I think I have found out the issue. this is not because that query nodes were not distributing the queries. It was happening based on round robin fashion. Each node was getting 17% of the queries. The problem with the two query nodes is that it is also an index node. The queries are using three indexes and those indexes are on those indexes only. That I think is spiking the memory and CPU. Ideally index nodes should have been on separate nodes.
3 Likes