Allow import-db within stages

[fabiannagel]

I'm missing the possibility to run import-db as part of my pipeline. Consider the following scenario:

stages:
  
  convert:
    cmd: python convert.py
    deps:
      - data/raw_input_data
    outs:
      - data/converted_input_data

  ingest:
    cmd: python ingest.py && dvc import-db --table ingested_data --conn pgsql -o data/database
    deps:
      - data/coverted_input_data
    outs:
      - data/database

  run:
    cmd: python app.py
    deps:
      - data/database

ingest consumes my converted input data, applies some transformations and populates the application database. The state of the database is persisted via import-db, which is the data dependency for running the application.

Right now, dvc repro throws the following error with this config:

Running stage 'ingest':
> dvc import-db --table ingested_data --conn pgsql -o data/database
ERROR: output 'data/database' is already specified in stage: 'ingest'.
Use `dvc remove ingest` to stop tracking the overlapping output.
ERROR: failed to reproduce 'ingest': failed to run: dvc import-db --table ingested_data --conn pgsql -o data/database, exited with 255

It would be great to have a flag telling dvc import-db that it is part of a pipeline such that overlapping outputs are not an issue.

[shcheklein]

Make sense to expand pipeline stages to be DB import (or other imports?), wdyt @skshetry ?

For now I would recommend to run the query directly via Python script. You can take a look into DbDependency implementation and get some SQL wrapper code from it (it should not be very complicated I think).