We're troubleshooting some performance issues with respect to highlighting, where certain queries that contain 10 - 40 terms take 2x - 3x as long with highlighting enabled.
We're wondering in which phase (and on which nodes) the highlighting is performed? We have three types of nodes:
- "client" nodes which have
node.master
andnode.data
set tofalse
- "data" nodes which have
node.master: false
andnode.data: true
- "master" nodes which have
node.master: true
andnode.data: false
All search and index requests connect to "client" nodes.
Is highlighting always preformed in the "gather" phase and if so, does that mean that in our set up, highlighting is only done on "client" nodes? Or does it depend on the highlighter that the query is using?