Skip to main content


Defer is a powerful feature that makes it possible to run a subset of models or tests in a sandbox environment without having to first build their upstream parents. This can save time and computational resources when you want to test a small number of models in a large project.

Defer requires that a manifest from a previous dbt invocation be passed to the --state flag or env var. Together with the state: selection method, these features enable "Slim CI". Read more about state.

An alternative command that accomplishes similar functionality for different use cases is dbt clone - see the docs for clone for more information.


dbt run --select [...] --defer --state path/to/artifacts
dbt test --select [...] --defer --state path/to/artifacts

When the --defer flag is provided, dbt will resolve ref calls differently depending on two criteria:

  1. Is the referenced node included in the model selection criteria of the current run?
  2. Does the reference node exist as a database object in the current environment?

If the answer to both is no—a node is not included and it does not exist as a database object in the current environment—references to it will use the other namespace instead, provided by the state manifest.

Ephemeral models are never deferred, since they serve as "passthroughs" for other ref calls.

When using defer, you may be selecting from production datasets, development datasets, or a mix of both. Note that this can yield unexpected results

  • if you apply env-specific limits in dev but not prod, as you may end up selecting more data than you expect
  • when executing tests that depend on multiple parents (e.g. relationships), since you're testing "across" environments


In my local development environment, I create all models in my target schema, dev_alice. In production, the same models are created in a schema named prod.

I access the dbt-generated artifacts (namely manifest.json) from a production run, and copy them into a local directory called prod-run-artifacts.


I've been working on model_b:



from {{ ref('model_a') }}
group by 1

I want to test my changes. Nothing exists in my development schema, dev_alice.


dbt run --select "model_b"
create or replace view dev_me.model_b as (



from dev_alice.model_a
group by 1


Unless I had previously run model_a into this development environment, dev_alice.model_a will not exist, thereby causing a database error.

I also have a relationships test that establishes referential integrity between model_a and model_b:

version: 2

- name: model_b
- name: id
- relationships:
to: ref('model_a')
field: id

(A bit silly, since all the data in model_b had to come from model_a, but suspend your disbelief.)

dbt test --select "model_b"
select count(*) as validation_errors
from (
select id as id from dev_alice.model_b
) as child
left join (
select id as id from dev_alice.model_a
) as parent on =
where is not null
and is null

The relationships test requires both model_a and model_b. Because I did not build model_a in my previous dbt run, dev_alice.model_a does not exist and this test query fails.