Command which 'run's, then 'test's (in the same command).

[doctoryes]

Describe the feature

On a large DBT project with many sources/models, DBT startup time affects development velocity when iterating on models. A common development pattern is to dbt run -m <mymodel>, immediately followed by (if the run succeeds) dbt test -m <mymodel>. A dbt run_then_test command (with a better name/design?) would be nice, which - in the context of a single DBT command - both runs and tests the model, saving the startup time of the dbt test.

Describe alternatives you've considered

I'm unaware of any alternatives here. This sort of functionality would need to be within DBT itself.

Additional context

Although a similar issue exists here:
#1054
this issue is different. I'm not asking for the results of the dbt run to be rolled-back if dbt test fails, which is typically unnecessary while iterating in development.

Who will this benefit?

This feature would aid any developers who are iterating on models for a large-scale DBT project that has significant DBT startup time overhead (~30-60 seconds).

[drewbanin]

Thanks for opening @doctoryes! We're still thinking about the exact mechanism we should employ here, but I very much buy this use-case and it's one that I'm keen to support!

[drewbanin]

To implement this feature, let's add a flag --test to the dbt run subcommand. If the --test flag is provided, then dbt should run tests for the selected models in DAG order during the run.

Notes:

• Only tests whose entire set of parents are selected in the run should be included. ie: this should not select a relationships test with one parent model selected and one parent model unselected.
  • When this happens, we should show a warning to the user that the test is not selected (see Node selection for @ on dbt test is incomplete #1827)
• If a test fails with error-level severity, then all of the dependents of the failed model should be skipped. If a test has more than one parent, then the children models of all ref'd parents should be skipped.
• If a test fails with warning-level severity, then the run should proceed, but note that the --warn-error flag can elevate test warnings to failures. Note: The --warn-error flag should also be included as a flag to dbt run for the purposes of this feature

[automagic]

Would it make sense to also specify --data so that we can run both schema and data tests inline? i.e --test --data or --data --test independently as part of the run command?

[jtcohen6]

That's a great question, which I would take a step further: how should we be passing task-specific flags to a generalized command? As I noted in #2743, this is even trickier for same-named flags that do subtly different things depending on the task type.

[fabrice-etanchaud]

Hello ! Having a little time to get back to dbt, I read only today that thread, and would like to suggest this link to our discourse design by contrat. this could help clarify (imho) the "build" idea. I tried to figure out how to add to dbt not only code testing, but mainly correctness validation.

[jtcohen6]

This was partially resolved by the addition of dbt build in #3490. There's one more piece here which is still TK: failing tests blocking downstream models from running. Closing in favor of #3597