On Wed, Nov 12, 2025 at 12:58 PM Alexander Kukushkin <[email protected]> wrote: > > On Wed, 12 Nov 2025 at 05:22, Amit Kapila <[email protected]> wrote: >> >> It is difficult to tell when this can happen but you detailed there is >> a theoretical possibility of the same. If we had an in-core cluster >> tool that manages nodes on its own which doesn't allow such scenarios >> to happen then we could possibly say that using such a tool it is safe >> to overwrite old primary's slots. > > > That's a lot of ifs, and none of them could be fulfilled in the foreseeable > future. > > Situation you describe is impossible. > When there is a split-brain and someone drops and re-creating logical slots > with the same names on the old primary - such node can't be joined as a > standby without pg_rewind. > In its current state pg_rewind wipes the pg_replslot directory, and therefore > there will be no replication slots. >
Say, the only operations that happened are slot-drop-recreate and or some operations on unlogged tables. Why then pg_rewind is required? -- With Regards, Amit Kapila.
