On Wed, Jan 15, 2020 at 6:38 PM Peter Geoghegan <p...@bowt.ie> wrote: > There are some outstanding questions about how B-Tree deduplication > [1] should be configured, and whether or not it should be enabled by > default. I'm starting this new thread in the hopes of generating > discussion on these high level questions.
It seems like the issue here is that you're pretty confident that deduplication will be a win for unique indexes, but not so confident that this will be true for non-unique indexes. I don't know that I understand why. It does seem odd to me to treat them differently, but it's possible that this is a reflection of my own lack of understanding. What do other database systems do? I wonder whether we could avoid the downside of having only one LP_DEAD bit per line pointer by having a bit per TID within the compressed tuples themselves. I assume you already thought about that, though. What are the characteristics of this system if you have an index that is not declared as UNIQUE but actually happens to be UNIQUE? -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
