Schema checks

Use a SodaCL schema check to validate column presence, absence, or position in a table, or the type of data column contains.

Use a schema check to validate the presence, absence or position of columns in a dataset, or to validate the type of data column contains.

checks for dim_product:
  - schema:
      name: Confirm that required columns are present
      warn:
        when required column missing: [weight_unit_measure_code, product_subcategory_key, made_up_column]
      fail:
        when required column missing:
          - product_key 
          - product_alternate_key
  - schema:
      warn:
        when forbidden column present: [credit_card]
        when wrong column type:
          standard_cost: money
      fail:
        when forbidden column present: [pii*]
        when wrong column type:
          reorder_point: smallint
  - schema:
      name: Columns out of order
      warn:
        when wrong column index:
          style: 1
      fail:
        when wrong column index:
          model_name: 22
  - schema:
      name: Any schema changes
      warn: 
        when schema changes: any

Define schema checks

✖️ Requires Soda Core Scientific (included in a Soda Runner) ✔️ Supported in Soda Core ✔️ Supported in Soda Library + Soda Cloud ✔️ Supported in Soda Cloud Agreements + Soda Runner ✔️ Available as a no-code check with a self-hosted Soda Runner connected to any Soda-supported data source, except Spark, and Dask and Pandas OR with a Soda-hosted Runner connected to a BigQuery, Databricks SQL, MS SQL Server, MySQL, PostgreSQL, Redshift, or Snowflake data source

In the context of SodaCL check types, schema checks are unique. Schema checks always employ alert configurations – specifying warn and/or fail alert conditions – with validation keys. Refer to Add alert configurations for exhaustive alert configuration details.

The validation key:value pairs in schema checks set the conditions for a warn or a fail check result. See a List of validation keys below.

For example, the following check uses the when required column missing validation key to validate that specific columns are present in a dataset; if any of columns in the list are absent, the check result is fail.

checks for dim_product:
  - schema:
      fail:
        when required column missing:
          - standard_cost
          - list_price
          - weight

In the example above, the value for the validation key is in a nested list format, but you can use an inline list of comma-separated values inside square brackets instead. The following example yields identical check results to the example above.

checks for dim_product:
  - schema:
      fail:
        when required column missing: [standard_cost, list_price, weight]

You can define a schema check with both warn and fail alert conditions, each with multiple validation keys. Refer to Configure multiple alerts for details. Be aware, however, that a single schema check only ever produces a single check result. See Expect one check result below for details.

The following example is a single check; Soda executes each of its validations during a scan. Note that unlike the nested list of column names in the example above, the nested key:value pairs that form the value for these validation keys are indented, but do not use a -.

checks for dim_product:
  - schema:
      warn:
        when forbidden column present: [standard_cost]
        when wrong column type:
          standard_cost: money
          weight: double precision
      fail:
        when forbidden column present: [sombrero]
        when wrong column type:
          reorder_point: smallint

Add a schema_name parameter to a schema check to address a situation in which you need to explicitly identify or override a dataset's schema in the data source.

checks for dim_employee:
   - schema:
      schema_name: staff.pr
      name: Required columns present
      warn:
        when required column missing: [last_name, birth_date]

Define schema evolution checks

✖️ Requires Soda Core Scientific (included in a Soda Runner) ✖️ Supported in Soda Core ✔️ Supported in Soda Library + Soda Cloud ✔️ Supported in Soda Cloud Agreements + Soda Runner ✔️ Available as a no-code check with a self-hosted Soda Runner connected to any Soda-supported data source, except Spark, and Dask and Pandas ✔️ Available as a no-code check with a Soda-hosted Runner connected to a BigQuery, Databricks SQL, MS SQL Server, MySQL, PostgreSQL, Redshift, or Snowflake data source

Rather than specifying exact parameters for column changes, you can use the when schema changes validation key to warn or fail when indistinct changes occur in a dataset.

Soda Cloud must have at least two measurements to yield a check result. In other words, the first time you run a scan to execute a schema evolution check, Soda returns no results because it has nothing against which to compare; the second scan that executes the check yields a check result.

checks for dim_customer:
  - schema:
      warn:
        when schema changes: any
      fail:
        when schema changes: 
         - column delete
         - column add
         - column index change
         - column type change

The output in Soda Cloud displays the output of all the alert states during the scan.

