Reliability

cache

PostgreSQL goes into recovery mode if any of its process (maintenance or backup) exit with an error, as it may have :

• corrupted the cache;
• write erroneous data to disk.

data file

When a block is read, its checksum is compared to the checksum stored in its header.

checksums, enabled by default starting v18

write anomaly

Page-level atomicity.

PostgreSQL always ask to write 8kb of data to disk, but the disk controller may have a different block size. If it requires, says, 10 disk block to accommodate 1 database block, and the write fails at the second block, the database block is lost. This is a "torn write". But, because it has also been written down on the WAL,

Source

https://github.com/NikolayS/postgres_dba