docker-compose.postgres.yml

# Runs the stack on PostgreSQL instead of SQL Server.
#
#   docker compose -f docker-compose.yml -f docker-compose.postgres.yml up -d
#
# Both are first-class: the same model, the same tests, the same CI checks. What differs is what
# the two databases genuinely do not share, and each of those is named in ProviderSpecificMapping
# rather than left to a convention - the filtered index that keeps one plan active, and the
# concurrency token, which PostgreSQL carries on its own `xmin` system column instead of a
# SQL Server `rowversion`.
#
# Migrations live in their own assembly per provider, because a migration is generated SQL rather
# than a description of intent: `nvarchar(200)` and `character varying(200)` cannot share a file.
# The application picks the assembly from Database__Provider at startup.
#
# Verified as far as this environment allows, and no further. The model, the mapping differences
# and the generated SQL are covered by tests and by CI on both providers; `docker compose config`
# renders this merge. What has NOT happened is either database actually running - the registry is
# filtered here, so no image was pulled and no migration was ever applied.
services:
  core:
    environment:
      Database__Provider: postgresql
      ConnectionStrings__Default: "Host=postgres;Database=echelon;Username=${POSTGRES_USER:-postgres};Password=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required (see .env.example)}"
      ConnectionStrings__Archive: "Host=postgres;Database=echelon_archive;Username=${POSTGRES_USER:-postgres};Password=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required (see .env.example)}"

    # Merged, not !override: overriding the whole map would drop what another override file put
    # there, and the two are meant to combine - postgres.yml plus single-instance.yml is a valid
    # deployment. `required: false` is what removes a dependency without erasing the map: the
    # profiled-out service is not started, and Compose warns instead of refusing the project.
    depends_on:
      mssql:
        condition: service_healthy
        required: false
      postgres:
        condition: service_healthy

  postgres:
    image: postgres:17-alpine
    environment:
      POSTGRES_USER: "${POSTGRES_USER:-postgres}"
      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required (see .env.example)}"
      # The archive is a second database, so it cannot be POSTGRES_DB - that creates only one.
      # Both are created by the init script below.
      POSTGRES_DB: echelon
    volumes:
      - pgdata:/var/lib/postgresql/data
      - ./scripts/postgres-init.sql:/docker-entrypoint-initdb.d/10-archive.sql:ro
    healthcheck:
      # -U, or pg_isready checks the OS user and reports ready before the role exists.
      test: ["CMD-SHELL", "pg_isready -U \"$$POSTGRES_USER\" -d echelon"]
      interval: 5s
      timeout: 3s
      retries: 10
      start_period: 10s
    deploy:
      resources:
        limits:
          cpus: "4"
          memory: 4G
    restart: unless-stopped

  # Kept in the model rather than deleted, so `docker compose config` still describes the full
  # stack and switching back is dropping this file.
  mssql:
    profiles: ["never"]

volumes:
  pgdata: