Checkpoint
The moment the database flushes all dirty buffers to disk and records how far into the transaction log it has come. After a crash only the log from that point onward needs replaying.
It's therefore a trade between ongoing work and recovery time. Infrequent checkpoints mean less writing during normal operation, since a hot page can be modified a hundred times before being written once — but a longer startup after a failure. Frequent checkpoints give a fast restart and smoother, though larger, write load. In PostgreSQL it's governed by checkpoint_timeout and max_wal_size, and checkpoint_completion_target spreads the writes across the interval so they don't arrive as an I/O spike. If you see "checkpoints are occurring too frequently" in the log, max_wal_size is set too low.