Avoiding auto-freezing and page corruptions
There are some aspects of VACUUM
whose reason to exist is complex to explain, and occasionally they have negative behavior. Let's look more deeply at those and find some solutions.
Getting ready
PostgreSQL performs regular sweeps to clean out old transaction identifiers, which is known as "freezing". It does this to defer transaction wraparound, which is discussed in more detail in the next recipe.
There are two routes that a row can take in PostgreSQL: a row version dies and needs to be removed by VACUUM
, or a row version gets old enough and needs to be frozen, which is also performed by the VACUUM
process.
Why do we care? Say, we load a table with 100 million rows, and everything is fine. When those rows have been there long enough to begin being frozen, the next VACUUM
operation on that table will rewrite all of them to freeze their transaction identifiers. Put that another way, autovacuum will wake up and start using lots of I/O to perform the...