Is Velox's query execution model pull-based or push-based?

[liujp]

when i read the execution in driver, it seems that the sink operator is first called in the control plan, as this shows, I think velox is a pull-base query engine.

[majetideepak]

I asked @oerling this question a while ago. I understood that Velox can be considered both a pull and push model because of the driver loop. It pulls(getOutput) from one operator and pushes(addInput) it to the next operator.

[liujp]

from the figure above, the difference of the pull/push engine is the order of the operator to call. in the pull mode, the top operator is called, and then the behind operators are called passively. In push mode the last operator is first called, and then push the data to the previous operator.
In velox driver, the process is the same as the pull mode.