Testing model outputs in dbt - Step-by-Step Execution

Concept Flow - Testing model outputs

Run dbt model

↓

Generate output table

↓

Run tests on output

↓

Test passes

→Continue

↓

Test fails

→Fix model or data

↓

Deploy or iterate

This flow shows how dbt runs a model, generates output, tests it, and either continues or fixes issues.

Execution Sample

dbt

select id, count(*) as cnt
from source_table
group by id

This SQL model groups data by id and counts rows per id.

Execution Table

Step	Action	Evaluation	Result
1	Run model SQL	Group by id and count rows	Output table with id and cnt columns
2	Run test: unique id	Check if all ids are unique	Pass if no duplicates
3	Run test: cnt > 0	Check if count is positive	Pass if all counts > 0
4	Test results	All tests pass	Model output is valid
5	End	No errors found	Ready for deployment

💡 All tests pass, so execution stops successfully.

Variable Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	Final
output_table	empty	table with id and cnt	same	same	same
test_unique_id	not run	not run	pass	pass	pass
test_cnt_positive	not run	not run	not run	pass	pass

Key Moments - 2 Insights

Why do we run tests after the model runs?

Tests check if the output data meets expectations, catching errors early as shown in steps 2 and 3 of the execution table.

What happens if a test fails?

If a test fails, the flow stops and you fix the model or data before continuing, preventing bad data from moving forward.

Visual Quiz - 3 Questions

Look at the execution table, what is the result of step 3?

A Test passes because all counts are positive

B Test fails due to negative counts

C Model output is empty

D Test not run yet

💡 Hint

Check the 'Result' column in row with Step 3 in execution_table

At which step do we verify that all ids are unique?

A Step 1

B Step 4

C Step 2

D Step 5

💡 Hint

Look at the 'Action' column in execution_table for the unique id test

If the test for unique ids failed, what would happen next?

A Continue to deploy model

B Fix model or data before continuing

C Ignore and run other tests

D Delete the model

💡 Hint

Refer to the concept_flow where test failure leads to fixing issues

Concept Snapshot

Testing model outputs in dbt:
- Run your SQL model to create output
- Use dbt tests to check data quality
- Tests include uniqueness, non-null, and value checks
- If tests pass, deploy or continue
- If tests fail, fix model or data before proceeding

Full Transcript

In dbt, after running a model that creates a table, we run tests to check the output data. These tests verify things like whether ids are unique and counts are positive. If all tests pass, the model output is valid and ready for deployment. If any test fails, we stop and fix the model or data. This process helps catch errors early and ensures data quality.

Practice

1. What is the main purpose of testing model outputs in dbt?

easy

A. To ensure the data is accurate and reliable

B. To speed up the data loading process

C. To create new tables automatically

D. To delete old data from the database

Solution

1. Step 1: Understand the goal of testing in dbt

   Testing checks if the data produced by models is correct and trustworthy.

2. Step 2: Identify the main benefit of testing outputs

   Accurate and reliable data helps users make good decisions and trust reports.

3. Final Answer:

   To ensure the data is accurate and reliable -> Option A

4. Quick Check:

   Testing = Accurate data [OK]

Hint: Testing checks data correctness, not speed or deletion [OK]

Common Mistakes:

• Thinking tests speed up loading
• Confusing testing with table creation
• Assuming tests delete data

2. Which of the following is the correct syntax to define a uniqueness test on a column user_id in a dbt model's schema.yml file?

easy

A. - model: users
columns:
- name: user_id
tests:
- unique

B. - name: users
columns:
- name: user_id
test: unique

C. - name: users
columns:
- user_id
tests:
- unique

D. - name: users
columns:
- name: user_id
tests:
- unique

Solution

1. Step 1: Check the correct key for model name

   The key to specify the model is name, not model.

2. Step 2: Verify column and test syntax

   Each column uses name and tests are listed under tests as a list.

3. Final Answer:

   - name: users columns: - name: user_id tests: - unique -> Option D

4. Quick Check:

   Correct schema.yml syntax = - name: users columns: - name: user_id tests: - unique [OK]

Hint: Use 'name' for model and column, 'tests' as list [OK]

Common Mistakes:

• Using 'model' instead of 'name' for model
• Writing 'test' instead of 'tests'
• Omitting 'name' for column

3. Given this dbt test defined in schema.yml for the model orders:

- name: orders
  columns:
    - name: order_id
      tests:
        - unique
        - not_null

What will happen if the orders table has two rows with the same order_id and one row with order_id as NULL when you run dbt test?

medium

A. The test will fail because of duplicate and NULL values in order_id

B. The test will pass because only one test can fail at a time

C. The test will fail only for duplicate values, NULLs are ignored

D. The test will pass because NULLs are allowed in unique tests

Solution

1. Step 1: Understand the tests applied

   The tests are unique and not_null on order_id.

2. Step 2: Analyze the data issues

   Two rows have the same order_id (violates uniqueness) and one row has NULL order_id (violates not_null).

3. Final Answer:

   The test will fail because of duplicate and NULL values in order_id -> Option A

4. Quick Check:

   Duplicates + NULLs = test fail [OK]

Hint: Both unique and not_null must pass for success [OK]

Common Mistakes:

• Assuming NULLs are allowed in unique tests
• Thinking only one test failure causes pass
• Ignoring NULL violation in not_null test

4. You wrote this test in your schema.yml file:

- name: customers
  columns:
    - name: email
      tests:
        - unique
        - not_null

But when you run dbt test, you get an error saying Invalid test configuration. What is the likely cause?

medium

A. The test name should be unique and not_null without dashes

B. The indentation of the tests list is incorrect

C. The model name should be under models key in schema.yml

D. The email column does not exist in the model

Solution

1. Step 1: Check the structure of schema.yml

   Tests must be defined under the models: key in schema.yml.

2. Step 2: Identify missing models: key

   The snippet misses the models: root key, causing invalid configuration.

3. Final Answer:

   The model name should be under models key in schema.yml -> Option C

4. Quick Check:

   Missing 'models:' key = config error [OK]

Hint: Always start schema.yml tests under 'models:' key [OK]

Common Mistakes:

• Removing dashes from test names
• Incorrect indentation of tests list
• Not placing model under 'models:'

5. You want to test that the status column in your transactions model only contains the values 'pending', 'completed', or 'failed'. Which test definition in schema.yml correctly enforces this?

hard

A. - name: transactions
columns:
- name: status
tests:
- values_in: ['pending', 'completed', 'failed']

B. - name: transactions
columns:
- name: status
tests:
- accepted_values:
values: ['pending', 'completed', 'failed']

C. - name: transactions
columns:
- name: status
tests:
- accepted_values: ['pending', 'completed', 'failed']

D. - name: transactions
columns:
- name: status
tests:
- unique
- not_null

Solution

1. Step 1: Identify the correct test for allowed values

   The accepted_values test checks if column values are in a list.

2. Step 2: Check correct syntax for accepted_values

   The test requires a dictionary with key values listing allowed values.

3. Final Answer:

   - name: transactions columns: - name: status tests: - accepted_values: values: ['pending', 'completed', 'failed'] -> Option B

4. Quick Check:

   accepted_values with 'values' key = - name: transactions columns: - name: status tests: - accepted_values: values: ['pending', 'completed', 'failed'] [OK]

Hint: Use accepted_values with 'values' list for allowed values [OK]

Common Mistakes:

• Using unique or not_null instead of accepted_values
• Omitting 'values:' key under accepted_values
• Using wrong test name like values_in