Column stores typically don’t use the primary key for anything except metadata. I think you’re thinking of distributed row stores that use the primary key to sh

Column stores typically don’t use the primary key for anything except metadata. I think you’re thinking of distributed row stores that use the primary key to shard. In a column store, you would apply the sort/cluster key to your time column, which would partition the table within nodes, not across nodes. This will speed up queries that specify a narrow time range, and will not cause any of the pathological behavior you describe. Some column stores allow you to chhose a distribution key, which specifies how data is distributed across nodes. It can be used to optimize joins but it’s dangerous—it can cause exactly the problem you describe.