trying to confirm (but i think this is the case),

dedupe

mode for realtime tables doesn’t support the metadata TTL like

upsert

mode does, correct? we’re probably ok with

upsert

except it’s unfortunate to lose the ability to enable star tree indexes, and use realtime to offline minion tasks. our use case is that we have very high cardinality tables, but some duplicate data can appear within a short time window if upstream producers encounter issues. we’d prefer not to buffer and eliminate the duplicates upstream since that would add latency and complexity.