Skip to content

Latest commit

 

History

History
194 lines (146 loc) · 13.6 KB

README.md

File metadata and controls

194 lines (146 loc) · 13.6 KB
Raw

dbt Constraints Package

This package generates database constraints based on the tests in a dbt project. It is currently compatible with Snowflake, PostgreSQL, Oracle, Redshift, and Vertica only.

How the dbt Constraints Package differs from dbt's Model Contracts feature

This package focuses on automatically generating constraints based on the tests already in a user's dbt project. In most cases, merely adding the dbt Constraints package is all that is needed to generate constraints. dbt's recent model contracts feature allows users to explicitly document constraints for models in yml. This package and the core feature are 100% compatible with one another and the dbt Constraints package will skip generating constraints already created by a model contract. However, the dbt Constraints package will also generate constraints for any tests that are not documented as model contracts. As described in the next section, dbt Constraints is also designed to provide join elimination on Snowflake.

Why data engineers should add referential integrity constraints

The primary reason to add constraints to your database tables is that many tools including DBeaver and Oracle SQL Developer Data Modeler can correctly reverse-engineer data model diagrams if there are primary keys, unique keys, and foreign keys on tables. Most BI tools will also add joins automatically between tables when you import tables that have foreign keys. This can both save time and avoid mistakes.

In addition, although Snowflake doesn't enforce most constraints, the query optimizer can consider primary key, unique key, and foreign key constraints during query rewrite if the constraint is set to RELY. Since dbt can test that the data in the table complies with the constraints, this package creates constraints on Snowflake with the RELY property to improve query performance. Some database query optimizers also consider not null constraints when building an execution plan.

Many databases including Snowflake, PostgreSQL, Oracle, SQL Server, MySQL, and DB2 can use referential integrity constraints to perform "Join Elimination" to remove tables from an execution plan. This commonly occurs when you query a subset of columns from a view and some of the tables in the view are unnecessary. In addition, on databases that do not support join elimination, some BI and visualization tools will also rewrite their queries based on constraint information, producing the same effect.

Finally, although most columnar databases including Snowflake do not use or need indexes, most row-oriented databases including PostgreSQL and Oracle require indexes on their primary key columns in order to perform efficient joins between tables. A primary key or unique key constraint is typically enforced on databases using such indexes. Having dbt create the unique indexes automatically can slightly reduce the degree of performance tuning necessary for row-oriented databases. Row-oriented databases frequently also need indexes on foreign key columns but that is something best added manually.

Please note

When you add this package, dbt will automatically begin to create unique keys for all your existing unique and dbt_utils.unique_combination_of_columns tests, foreign keys for existing relationship tests, and not null constraints for not_null tests. The package also provides three new tests (primary_key, unique_key, and foreign_key) that are a bit more flexible than the standard dbt tests. These tests can be used inline, out-of-line, and can support multiple columns when used in the tests: section of a model. The primary_key test will also cause a not null constraint to be created on each column.

Disabling automatic constraint generation

The dbt_constraints_enabled variable can be set to false in your project to disable automatic constraint generation. By default dbt Constraints only creates constraints on models. To allow constraints on sources, you can set dbt_constraints_sources_enabled to true. The package will verify that you have sufficient database privileges to create constraints on sources.

vars:
  # The package can be temporarily disabled using this variable
  dbt_constraints_enabled: true

  # The package can also add constraints on sources if you have sufficient privileges
  dbt_constraints_sources_enabled: false

  # You can also be specific on which constraints are enabled for sources
  # You must also enable dbt_constraints_sources_enabled above
  dbt_constraints_sources_pk_enabled: true
  dbt_constraints_sources_uk_enabled: true
  dbt_constraints_sources_fk_enabled: true
  dbt_constraints_sources_nn_enabled: true

Installation

  1. Add this package to your packages.yml following these instructions. Please check this link for the latest released version.
packages:
  - package: Snowflake-Labs/dbt_constraints
    version: [">=0.6.0", "<0.7.0"]
# <see https://github.com/Snowflake-Labs/dbt_constraints/releases/latest> for the latest version tag.
# You can also pull the latest changes from Github with the following:
#  - git: "https://github.com/Snowflake-Labs/dbt_constraints.git"
#    revision: main

  1. Run dbt deps.

  2. Optionally add primary_key, unique_key, or foreign_key tests to your model like the following examples.

  - name: DIM_ORDER_LINES
    columns:
      # Single column inline constraints
      - name: OL_PK
        tests:
          - dbt_constraints.primary_key
      - name: OL_UK
        tests:
          - dbt_constraints.unique_key
      - name: OL_CUSTKEY
        tests:
          - dbt_constraints.foreign_key:
              pk_table_name: ref('DIM_CUSTOMERS')
              pk_column_name: C_CUSTKEY
    tests:
      # Single column constraints
      - dbt_constraints.primary_key:
          column_name: OL_PK
      - dbt_constraints.unique_key:
          column_name: OL_ORDERKEY
      - dbt_constraints.foreign_key:
          fk_column_name: OL_CUSTKEY
          pk_table_name: ref('DIM_CUSTOMERS')
          pk_column_name: C_CUSTKEY
      # Multiple column constraints
      - dbt_constraints.primary_key:
          column_names:
            - OL_PK_COLUMN_1
            - OL_PK_COLUMN_2
      - dbt_constraints.unique_key:
          column_names:
            - OL_UK_COLUMN_1
            - OL_UK_COLUMN_2
      - dbt_constraints.foreign_key:
          fk_column_names:
            - OL_FK_COLUMN_1
            - OL_FK_COLUMN_2
          pk_table_name: ref('DIM_CUSTOMERS')
          pk_column_names:
            - C_PK_COLUMN_1
            - C_PK_COLUMN_2

