In many cases, people want to drop a range of keys. For example, in MyRocks, we encoded rows from one table by prefixing with table ID, so that when we need to drop a table, all keys with the prefix needs to be dropped. For another example, if we store different attributes of a user with key with the format , then if a user deletes the account, we need to delete all the keys prefixing [user_id].

The first way is to issue a DeleteFilesInRange() to the range. The command will remove all SST files only containing keys in the range to delete. For a large chunk, it will immediately reclaim most space, so it is a good solution to problem 1. One thing needs to be taken care of is that after the operations, some keys in the range may still exist in the database. If you want to remove all of them, you should follow up with other operations, but it can be done in a slower pace. Also, be aware that, DeleteFilesInRange() will be removed despite of existing snapshots, so you shouldn't expect to be able to read data from the range using existing snapshots any more.

Problem 2 is a harder problem to solve. One way is to apply DeleteFilesInRange() + so that all keys and tombstones for the range are dropped. It works for large ranges, but it is too expensive if we frequently drop small ranges. The reason is that DeleteFilesInRange() is unlikely to delete any file and CompactRange() will delete far more data than needed because it needs to execute compactions against all SST files containing any key in the deletion rang. For the use cases where CompactRange() is too expensive, there are still two ways to reduce the harm: