Are there any “soft” limits on Starrocks that users should be wary of? Would a cluster struggle to have tens of thousands of tables? Would a table struggle having tens of thousands of columns?
Not really. It is designed to be scalable. In your post you refer to metadata processes like creating tables. Metadata processes are handled by the FE. Extreme amounts of metadata can increase pressure on FE, but this is not common and usually easy to overcome by increasing the size (memory and cpu) of the FE node. As a general rule of thumb we try to put the cluster pressure into the BE as much as possible since you can add a node to accommodate pressure. The best way to prepare for prod at scale is to test your data under prod-like pressure conditions to understand the limits of your cluster.