While partitioning a table into shards, what is the most important factor to consider while selecting a partition key

1. Should we try to make sure data is evenly divided across the shards, so that we don’t end up in “hot” shards
2. Should we try to route most of our queries to the same partition?

Is there any article regarding this? I am aware that there is no one-fit answer for all use cases, but still, any guidelines or best practices to follow?