Incremental loading#

Incremental loading is the act of loading only new or changed data and not old records that we have already loaded. It enables low-latency and low-cost data transfer.

The challenge of incremental pipelines is that if we do not keep track of the state of the load (i.e., which increments were loaded and which are to be loaded), we may encounter issues. Read more about state here.

Choosing a write disposition#

The 3 write dispositions:#

Full load: replaces the destination dataset with whatever the source produced on this run. To achieve this, use write_disposition='replace' in your resources. Learn more in the full loading docs.
Append: appends the new data to the destination. Use write_disposition='append'.
Merge: Merges new data into the destination using merge_key and/or deduplicates/upserts new data using primary_key. Use write_disposition='merge'.

How to choose the right write disposition#

![write disposition flowchart](https://storage.googleapis.com/dlt-blog-images/flowchart_for_scd2.png)

The "write disposition" you choose depends on the dataset and how you can extract it.

To find the "write disposition" you should use, the first question you should ask yourself is "Is my data stateful or stateless"? Stateful data has a state that is subject to change - for example, a user's profile. Stateless data cannot change - for example, a recorded event, such as a page view.

Because stateless data does not need to be updated, we can just append it.

For stateful data, comes a second question - Do you need to track history of change ? If yes, you should use slowly changing dimensions (Type-2), which allow you to maintain historical records of data changes over time.

If not, then we need to replace the entire dataset. However, if we can request the data incrementally, such as "all users added or modified since yesterday," then we can simply apply changes to our existing dataset with the merge write disposition.