> > > I've had this idea before, and even wrote a quick POC at one point, but > I had it simply throw a warning rather than an > > error. That avoids the need for any GUC, which I agree is not a good > idea. And it still allows people to create a > > duplicate index if they really want to. > > > > I also appreciate your suggestion regarding the GUC parameter. You've > convinced me that a warning might be a more appropriate approach. A > warning > would still alert users to the potential issue of creating a redundant > index, > while allowing them to proceed if they have a specific reason to do so.
+1 to a warning. One reason they might want to create a duplicate index is to reduce index bloat, especially on partial indexes. Granted, we've had REINDEX CONCURRENTLY since v12, but some reindexing workflows are older than that, and I've seen a few that put a "freshness date" into the index name as a hint to future DBAs about the urgency of the next reindexing. A canceled reindex concurrently could in the past (and maybe still can) leave an invalid index of indeterminate name, and therefore harder to clean up, so that's a possible reason to prefer duplicate-then-swap index creation over reindex.
