Snowflake provides cloud-based data storage and analytics as a service. Among other infrastructures, Snowflake runs on Amazon S3. If all of your source datasets and outputs are in Snowflake locations and other conditions are met, then the entire execution of the transformations can occur in Snowflake.
Transferring the execution steps from the to Snowflake yields the following benefits:
In this scenario, the recipe steps are converted to SQL, which is sequentially executed your source data in temporary tables, from which the results that you have defined for your output are written.
Tip: When running a job in Snowflake, your data never leaves Snowflake.
Tip: Execution on datasets created with custom SQL is supported.
If the requirements and limitations are met, the automatically executes the job in Snowflake.
Spark + Snowflakemust be selected as running environment. See Run Job Page.
Jobs are executed in the virtual warehouse that is specified as part of the Snowflake connection.
NOTE: Job execution requires significantly more resources than ingest or publish jobs on Snowflake. Before you begin using Snowflake, you should verify that your Snowflake virtual warehouse has sufficient resources to handle the expected load. For more information, see Snowflake Connections.
For customer-managed deployments, the following additional requirements apply:
If you are executing a job on Snowflake that utilizes multiple connections, the following requirements must also be met for execution of the job on Snowflake:
Snowflake as a running environment requires that pushdowns be enabled for the workspace and for the specific flow for which the job is executed. If the flow and the workspace are properly configured, the job is automatically executed in Snowflake.
NOTE: Snowflake is not a running environment that you explicitly select or specify as part of a job. If all of the requirements are met, then the job is executed in Snowflake when you select EMR.
All recipe steps, including all functions in the recipe, must be translatable to SQL.
NOTE: When attempting to execute a job in Snowflake, executes each recipe in Snowflake, until it reaches a step that cannot be executed there. At that point, data is transferred to EMR, where the remainder of the job is executed.
If the schemas have changed for your datasets, pushdown execution on Snowflake is not supported. falls back to submitting the job through another running environment.
Visual profiles are unloaded to a stage in an S3 bucket.
If no stage is named, a temporary stage is be created in the
PUBLIC schema. The connecting user must have write access to
NOTE: Creating a temporary stage requires temporary credentials from AWS. These credentials are valid for 1 hour only. If a job is expected to run longer than one hour, you should define a named stage.
For more information, see Snowflake Connections.
The following setting must be enabled in the workspace. Select User menu > Admin console > Workspace settings.
|Logical and physical optimization of jobs|
When enabled, the attempts to optimize job execution through logical optimizations of your recipe and physical optimizations of your recipes interactions with data.
For more information, see Workspace Settings Page.
You must enable the Snowflake optimizations in your flow. In Flow View, select More menu > Optimization settings.
NOTE: All general optimizations must be enabled for your flow, as well as the following optimizations, which are specific to Snowflake.
|Snowflake > Column pruning from source|
When enabled, job execution performance is improved by removing any unused or redundant columns from the source database.
|Snowflake > Filter pushdown|
When this setting is enabled, the optimizes job performance on this flow by pushing data filters directly on the source database.
|Snowflake > Full pushdown||When this setting is enabled, all supported pushdown operations, including full transformation and profiling job execution, is pushed down to Snowflake, where possible.|
For more information, see Flow Optimization Settings Dialog.
You must select
Snowflake + Spark as your running environment in the Run Job page.
NOTE: If this running environment option does not appear in the Run Job page, then all required optimization settings have not been enabled for the workspace or the flow (see above) or the data or recipes do not meet the criteria for execution.
See Run Job Page.
Tip: After launching the job, you can monitor job execution through the Job Details page, which includes a link to the corresponding job in the Snowflake console.
The following transformations and functions are not currently supported for execution in Snowflake.
NOTE: If your recipe contains any of the following transformations or functions, full job execution in Snowflake is not possible at this time. These transformations are expected to be supported and removed from this list in future releases.
For more information on limitations on specific push-downs, see Flow Optimization Settings Dialog.
The following Snowflake data types are not supported for input into :
The following functions are not currently supported for execution in Snowflake.
The following functions are not currently supported for execution in BigQuery.
For more information, see Aggregate Functions.
NUMFORMAT: Only supported when used for rounding.
For more information, see Math Functions.
For more information, see String Functions.
NOTE: When the IFMISSING function immediately follows the PREV function in your recipe steps, Snowflake generates an incorrect value. This is a known issue and will be fixed in a future Snowflake release.
For more information, see Window Functions.
To verify execution in Snowflake, please do the following:
For more information, see Job Details Page.