How does captain scheduler_load_based calculation on its members to distribute scheduled saved searches ? Is it based on number of searches currently running or available cores or memory etc
Each members send the number of searches they are running for the past 1m, 5m, 15m to the captain. It does a heuristics on that
Captain selects a peer by two approach
load based :
Each member sends its load over its heartbeat to captain (1m, 5m, 15m) just the searches it have seen during this period. Captain picks one with a simple heuristics.
Round robin:
Just pick next in a circular loop on the UP peers.