Inside the shared buffer, there are pages that have been modified, but their changes have not been written to disk yet; those pages are called dirty pages. A checkpoint is a specific point in time when all of the dirty pages in the shared buffer are written to disk by the checkpointer. So, if there is a crash, the database recovery will start from the most recent checkpoint where all the data was surely consistent. The PostgreSQL checkpointer flushes all of the dirty pages at certain intervals to create a checkpoint for necessary recovery.
There are two types of writers in PostgreSQL: the WAL writer and the background writer. Between the two checkpoints, the background writer flushes some dirty pages to disk so that there are always enough clean pages to reduce the amount of blocking that occurs during the queries.Â
PostgreSQL does not write the updates directly to the data files on disk but on commit; these updates are written to the WAL file sequentially, and later, the checkpointer will write all the dirty blocks to each of the respective data files because data file writing cannot be sequential, so users will not have to wait for the delay to locate the data block inside a data file.
Hence, the WAL writer will write our updates from the shared buffer to the WAL file.
