How exactly am I supposed to manage volumes with podman / docker?

[enricojr01]

SO I have the following minimal compose file for a postgres instance with pgadmin for admin stuff:

services:
  main_postgres:
    image: pgrouting/pgrouting:17-3.5-main
    restart: always
    shm_size: 128mb

    ports:
      - "5432:5432"

    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: admin

  main_pgadmin:
    image: dpage/pgadmin4:9
    restart: always
    shm_size: 256mb

    ports:
      - "8180:80"

    environment:
      PGADMIN_DEFAULT_EMAIL: admin@admin.com
      PGADMIN_DEFAULT_PASSWORD: admin

    depends_on:
      - main_postgres

I added a "volume" section to the main_postgres entry (mistakenly) because I wanted to bind mount a local directory containing .csv files for import. It should have been to the main_pgadmin directory in hindsight, but that's beside the point. This appears to have erased all my data on both containers, because when I start things up with podman compose up -d pgadmin no longer has any of my saved servers / SQL tabs, and when I connect to postgres none of my tables are there.

The data loss isn't that concerning because I already have exports I can restore from but I need clarification on a few things so that this doesn't happen again:

1. I was under the impression that podman/docker will create and manage volumes for you if none are specified. Is this correct?
2. Given the above, I'm guessing that my adding the "volume" section caused the containers to rebuild somehow, discarding the data. Am I correct, or am I barking up the wrong tree here?
3. Am I correct in assuming that I should have made named volumes from the start, by adding the following to the compose file:

// obviously they'd have to be referenced in their respective sections too
volumes:
    postgres_data
    pgadmin_data

4. If I create named volumes like this, would I be able to copy files to them from Windows Explorer i.e. without using the podman cp command?

I'm normally not the devops guy, I'm just a regular dev that has to use docker sometimes, any help with this is much appreciated.

[de-lazurenko]

Great questions - let me go through each one.

1. Do Podman/Docker create volumes automatically if none are specified?Yes, but only if the image declares a VOLUME instruction in its Dockerfile. Both postgres and pgadmin4 do this - so out of the box, they get anonymous volumes (random hashes like a3f9b2c1...). These persist across podman compose down restarts, but are easy to accidentally orphan.

2. Did adding the volumes: section cause data loss?Most likely yes, but not because containers "rebuilt" - the image layers stay the same. What probably happened is that your new bind mount shadowed the anonymous volume that was previously holding your data. When Podman saw a new mount point for /var/lib/postgresql/data, it used that (empty local directory) instead of the old anonymous volume. The old anonymous volume still exists on your system - you can check with:

podman volume ls

You might be able to recover your data from there if the hash-named volume is still listed.

3. Should you have used named volumes from the start?

Absolutely - named volumes are the standard recommendation for exactly this reason. Your compose snippet is almost right, just missing the colons:

volumes:
  postgres_data:
  pgadmin_data:

And reference them in each service:

main_postgres:
  volumes:
    - postgres_data:/var/lib/postgresql/data

main_pgadmin:
  volumes:
    - pgadmin_data:/var/lib/pgadmin

Named volumes survive podman compose down, are easy to inspect by name, and won't get accidentally orphaned.

4. Can you copy CSV files to a named volume from Windows Explorer?
   Not directly - named volumes live inside Podman's storage and aren't exposed as a regular folder in Explorer. Your options are:

podman cp - simplest for one-off copies: podman cp myfile.csv main_postgres:/tmp/
Bind mount a local folder - the right approach for your CSV use case. Instead of a named volume for the import directory, just mount a Windows path directly:

  volumes:
    - C:/Users/you/csv_data:/tmp/imports

Then you can drop files into that folder from Explorer and access them inside the container at /tmp/imports. Use named volumes only for persistent state (database files, pgadmin config), and bind mounts for input data you manage yourself.
Hope that clears things up!

[enricojr01]

Bind mount a local folder - the right approach for your CSV use case. Instead of a named volume for the import directory, just mount a Windows path directly:

  volumes:
    - C:/Users/you/csv_data:/tmp/imports

So I take it you can have two entries in the volumes: section like so?

main_pgadmin:
    volumes:
        - pgadmin_data:/var/lib/pgadmin
        - C:/Files/csv_data:/var/lib/pgadmin/wherever

[de-lazurenko]

Yes, exactly - you can mix named volumes and bind mounts freely in the same volumes: section. Podman/Docker treats each entry independently.
One thing to flag in your example though: you're mounting the bind mount inside /var/lib/pgadmin/wherever, which is already covered by the named volume mounted at /var/lib/pgadmin. That works, but be aware of the layering - the named volume owns the parent directory, and the bind mount will shadow whatever is at that specific subdirectory path inside it.
For your CSV import use case, it's cleaner to mount somewhere neutral like /tmp/imports rather than nesting inside pgadmin's data directory:

main_pgadmin:
  volumes:
    - pgadmin_data:/var/lib/pgadmin        # named volume for pgadmin state
    - C:/Files/csv_data:/tmp/imports       # bind mount for your CSV files

This keeps concerns separated - pgadmin manages its own data, and you have a clean drop zone for files you control. Then inside the container you'd reference your CSVs at /tmp/imports/yourfile.csv.

[enricojr01]

Noted, thank you very much. This was super helpful.