balancer_defragmentation_merge_chunks.js is currently cycling throw all the CSRS nodes to catch the first one that enter a specific failpoint. In order to reduce this waiting time we should always start from the current CSRS primary node.