Debug and cleanup

What you will be able to do

• Use compose logs and exec to debug a service
• Read docker system df
• Remove leftover containers and know when not to prune volumes

Why this matters for DE and AI

Failed jobs fill disks with images and anonymous volumes. No space left on device on a GPU box is often leftover Docker layers, not the dataset. You already practiced df in Linux. This lesson is the Docker-shaped version — plus the rule: do not prune volumes unless you mean to wipe databases.

Concepts

Debug order

1. docker compose ps -a — is it running, restarting, or Exited (1)?
2. docker compose logs SERVICE — what did it print?
3. docker compose exec SERVICE sh — only if it is still up.
4. docker inspect — mounts and env when the YAML and the engine disagree.

Disk

docker system df

Shows images, containers, and volumes. Images grow every time you build without cleaning tags.

Cleanup ladder (small to large)

Command	What it removes
docker compose down	Project containers and default network
docker rm stopped containers	One name
docker image prune	Dangling images (no tag)
docker system prune	Stopped containers, unused networks, dangling images
docker system prune -a	Also unused tagged images
docker compose down -v / docker volume prune	Volumes — Postgres data, named volumes

docker rm -f NAME force-stops and removes one container. Fine for a named lab sleeper. Not a substitute for reading logs.

Practice on your machine

If the Compose database from the last lesson is down, start it:

cd ~/dcubes/docker-lab/compose-db
docker compose up -d
docker compose ps -a
docker compose logs --tail=30 db

What you should see: db Up (healthy) if the healthcheck passed. Logs include ready/accepting connections.

docker compose exec db pg_isready -U dcubes -d lab

What you should see: accepting connections

docker system df

Images will be the largest line after Postgres. Numbers differ. The habit is to look.

List volumes from this project:

docker volume ls | grep compose-db || docker volume ls

Compose prefixes volume names with the project directory (compose-db_pgdata or similar).

Stop the stack without wiping the database:

docker compose down
docker compose ps

Containers gone, volume still listed in docker volume ls.

Optional tidy of unused containers and dangling images only:

docker container prune -f
docker image prune -f

-f skips the yes/no prompt. We still did not pass --volumes.

Remove the named sleeper leftovers if any still exist from earlier lessons:

docker ps -a --filter name=dcubes-

If names appear, docker rm -f those names. Do not rm -f production containers; you have none on this laptop if you followed the module.

Common mistakes

• Pruning to fix a crash. Read logs. Prune does not explain connection refused.
• down -v out of habit. That resets Postgres. Use it before a demo when you want empty.
• Ignoring docker system df until the VM is full. Check after a week of pulling tags.

Next

Capstone: First Compose shift — load the movies CSV into Postgres with Compose.