Pullup is a compute method where all of the processing executes inside the Apache Spark compute engine. Source data is stored inside a database, where Spark reads it out, and the parameters you set in the Explorer workflow are partitioned and sorted. The profile results of the job are then recorded in the DQ Metastore.
Depending on the size of your dataset and the number of DQ checks performed, this process can greatly slow run times because Spark has its own compute resources, such as memory and CPUs. Pullup has limited support for profiling because it cannot run without Spark. In some cases, you might need to partition your job into smaller segments for Spark to process in parallel and offset some of the compute load. A workaround to some of these Spark processing limitations is to use Pushdown mode instead.
Before the introduction of Pushdown, Pullup was the standard mode for creating and running DQ jobs. Unlike Pushdown, Pullup data sources and datasets are not identified throughout the app with an icon. Therefore, any dataset without the Pushdown icon 
next to it is considered a Pullup dataset.
Benefits of Pullup
- Highly configurable at the Spark-level.
- All DQ Layers are available.
- All connections listed on the Supported Connections page are available for use.
Prerequisites for using Pullup
Before running Pullup jobs, a Collibra DQ user with Admin permissions must: