backend needs a real runtime. → T1-FN Q3