I've was a primary contributor on a migration of time series data to Scylla. As an anecdote, I once emailed our business contact about tracking down why we appeared to have data inconsistencies between our new (Scylla backed) and old system. I thought the e-mail got lost since we never heard back...until 8 months later (long after we had de-prioritized the migration since our old system was "good enough") asking if we had tried the newly released version which fixed a data loss issue.

Blew. My. Mind. Not only because of the radio silence and then dropping back in out of the blue as if no time had passed, but also because they had a data loss issue.

So rechecked out my previous branch, upgraded Scylla versions and sure enough the data differences we were noticing before appeared to be resolved. I couldn't believe the amount of time I had spent combing through my code to see if I had a hard to detect bug somewhere...but nope, it was ScyllaDB (although I am sure there were plenty of other bugs...just they weren't the cause of this specific symptom).

I am actually a fan of ScyllaDB and what is trying to do. Performance was great (as advertised) and management was simple enough; but they are going to need to work pretty hard to convince me "instability" is just rumor after that experience not too many years ago.