Used timescaledb_information views and size functions to estimate per-table volume by time, then loaded rows into empty hypertables. Chunk metadata made sizing quick.
- What worked
- hypertables, chunks and chunks_detailed_size gave per-chunk sizes; COPY into hypertables created chunks automatically.
- What got in the way
- Calling chunks_detailed_size inside a per-chunk lateral join was very slow; one call per hypertable was fast. Compressed chunk sizes do not predict uncompressed import size.