Optional check configurations

Supported	Configuration	Documentation
✓	Define a name for a schema check; see example.	Customize check names
✓	Add an identity to a check.	Add a check identity
✓	Define alert configurations to specify warn and fail alert conditions; see example.	Add alert configurations
	Apply an in-check filter to return results for a specific portion of the data in your dataset.	-
✓	Use quotes when identifying dataset or column names; see example. Note that the type of quotes you use must match that which your data source uses. For example, BigQuery uses a backtick (`) as a quotation mark.	Use quotes in a check
✓	Use wildcard characters ( % or * ) in values in the check; see example.	See note in example below.
✓	Use for each to apply schema checks to multiple datasets in one scan; see example.	Apply checks to multiple datasets
✓	Apply a dataset filter to partition data during a scan; see example.	Scan a portion of your dataset

Example with check name

checks for dim_product:
  - schema:
      name: Confirm that required columns are present
      warn:
        when required column missing: [weight_unit_measure_code, product_subcategory_key]

Example with alert configuration

checks for dim_product:
  - schema:
      warn:
        when forbidden column present: [standard_cost]

Example with quotes

checks for dim_product:
  - schema:
      warn:
        when wrong column type:
          standard_cost: "money"

Example with wildcards

You can use * or % as wildcard characters in a list of column names. If the column name begins with a wildcard character, add single quotes as per the example below.

checks for dim_product:
  - schema:
      fail:
        when forbidden column present:
          - credit_card
          - obsolete_%
          - '%SALARY%'
          - pii*

Example with for each

for each dataset T:
  datasets:
    - dim_product_%
  checks:
    - schema:
       warn:
         when schema changes: any

Example with dataset filter

filter CUSTOMERS [daily]:
  where: TIMESTAMP '{ts_start}' <= "ts" AND "ts" < TIMESTAMP '${ts_end}'

checks for CUSTOMERS [daily]:
  - schema:
      fail:
        when forbidden column present:
          - credit_card

List of validation keys

Validation key	Values
when required column missing	one or more column names in an inline list of comma-separated values, or a nested list
when forbidden column present	one or more column names in an inline list of comma-separated values, or a nested list
when wrong column type	nested key:value pair to identify column:expected_data_type
when wrong column index	nested key:value pair to identify column:expected_position_in_dataset_index
when schema changes	any as an inline value column add as a nested list item column delete as a nested list item column index change as a nested list item column type change as a nested list item

Expect one check result

Be aware that a check that contains one or more alert configurations only ever yields a single check result; one check yields one check result. If your check triggers both a warn and a fail, the check result only displays the more severe, failed check result.

Using the following example, Soda Library, during a scan, discovers that the data in the dataset triggers both alerts, but the check result at the bottom is Oops! 1 failures. Nonetheless, the results in the Scan summary section of the CLI output still display both the warn and fail alerts as having been triggered.

checks for dim_product:
  - schema:
      name: Required columns all present
      warn:
        when required column missing: [weight_unit_measure_code, product_subcategory_key, made_up_column]
      fail:
        when required column missing: [pretend_column]

Soda Library 1.0.x
Soda Core 3.0.x
Scan summary:
1/1 check FAILED: 
    dim_product in adventureworks
      Required columns all present [FAILED]
        fail_missing_column_names = [pretend_column]
        warn_missing_column_names = [made_up_column]
        schema_measured = [product_key integer, product_alternate_key character varying ...]
Oops! 1 failures. 0 warnings. 0 errors. 0 pass.
Sending results to Soda Cloud
Soda Cloud Trace: 7845***

Example: Detect PII

To address a common use case, you may wish to use a schema check in combination with a for each configuration and wildcard characters to automatically detect columns that contain personally identifiable information (PII) in your datasets, as in the following example.

for each dataset R:
  tables:
    # Apply the check to any dataset that begins with retail.
    - retail%
  checks:
    - schema:
        fail:
          when forbidden column present: ['*name*', '*address*', '*phone*', '*email*']

Go further

• Learn more about SodaCL metrics and checks in general.

• Use a reference check to validate matching contents between datasets.

• Reference tips and best practices for SodaCL.

Need help? Join the Soda community on Slack.