Is that normal that TSI-based shards uses a heck of amount of memory?
I upgraded from 1.4.3 to 1.5.4 and followed steps documented here https://docs.influxdata.com/influxdb/v1.6/administration/upgrading/ specifically section Upgrading from 1.3 / 1.4 (no TSI Preview) to 1.5.x (TSI enabled)
After I started influxd the memory usage skipped above 98% and kept hovering there. Then eventually process crashed due to OOM.
So I followed steps in procedure Switching from disk (TSI-based) index to in-memory (TSM-based) index
And memory usage got back to its previous level.