[Bug] on-start-run executes during compilation

[tekumara]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

on-start-run executes macros during dbt docs and dbt compile even when inside an if execute conditional.

Expected Behavior

dbt docs and dbt compile don't run macro code inside an if execute conditional

Steps To Reproduce

macros/run_start.sql

{% macro run_start() %}
  {% if execute %}
  {{ dbt_utils.log_info("🏁 \033[1m Let's go! 🏁 \033[0m")}}
  {% set results = run_query('select 1 from foobar') %}
  {% endif %}
{% endmacro %}

dbt_project.yml

on-run-start: "{{ run_start() }}"

Relevant log output

Running with dbt=0.21.1
Found 9 models, 7 tests, 0 snapshots, 0 analyses, 572 macros, 2 operations, 8 seed files, 4 sources, 0 exposures

12:24:46 | Concurrency: 4 threads (target='dev')
12:24:46 | 
12:24:46 + 🏁  Let's go! 🏁 
Encountered an error:
Runtime Error
  Database Error in operation jaffleshop-on-run-start-0 (./dbt_project.yml)
    002003 (42S02): SQL compilation error:
    Object 'FOOBAR' does not exist or not authorized.

Environment

- OS: darwin
- Python: 3.9
- dbt: 0.21.0

What database are you using dbt with?

snowflake

Additional Context

No response

[jtcohen6]

@tekumara Thanks for opening!

The compile task (and thereby the docs generate task, too) compiles SQL, including macros + hooks, in execute mode. The distinction between run and compile is subtle: compile is a subset of run functionality; model SQL is fully compiled for both (including execute-time introspective queries); the only step excluded in compile is the materialization logic itself.

We do need to update the docs for on-run-* hooks to reflect this reality. I just opened a PR to do that: dbt-labs/docs.getdbt.com#948

The good news is, in v1.0, there is now a context variable available from the flags module (docs), flags.WHICH, which will return the name of the currently running task/command. So your macro code could look like:

{% macro run_start() %}
  {% if execute and flags.WHICH in ('run', 'build') %}
  {{ dbt_utils.log_info("🏁 \033[1m Let's go! 🏁 \033[0m")}}
  {% set results = run_query('select 1 from foobar') %}
  {% endif %}
{% endmacro %}

Be advised, you should not use any values of flags in code needed at parse time (dependencies + configurations).

[jtcohen6]

Since there's a viable workaround in v1.0, I'm going to close this as a wontfix

[tekumara]

Thanks for the explanation and workaround @jtcohen6!

If the on-run-* hooks run for most everything, and not specifically run, maybe there's a better name for them?

[jtcohen6]

That's fair! on-task-start? on-execute-start? There are other places where we use the word run to refer to any task / execution, such as run_results.json

[tekumara]

Or how about on-dbt-start perhaps?