Hi, On Mon, Aug 31, 2026 at 8:57 PM shihao zhong <[email protected]> wrote: > > Hi hackers, > > In the autovacuum freeze scores added by d7965d65fc5, raising a freeze > weight can make a table's score smaller instead of larger. When a table > gets old, its score is raised to a power to grow fast, but a larger weight > makes this start while the score is still below one, where a power makes it > smaller. So a bigger weight can push a table near wraparound down the list. > The safety net still works; only the order changes. > > The attached patch raises the score to a power only when that makes it > bigger, with a test. New in v19, so no backport needed.
Thanks for reporting this. This is already being discussed here: https://www.postgresql.org/message-id/apWcbq91BAfhmsKU%40nathan. The TAP test case looks interesting and will definitely be useful. You may want to propose it in that thread or here. -- Bharath Rupireddy Amazon Web Services: https://aws.amazon.com
