On Wed, May 15, 2024 at 6:31 AM Bertrand Drouvot <bertranddrouvot...@gmail.com> wrote: > Please find attached v6 (only diff with v5 is moving the tests as suggested > above).
I don't immediately know what to think about this patch. I've known about this issue for a long time, but I didn't think this was how we would fix it. I think the problem here is basically that we don't lock namespaces (schemas) when we're adding and removing things from the schema. So I assumed that if we ever did something about this, what we would do would be add a bunch of calls to lock schemas to the appropriate parts of the code. What you've done here instead is add locking at a much lower level - whenever we are adding a dependency on an object, we lock the object. The advantage of that approach is that we definitely won't miss anything. The disadvantage of that approach is that it means we have some very low-level code taking locks, which means it's not obvious which operations are taking what locks. Maybe it could even result in some redundancy, like the higher-level code taking a lock also (possibly in a different mode) and then this code taking another one. I haven't gone through the previous threads; it sounds like there's already been some discussion of this, but I'm just telling you how it strikes me on first look. -- Robert Haas EDB: http://www.enterprisedb.com
