PostgreSQL 10 High Performance - Third Edition

Spread checkpoints

Starting in PostgreSQL 8.3, the spread checkpoint feature, tuned by checkpoint_completion_target, aims to reduce the amount of writes in any unit of time by spreading them out over a much longer period of time. For recovery to work, it's only necessary that a checkpoint finish before the next one can start. There's no reason you can't have one checkpoint start almost immediately after the previous one, executing and slowly writing everything that happened while the previous checkpoint was active. This is exactly what the spread checkpoint feature does. It estimates when the next checkpoint is coming based on whether checkpoints seem to be happening because of the timeout being reached or because the checkpoint segment limit ...