This page contains a set of tips for how to improve the overall performance of job execution.
Filter data early
If you know that you are dropping some rows and columns from your dataset, add these transform steps early in your recipe. This reduction simplifies working with the content through the application and, at execution, speeds the processing of the remaining valid data. Since you may be executing your job multiple times before it is finalized, it should also speed your development process.
- To drop columns:
- To delete rows:
deletetransform with a
rowparameter value to identify the rows to remove. For example, the following removes all rows that lack a value for the
You can paste Wrangle steps into the Transformer Page.
Similarly, you can use the
keeptransform to retain the rows of interest, dropping the rows that do not match. For example, the following transform keeps all rows that lack a value in the
Perform joins early
After you have filtered out unneeded rows and columns, join operations should be performed in your recipe. These steps bring together your data into a single consistent dataset. By doing them early in the process, you reduce the chance of having changes to your join keys impacting the results of your join operations. See Join Page.
Perform unions late
Union operations should be generally performed later in the recipe so that you have a small chance of changes to the union operation, including dataset refreshes, affecting the recipe and the output.
NOTE: If your dataset requires a significant amount of data cleaning, you should perform your unions early in your recipe, so that all cleaning steps can be applied once across the dataset.
See Union Page.