I've been indexing some documents on a cluster of 3 nodes (1 master-only and 2 regular). There is an external ETL process running on all servers, generating nearly the same load.
Looking at the OS load diagram below, I'm asking myself, why does one of the main ES nodes get double load as the other one?
There are no ingestion pipelines configured.
Does that mean, that the actual document indexing happens on one node of the cluster only, which requires, of course, more computation power, and the other node(s) gets the "prepared" Lucene segments afterwards? Why don't they share that processing task?
How can we check, which node is actually the "processing" node? Can we switch the role somehow?