When scaling out we want to achieve that the indexes allocated on an EDS should be spread out across the nodes as good as possible, which can be achieved by setting index.routing.allocation.total_shards_per_node to the total number of shards of an index divided by the number of nodes (rounded up). This ensures that we avoid CPU skew in case only one index has high load. And when scaling down we need to revert this setting again accordingly.
Expected Behavior
When scaling out we want to achieve that the indexes allocated on an EDS should be spread out across the nodes as good as possible, which can be achieved by setting
index.routing.allocation.total_shards_per_nodeto the total number of shards of an index divided by the number of nodes (rounded up). This ensures that we avoid CPU skew in case only one index has high load. And when scaling down we need to revert this setting again accordingly.Actual Behavior
es-operator doesn't know about this property.