If the inserts are generated faster than they are processed, they would certainly not finish the queue. The SQL point values table hasn't been performance tested in quite some time. The NoSQL backend is so much faster it isn't really worth it. Our licensing in version 3 is such that only free licenses don't have an unrestricted NoSQL license. If using SQL you really need to keep that table small.
I've not had issues with Aurora. I'm not sure that was the AWS RDS I've used by name but I've not had issues with the Amazon RDS in general.
Using NoSQL would have behaved fine in the situation you describe to store data sufficient to charting it that way. Rollups should chart even if the value was log on change.