On Mon, Sep 04, 2023 at 08:00:52PM +0200, Jim Jones wrote:
> LGTM. It applies and builds cleanly, all tests pass and documentation also
> builds ok. The CFbot seems also much happier now :)
+ /*
+ * Open and lock the relation. ShareLock is sufficient since we only need
+ * to prevent schema and data changes in it. The lock level used here
+ * should match catalog's reindex_relation().
+ */
+ rel = try_table_open(relid, ShareLock);
I was eyeing at 0003, and this strikes me as incorrect. Sure, this
matches what reindex_relation() does, but you've missed that
CONCURRENTLY takes a lighter ShareUpdateExclusiveLock, and ShareLock
conflicts with it. See:
https://www.postgresql.org/docs/devel/explicit-locking.html
So, doesn't this disrupt a concurrent REINDEX?
--
Michael
