[pgpool-hackers: 160] Suggested patch for check_temp_table setting

[Matt Solnit]

Hi everyone.  I've been experimenting with the new "check_temp_table" setting
that was introduced in version 3.2.  Setting it to "off" definitely reduces the
number of queries sent to the master, which is great.  However, pgpool-II is
still preventing system catalog queries from being load-balanced.  From the
comments, it sounds like the only reason for doing this is to accommodate temporary
tables:

pool_query_context.c, lines 422-437:
  /*
   * If system catalog is used in the SELECT, we
   * prefer to send to the primary. Example: SELECT
   * * FROM pg_class WHERE relname = 't1'; Because
   * 't1' is a constant, it's hard to recognize as
   * table name.  Most use case such query is
   * against system catalog, and the table name can
   * be a temporary table, it's best to query
   * against primary system catalog.
   * Please note that this test must be done *before*
   * test using pool_has_temp_table.
   */
  else if (pool_has_system_catalog(node))
  {
  	pool_set_node_to_be_sent(query_context, PRIMARY_NODE_ID);
  }

Is there any reason not to use check_temp_table here as well?  The following
change is working fine for me, and has significantly reduced the amount of
traffic going to the master, especially during application start-up (when Hibernate
and the JDBC driver seem to generate a lot of pg_catalog queries).

diff --git a/pool_query_context.c b/pool_query_context.c
index 6dbe397..bb84563 100644
--- a/pool_query_context.c
+++ b/pool_query_context.c
@@ -434 +434 @@ void pool_where_to_send(POOL_QUERY_CONTEXT *query_context, char *query, Node *no
-                                       else if (pool_has_system_catalog(node))
+                                       else if (pool_config->check_temp_table && pool_has_system_catalog(node))

-- Matt