(Just reposting this from the Slack channel as suggested by Nikita.)
fulltext support for rowstore tables projected? Is it on the roadmap?
@Alex Gunnarson columnstore tables are getting a lot more powerful. We will have secondary indexes in 7.0 so it may work for you. What’s your use case ?
Exciting @nikita! So it’s more of a theoretical question for us at this time. We’re building an EAVT (entity/attribute/value/“time” — i.e. temporally-based transaction ID) store on top of MemSQL and we were wanting to keep everything in memory for now for speed and simplicity. We’re using one table for each built-in datatype that the “V” in EAVT can be — 1) tinyints/booleans, 2) smallints, 3) mediumints, 4) ints, 5) longs, 6) floats, 7) references/UUIDs, 8)opaque binary data, 9) strings, and 10) fulltext-indexed strings — rather than e.g. having one varbinary EAVT table. This makes it so can make proper comparisons/aggregations among values. Currently, the fulltext-indexed-strings EAVT table is a columnstore and this seems to be working really well.
The fact that columnstore tables are getting secondary indexes will be really helpful to us! This is because at some point we’d like to test a hypothesis. We’re thinking that rather than storing all rows in memory, including non-current states of entities, we might want to offload historical data to columnstore tables. This would significantly reduce the size of the in-memory tables, which would make it cheaper for us, and the reduced table size would speed up queries on current entity-states.