Incremental checkpoint
An incremental checkpoint snapshots only what changed since the last one: cheaper to write, more complex to recover.
Incremental checkpoint - snapshotting only the state that changed since the last snapshot, instead of the whole thing every time. An incremental checkpoint of a terabyte of RocksDB state uploads just the new SST files created since the previous one, which turns a full-state upload every interval into a trickle, and makes very large state feasible to protect at all.
The bill arrives at recovery, and in operations. A restore no longer reads one self-contained snapshot but a chain of deltas referencing files shared across checkpoints, so deleting an old checkpoint's directory can break a newer one, reported checkpoint size stops meaning what you think it means, and recovery time depends on history rather than on state size. A real optimisation, with a real new failure surface attached.
Do you need it? Only when a full snapshot is too big to write inside your interval, which mostly means your state has outgrown one machine. Ubik takes the other exit: keep state on one box, write one atomic checkpoint file, resume from it. At the sizes a single process handles, the full snapshot stays cheap and the word "incremental" never comes up.