This dbt package is for Snowflake ❄️ only.

It is reproducing dbt incremental materialization leveraging on Snowflake streams to :

• Improve model performance 💨. Stream accelerate source table scan focusing new data only
• Optimise FinOps 💰. Snowflake warehouse will be started only if stream is not empty.
• Simplify code. is_incremental() is generally not required as stream natively collect "new" rows

New to dbt packages? Read more about them here.

1. Include this package in your packages.yml. An example can be found here in the working example project
2. Run dbt deps

Use incremental_stream materialisation like dbt incremental model :

• Replace ref by stream_ref macro to add a stream on a dbt model
• Replace sourceby stream_source to add a stream on a source
• incr_stream.get_stream_metadata_columns() must be included to retreive METADATA columns of each STREAMS. Hence the materialization can deal accordingly with the changes on inputs tables (INSERT, UPDATE, DELETE) but Metadata columns (METADATA$ACTION, METADATA$ISUPDATE, METADATA$ROW_ID) will not be included in the model
• Like in dbt incremental model unique_key optional parameter is supported. DBT will perform a MERGE with a unique_key an INSERT instead.
• Like in dbt incremental model --full-refresh rebuild the target model based on the source tables
• An optional dbt Project Variable can be added to create streams at a specific timestamp using Snowflake Time Travel with Stream. The below example shows how to force to recreate a given model and related streams (using --full-refresh option) at given TIMESTAMP using --vars '{TIMESTAMP: 2024-04-13 00:00:00}'. Date format is : yyyy-mm-dd hh24:mi:ss

dbt run --select 'conso_client_source'  --vars '{TIMESTAMP: 2024-04-13 00:00:00}' --full-refresh

{{-
   config(
       materialized='incremental_stream',
       unique_key=['ID']
   )
   select 
       *
   from incr_stream.stream_source('src', 'src_table')
-}}

A test model conso_client.sql following this pattern can be found in the integration_tests sub DBT project included in the package.

{{-
    config(
        materialized='incremental_stream',
        unique_key=['column1']
    )
    select 
        column1 as c_1,
        ...,
        columnN as c_n
        {{ incr_stream.get_stream_metadata_columns() }}
    from incr_stream.stream_ref('input_model')
-}}

A test model conso_client_source.sql following this pattern can be found in the integration_tests sub DBT project included in the package.

Exemple with source

{{-
    config(
        materialized='incremental_stream',
        unique_key=['column1']
    )
    select 
        column1 as c_1,

        ...,
        columnN as c_n
        {{ incr_stream.get_stream_metadata_columns() }}
    from incr_stream.stream_source('src', 'src_table')
-}}

Can be found in this test model: conso_client_insert

A test model conso_client_multiple_streams.sql implementing this pattern can be found in the integration_tests sub DBT project included in the package.

stream_ref and stream_source macro include created streams in the compiled SQL code of the model.

• Stream created in {{this}}.database and {{this}}.schema
• Stream name follows this naming convention : S_{{this.name}}_{{source_table_name}}
• get_stream_metadata_columns is used to easily retreive metadata columns from the stream (METADATA$ACTION, METADATA$ISUPDATE, METADATA$ROW_ID)

Incremental materialization will perform the following tasks :

1. Create {{this}} table IF NOT EXISTS
2. Create required stream IF NOT EXISTS
3. CREATE OR REPLACE a view to perform the transformation developped in the model
4. If stream is empty materialization stops and prevent Snowflake Warehouse RESUME
5. MERGE the view in {{this}} table based on unique_key provided

integration_test contains a working example with models and a test:

• To validate incremental_stream and incremental materialization produce the same output
• To do performance test

And more to explore in sub DBT Project integration_tests

# 1. Add 30 new random clients to ADD_CLIENTS table 
# 2. Merge it in CONSO_CLIENT and CONSO_CLIENT_INCR
# 3. Test CONSO_CLIENT and CONSO_CLIENT_INCR equals
cd ./integration_test
dbt build

Python 🐍 incremental model :

1. Creating a ADD_CLIENTS table if not exists in a STG schema
2. And adding random clients using Faker library.

ADD_CLIENTS Table

Sample commands

# Add 30 random clients in add_clients model (create table if needed)
dbt run --select add_clients

# Add 100 random clients in add_clients model (create table if needed)
dbt run --select add_clients --vars '{nb_clients: 100}'

# Rebuild ADD_CLIENTS table with 30 new random clients
dbt run --select add_clients --full-refresh

A sample model leveraging on incremental_stream custom materialization

1. Collecting lastest data ingested in add_clients
2. De-duplicating it based on ID with most recent LOADED_AT from add_clients stream or table
3. MERGE data in CONSO_CLIENTS table with ID as unique key

Sample commands

# Merge last clients inserted in add_clients model
dbt run --select conso_clients

# Drop CONSO_CLIENTS table and rebuild it based on ADD_CLIENTS table
dbt run --select conso_clients --full-refresh

Current known limitations :

• When a stream_ref or a stream_source table is dropped and recreated stream must be dropped. If not you will observe the following error

 091901 (01000): Base table 'DB_NAME.SCH_NAME.XXX_TABLE' dropped, cannot read from stream 'S_YYY_XXX_TABLE' in line 1 position 21

• on_schema_change parameter might work in certain conditions but is not currently supported

Thanks to jeremiahhansen ❄️ for the inspiring implementation (done in 2020) of streams & tasks on Snowflake