when is database denormalization a good idea

hey all, ive been thinking about database design lately. we often hear about normalization being the 'right' way to structure relational databases to avoid redundancy and ensure data integrity. but ive also seen instances where denormalization seemed to offer performance benefits, especially for read-heavy workloads. are there specific scenarios or trade-offs where denormalizing your database tables actually makes sense? what are the key risks to watch out for when you go down that path?