Can you explain, what indexing library are you using & how it compares with lucene/trinity and solr/elastic/vespa.ai sharding/replication/performance/efficiency ? Even vespa claims that you need memory mostly storage to have nice SLA, while google is said to have written special index just for ssd-storage (they have a big one).

Usually columnar store is used for aggregations and not for search (inverted-index for that)?

Does it support realtime index or just near-real-time ?