i'm assuming you're talking about the real postgres integration that LL is working on, not the KV joke. There is performance testing from alby from a year and a half ago for the KV one
i missed the latest LLabs roundtable but there are generally several problems to performance:
  • external database adds network latency and round trips over boltdb
  • you need very good architecture, schema and implementation due to row/table locking with updates etc
  • if you have synchronous replication for high availability you inject more latency into the process
Generally I don't think you'd want to be the early adopter of this if you want a high performance node since i'd expect a mixed bag of results and issues before the database gets polished enough. I'm also not sure postgresql integration, if done with 1:1 mapping is ever gonna be as high performance as boltdb. But I have not kept up2date with latest developments and architecture changes in lnd to see how they envisioned this migration and getting the performance up for it.