We have secondary time dimension in our data, which is not a time-series (updated randomly). We were able to use single-dimension partitioning to make querying by that dimension faster, and it works OK for historical nodes. However, real-time layer doesn’t have this partitioning, therefore it represents a real bottleneck during query - it takes 500% more time than querying historical nodes for the same period.
Is there any ingestion configuration that may help speed it up? We’re ingesting our data using Tranquility API.