example.yaml

# pgctl config. Copy this, change every value, point --config at it.
#
# Resolved from --config, else $PGCTL_CONFIG, else ./pgctl.yaml.
#
# No credentials live here, and none are needed: pgctl reads the object-storage
# credentials out of your running postgres container's own environment, which is
# how `docker exec` into it can already reach the bucket. This file is not
# sensitive.

targets:
  # The name is what you pass to --env.
  example-production:
    # The machine pgctl operates the database from. Reached over ssh unless
    # --local. It need not match the target name.
    host: db.example.com

    # Optional. Who to ssh in as. Omit and ssh uses your local user, as it would
    # on the command line. --ssh-user overrides this.
    ssh_user: deploy

    # Your live postgres container on that host. pgctl reaches barman by
    # exec-ing into it, and reads the object-storage credentials from its
    # environment.
    container: example-postgres

    # The image the throwaway restore/drill container boots. It needs
    # barman-cloud-* on its PATH, and its major version must match the backup's
    # -- the drill asserts PG_VERSION against it, so a mismatch fails every
    # drill.
    image: ghcr.io/example/postgres-barman:18

    db: example
    # NOT necessarily "postgres". If your cluster was initialized with
    # POSTGRES_USER=example then no "postgres" role exists, and assuming
    # otherwise is what fails every backup, silently.
    role: example

    # The barman server name: which set of backups in the bucket is this one.
    server: example-production

    bucket: s3://example-backups/api
    endpoint: https://s3.example.com

    # Both the prune window and the drill's staleness bound. A backup older than
    # this means backups have been broken for longer than you keep them, so the
    # drill refuses to pass on it. Must be greater than 0.
    retention_days: 30

    # What the drill asks the restored database to prove it holds real data.
    # Pick something that is never legitimately empty.
    smoke_sql: SELECT count(*) FROM public.users

    # Optional. Where PGDATA sits inside `image`, relative to its postgres
    # volume. Defaults to 18/docker. Postgres nests the data dir by major
    # version, so set this if your image is not 18.
    # pgdata_subpath: 17/docker

  # A second target: preview, kept for a week rather than a month.
  example-preview:
    host: preview.example.com
    container: example-postgres
    image: ghcr.io/example/postgres-barman:18
    db: example_preview
    role: example
    server: example-preview
    bucket: s3://example-backups-preview/api
    endpoint: https://s3.example.com
    retention_days: 7
    smoke_sql: SELECT count(*) FROM public.users