Re: [Postgres-xc-developers] Patch for pg_catalog-based views and pgadmin

[Pavan D. <pav...@gm...>]

On Wed, May 12, 2010 at 3:19 AM, Mason Sharp
<mas...@en...>wrote:

> I started the task to recognize more "Postgres-XC safe" queries.
>
> I initially focused on native pg_catalog based views like pg_settings. This
> is because it used by pgadmin. With the attached patch, pgadmin no longer
> complains about such queries because they are now supported.
>
> The PostgreSQL query rewriter appears to convert the view into a subquery.
> I modified the XC planner to check for pg_catalog table usage in the FROM
> clause (and subqueries thereof).
>
>
FWIW here are couple of things that I noticed:

+   foreach(item, rtable)
+   {
+       RangeTblEntry *rte = (RangeTblEntry *) lfirst(item);
+
+       if (rte->rtekind == RTE_RELATION)
+       {
+           if (rte->relid > FirstNormalObjectId)
+               return false;
+       } else if (rte->rtekind == RTE_SUBQUERY)
+           return contains_only_pg_catalog (rte->subquery->rtable);
+   }

This code will break out of the loop if the subquery contains only
pg_catalog tables. But other rtable entries may contain reference to
non-system tables.

The other thing I am worried about is whether its safe to just rely on
FirstNormalObjectId to decide if a relation belongs to pg_catalog or not. A
user can create a table in pg_catalog (though I think it requires changes to
GUC) and that may get an OID larger than the FirstNormalObjectId or we may
add a new default schema and create tables in the schema as part of initdb
(we do that routinely in EDBAS for example). Should we just explicitly
lookup the namespace of the relation instead ?

Thanks,
Pavan

-- 
Pavan Deolasee
EnterpriseDB     http://www.enterprisedb.com