The visualizations feedback for individual nodes is often delayed due to ongoing long computations. But in many cases we could do much better and provide data earlier.

A small reproducible example is visible in this screenshot:

On startup two of the first nodes have been executed and the program currently loads some big csv file.
In the meantime a visualization for .length is stuck waiting for data. The latter will only be provided once a visualization request is accepted and kicks of another execution. The data is already cached (from previous execution) and the user gets the final answer:

The wait, sometimes a really long one, is undesired as the value for .length has already been computed and the "stuck" node is after it, in terms of node order.

To improve we could

1. Leverage the functionality of safepoints to interrupt long executions and register visualization requests
2. Make runtime cache values reactive so that any update to RHS expressions automatically sends the visualization data, if needed.