Dependencies and Requirements

  • The package's macros depend on the results and graph object schemas of dbt >=1.0.0

  • The package currently only includes macros for creating constraints in Snowflake, PostgreSQL, and Oracle. To add support for other databases, it is necessary to implement the following seven macros with the appropriate DDL & SQL for your database. Pull requests to contribute support for other databases are welcome. See the <ADAPTER_NAME>__create_constraints.sql files as examples.

<ADAPTER_NAME>__create_primary_key(table_model, column_names, verify_permissions, quote_columns=false, constraint_name=none, lookup_cache=none)
<ADAPTER_NAME>__create_unique_key(table_model, column_names, verify_permissions, quote_columns=false, constraint_name=none, lookup_cache=none)
<ADAPTER_NAME>__create_foreign_key(pk_model, pk_column_names, fk_model, fk_column_names, verify_permissions, quote_columns=false, constraint_name=none, lookup_cache=none)
<ADAPTER_NAME>__create_not_null(pk_model, pk_column_names, fk_model, fk_column_names, verify_permissions, quote_columns=false, lookup_cache=none)
<ADAPTER_NAME>__unique_constraint_exists(table_relation, column_names, lookup_cache=none)
<ADAPTER_NAME>__foreign_key_exists(table_relation, column_names, lookup_cache=none)
<ADAPTER_NAME>__have_references_priv(table_relation, verify_permissions, lookup_cache=none)
<ADAPTER_NAME>__have_ownership_priv(table_relation, verify_permissions, lookup_cache=none)

dbt_constraints Limitations

Generally, if you don't meet a requirement, tests are still executed but the constraint is skipped rather than producing an error.

  • All models involved in a constraint must be materialized as table, incremental, snapshot, or seed.

  • If source constraints are enabled, the source must be a table. You must also have the OWNERSHIP table privilege to add a constraint. For foreign keys you also need the REFERENCES privilege on the parent table with the primary or unique key. The package will identify when you lack these privileges on Snowflake and PostgreSQL. Oracle does not provide an easy way to look up your effective privileges so it has an exception handler and will display Oracle's error messages.

  • All columns on constraints must be individual column names, not expressions. You can reference columns on a model that come from an expression.

  • Constraints are not created for failed tests. See how to get around this using severity and config: always_create_constraint: true in the next section.

  • primary_key, unique_key, and foreign_key tests are considered first and duplicate constraints are skipped. One exception is that you will get an error if you add two different primary_key tests to the same model.

  • Foreign keys require that the parent table have a primary key or unique key on the referenced columns. Unique keys generated from standard unique tests are sufficient.

  • The order of columns on a foreign key test must match between the FK columns and PK columns

  • The foreign_key test will ignore any rows with a null column, even if only one of two columns in a compound key is null. If you also want to ensure FK columns are not null, you should add standard not_null tests to your model which will add not null constraints to the table.

  • Referential constraints must apply to all the rows in a table so any tests with a config: where: property will be skipped when creating constraints. See how to disable this rule using config: always_create_constraint: true in the next section.

Advanced: config: always_create_constraint: true property

There is an advanced option to force a constraint to be generated when there is a config: where: property or if the constraint has a threshold. The config: always_create_constraint: true property will override those exclusions. When this setting is in effect, you can create constraints even when you have excluded some records or have a number of failures below a threshold. If your test has a status of 'failed', it will still be skipped. Please see dbt's documentation on how to set a threshold for failures.

Caveat Emptor:

  • You will get an error if you try to force constraints to be generated that are enforced by your database. On Snowflake that is only a not_null constraint but on databases like Oracle, all the generated constraints are enforced.
  • This feature could cause unexpected query results on Snowflake due to join elimination.

This is an example using the feature:

  - name: dim_duplicate_orders
    description: "Test that we do not try to create PK/UK on failed tests"
    columns:
      - name: o_orderkey
        description: "The primary key for this table"
      - name: o_orderkey_seq
        description: "duplicate seq column to test UK"
    tests:
      # This constraint should be skipped because it has failures
      - dbt_constraints.primary_key:
          column_name: o_orderkey
          config:
            severity: warn
      # This constraint should be still generated because always_create_constraint=true
      - dbt_constraints.unique_key:
          column_name: o_orderkey
          config:
            warn_if: ">= 5000"
            error_if: ">= 10000"
            always_create_constraint: true
      # This constraint should be still generated because always_create_constraint=true
      - dbt_constraints.unique_key:
          column_name: o_orderkey_seq
          config:
            severity: warn
            always_create_constraint: true

Primary Maintainers

This is a community-developed package, not an official Snowflake offering. It comes with no support or warranty. However, feel free to raise a github issue if you find a bug or would like a new feature.

Legal

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this package except in compliance with the License. You may obtain a copy of the License at: http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.