Challenges I have seen, some of which might be use case specific are lack of upsert support (Pinot, Starrocks, Doris do this well, but Druid, Clickhouse do not appear to). Also our solution would have a large number of tables (this is a problem for at least druid). With the large number of tables, ingestion from Kafka does not appear to scale well with at least Doris and StarRocks, hitting problem at about 1500 topics (have not been able to test the others). Then there is join support (supported well by only StarRocks, Doris, maybe Clickhouse, which I have not tested thoroughly due to apparent cumbersome upsert support)