I hope their internal report has more action-items...<p>They should have had probers running over test accounts on the production system to detect this kind of thing much quicker.<p>Cassandra's config layer needs some kind of check that it isn't allowed to make a change which turns a working service into a non-working one (adding 6 nodes while auto-bootstrap is set to false <i>always</i> leads to data inconsistency - so why did the config system allow such a transition?)<p>Doesn't Cassandra support some kind of autoscaling like spanner/bigtable? Something that happens automatically a couple of times every day is far less likely to go wrong. Less human effort too.