Also called: streaming vs materializing node, blocking operator
In plain English
A pipelined node can hand up a result row as soon as it has one — a sequential scan, an index scan, a nested loop. A blocking node must read all of its input before it can return even the first row — a sort, a hash build, an aggregate over unsorted data.
Why it matters
This split is the run-time reason behind startup cost. A blocking node makes a query “freeze” until its input is fully consumed, and a LIMIT above it saves nothing. Replacing a Sort with an index that supplies order, or keeping the blocking node inside work_mem, is how you make a stalling query feel instant.