The biggest gotcha listed there is how Redshift gets bogged down if you're loading a lot of tables, frequently. You can't run a production Redshift with lots of tables at <15m latency. But in most cases, Redshift is still an overall better choice than a timeseries database because:
* It has all of SQL, including JOINs
* You can use it for both timeseries data and all your other data.
* It has all of SQL, including JOINs
* You can use it for both timeseries data and all your other data.