VegaFusion: Automatic Server-Side Scaling for Interactive Vega Visualizations

Abstract

Figure 1: A generic Vega specification is automatically partitioned by the VegaFusion Planner into a runtime specification for the VegaFusion Middleware (describing operations on large datasets) and a client specification for Vega (describing the visualization of the output of these operations as well as client-side interactions). The Middleware dynamically responds to interaction signals from Vega by querying an out-of-browser, natively-compiled VegaFusion Runtime instance and relaying the results back to Vega. The pseudocode on the right illustrates a typical partition for a brushed histogram specification, including the dependencies between data and signals.