To build on irakli's answer, here's an updated solution:

• use newer version 2 Docker Compose file
• separate volumes section
• extra settings deleted

docker-compose.yml

version: '2'services:  postgres9:    image: postgres:9.4    expose:      - 5432    volumes:      - data:/var/lib/postgresql/datavolumes:  data: {}

demo

Start Postgres database server:

$ docker-compose up

Show all tables in the database. In another terminal, talk to the container's Postgres:

$ docker exec -it $(docker-compose ps -q postgres9 ) psql -Upostgres -c '\z'

It'll show nothing, as the database is blank. Create a table:

$ docker exec -it $(docker-compose ps -q postgres9 ) psql -Upostgres -c 'create table beer()'

List the newly-created table:

$ docker exec -it $(docker-compose ps -q postgres9 ) psql -Upostgres -c '\z'                         Access privileges Schema |   Name    | Type  | Access privileges | Column access privileges --------+-----------+-------+-------------------+-------------------------- public | beer      | table |                   | 

Yay! We've now started a Postgres database using a shared storage volume, and stored some data in it. Next step is to check that the data actually sticks around after the server stops.

Now, kill the Postgres server container:

$ docker-compose stop

Start up the Postgres container again:

$ docker-compose up

We expect that the database server will re-use the storage, so our very important data is still there. Check:

$ docker exec -it $(docker-compose ps -q postgres9 ) psql -Upostgres -c '\z'                         Access privileges Schema |   Name    | Type  | Access privileges | Column access privileges --------+-----------+-------+-------------------+--------------------------public | beer      | table |                   | 

We've successfully used a new-style Docker Compose file to run a Postgres database using an external data volume, and checked that it keeps our data safe and sound.

storing data

First, make a backup, storing our data on the host:

$ docker exec -it $(docker-compose ps -q postgres9 ) pg_dump -Upostgres > backup.sql

Zap our data from the guest database:

$ docker exec -it $(docker-compose ps -q postgres9 ) psql -Upostgres -c 'drop table beer'

Restore our backup (stored on the host) into the Postgres container.

Note: use "exec -i", not "-it", otherwise you'll get a "input device is not a TTY" error.

$ docker exec -i $(docker-compose ps -q postgres9 ) psql -Upostgres < backup.sql

List the tables to verify the restore worked:

$ docker exec -it $(docker-compose ps -q postgres9 ) psql -Upostgres -c '\z'                         Access privilegesSchema |   Name    | Type  | Access privileges | Column access privileges --------+-----------+-------+-------------------+--------------------------public | beer      | table |                   | 

To sum up, we've verified that we can start a database, the data persists after a restart, and we can restore a backup into it from the host.

Thanks Tomasz!

It looks like the PostgreSQL image is having issues with mounted volumes. FWIW, it is probably more of a PostgreSQL issue than Dockers, but that doesn't matter because mounting disks is not a recommended way for persisting database files, anyway.

You should be creating data-only Docker containers, instead. Like this:

postgres9:  image: postgres:9.4  ports:    - 5432:5432  volumes_from:    - pg_data  environment:    POSTGRES_PASSWORD: postgres    POSTGRES_USER: postgres    PGDATA : /var/lib/postgresql/data/pgdatapg_data:  image: alpine:latest  volumes:    - /var/lib/postgresql/data/pgdata  command: "true"

which I tested and worked fine. You can read more about data-only containers here: Why Docker Data Containers (Volumes!) are Good

As for: how to import initial data, you can either:

1. docker cp, into the data-only container of the setup, or
2. Use an SQL dump of the data, instead of moving binary files around (which is what I would do).