Opscraft
Docker · Lesson 6 of 11 · 3 min read

Storage and data

Three ways to mount data

Type Syntax Managed by Use for
Named volume -v pgdata:/var/lib/postgresql/data or --mount type=volume,src=pgdata,dst=... Docker (/var/lib/docker/volumes/) Persistent app data, databases
Bind mount -v /host/path:/container/path or --mount type=bind,... You (any host path) Dev hot-reload, config files, host sockets
tmpfs --tmpfs /tmp:size=64m Memory Scratch data, secrets that must not touch disk
Enlarge diagramHOSTcontainer/var/lib/postgresql/data/app/config/tmpnamed volumemanaged by Dockerbind mount/home/me/configtmpfsRAM, gone on stop
Figure 1. Three ways to give a container data. Volumes are managed by Docker; bind mounts expose a host folder; tmpfs lives in memory.
Swipe sideways to see all of it, or tap to open it full screen.
  • Volumes survive docker rm; docker rm -v or docker volume rm deletes them. docker volume prune removes unused ones (dangerous on shared hosts).
  • An empty named volume mounted over a directory that has content in the image is pre-populated with that content; a bind mount simply hides it.
  • Bind mounts expose host files with host UIDs — permission mismatches (container runs as UID 10001, host dir owned by 1000) cause "permission denied".
  • Add :ro for read-only mounts.
  • Back up a volume by running a throwaway container: docker run --rm -v pgdata:/data -v $PWD:/backup alpine tar czf /backup/pgdata.tgz -C /data .

The container's writable layer

Data written outside volumes lives in the container layer: lost when the container is removed, slower (copy-on-write), and it bloats /var/lib/docker. Production containers ideally run with a read-only root filesystem (--read-only plus tmpfs for /tmp) — it blocks attackers from writing tools into the container and proves the app has no hidden state.

Databases in containers

Fine for dev and CI. In production it's possible (volumes, backups, careful upgrades), but on AWS a managed service (RDS/Aurora) usually wins: automated backups, Multi-AZ failover, patching. In Kubernetes, stateful databases need StatefulSets, persistent volumes and ideally an operator.

Disk housekeeping

docker system df                     # space by images, containers, volumes, build cache
docker system prune                  # stopped containers, unused networks, dangling images, build cache
docker system prune -a --volumes     # aggressive: all unused images and volumes
docker builder prune --filter until=72h

/var/lib/docker filling up is one of the most common Docker host incidents. Put it on a separate volume, cap logs, prune on a schedule (carefully), and alert on disk usage.