Re: Faster "SET search_path"

On Tue, Nov 14, 2023 at 08:13:25PM -0800, Jeff Davis wrote:
> On Thu, 2023-10-19 at 19:01 -0700, Jeff Davis wrote:
> > 0003: Cache for recomputeNamespacePath.
>
> Committed

Commit f26c236 wrote:
> Add cache for recomputeNamespacePath().
>
> When search_path is changed to something that was previously set, and
> no invalidation happened in between, use the cached list of namespace
> OIDs rather than recomputing them. This avoids syscache lookups and
> ACL checks.

> --- a/src/backend/catalog/namespace.c
> +++ b/src/backend/catalog/namespace.c

> @@ -109,11 +110,13 @@
> * activeSearchPath is always the actually active path; it points to
> * to baseSearchPath which is the list derived from namespace_search_path.
> *
> - * If baseSearchPathValid is false, then baseSearchPath (and other
> - * derived variables) need to be recomputed from namespace_search_path.
> - * We mark it invalid upon an assignment to namespace_search_path or receipt
> - * of a syscache invalidation event for pg_namespace. The recomputation
> - * is done during the next lookup attempt.
> + * If baseSearchPathValid is false, then baseSearchPath (and other derived
> + * variables) need to be recomputed from namespace_search_path, or retrieved
> + * from the search path cache if there haven't been any syscache
> + * invalidations. We mark it invalid upon an assignment to
> + * namespace_search_path or receipt of a syscache invalidation event for
> + * pg_namespace or pg_authid. The recomputation is done during the next

You're caching the result of object_aclcheck(NamespaceRelationId, ...), so
pg_auth_members changes and pg_database changes also need to invalidate this
cache. (pg_database affects the ACL_CREATE_TEMP case in
pg_namespace_aclmask_ext() and affects ROLE_PG_DATABASE_OWNER membership.)