PostgreSQL

PostgreSQL is the awkward case: a tar of a running data directory can capture a half-written state. There are two correct answers, and Dockup supports both.

Option 1 — stop the database (simplest)

Postgres flushes cleanly on shutdown, so a stopped container's volume is consistent. This costs a few seconds of downtime per backup.

services:
  postgres:
    image: postgres:17
    volumes:
      - pgdata:/var/lib/postgresql/data
    labels:
      dockup.stop: "true"          # Dockup may stop me

volumes:
  pgdata:
    labels:
      dockup.enable: "true"
      dockup.schedule: daily
      dockup.target: nas
      dockup.keep-daily: "7"
      dockup.keep-monthly: "6"

Dockup stops the container, archives the volume, and starts it again — even when the archive fails. stop-timeout (default 30s) is the grace period before Docker kills it.

Option 2 — dump into the volume (no downtime)

Use a pre-hook to write a logical dump inside the volume, and a post-hook to remove it. The archive then contains a consistent dump instead of raw files.

# dockup.yml
hooks:
  postgres:
    pre:
      - pg_dump -U postgres -Fc -f /var/lib/postgresql/data/dockup.pgcustom app
    post:
      - rm -f /var/lib/postgresql/data/dockup.pgcustom

services:
  postgres:
    image: postgres:17
    volumes:
      - pgdata:/var/lib/postgresql/data
    labels:
      dockup.hook: postgres       # run the hook set above

volumes:
  pgdata:
    labels:
      dockup.enable: "true"
      dockup.schedule: daily
      dockup.target: nas,offsite
      dockup.exclude: dockup.pgcustom   # already inside the dump

Hooks run through sh -c inside the container, so they see the container's PATH and its PG* environment.

A failing pre-hook aborts the run

Dockup does not stop anything or archive anything when a pre command fails; the run is recorded as failed. That is deliberate: an aborted pg_dump should never produce a backup that looks successful.

Restoring a logical dump

Restore the volume with dockup restore, then load the dump:

$ docker compose exec postgres pg_restore -U postgres -d app --clean --if-exists /var/lib/postgresql/data/dockup.pgcustom

Choosing between them

	Stop	Dump
Downtime	seconds per run	none
Restore	whole cluster, exactly as it was	one database, portable across versions
Archive size	full data directory	dump only (usually smaller)
Works for	every database	databases pg_dump can read

Running both is also sensible: dockup.stop: "true" and dockup.hook: postgres gives a clean shutdown plus a portable dump in the same archive.