Very high parse times for query rewrite using cube materialized views

Question

We recently upgraded to version 11.2.0.2 (both AWM and Oracle database server). We are using cube materialized views with query rewrite enabled. Some observations of changes that took place when we rebuilt all the dimensions and cubes in this version:
1. Queries against the base tables take about 35 seconds to parse. Then they execute in a tenth of a second. Even simple queries that just get a sum of the amount from the fact table (which is joined to all the dimensions) takes that long to parse. Once parsed, the queries fly.
2. I noticed that the materialized views used to use grouping sets in the group by clause in version 11.2.0.1, but now they use group by rollup, rollup, rollup...
If we disable query rewrite on the MV or for my session, parse times drop to less than a second. Ideas?

Answer 1

There does appear to be a slow down in parse times between 11.1.0.7 and 11.2. We are still investigating this, but in the meantime here is a way to force the code in 11.2 to generate a GROUPING SETS clause instead of the new ROLLUP syntax.
The trick is to create a dummy hierarchy containing only the leaf level. This is necessary for all dimensions that currently have a single hierarchy. As a simple example I created a dimension, PROD, with three levels, A, B, and C, in a single hierarchy. I then created a one dimensional cube, PC. Here is the SELECT statement for the MV in 11.2. Note the ROLLUP clause in the GROUP BY.
SELECT
  GROUPING_ID(T3."CLASS_ID", T3."FAMILY_ID", T3."ITEM_ID")  SYS_GID,
  (CASE GROUPING_ID(T3."CLASS_ID", T3."FAMILY_ID", T3."ITEM_ID")
   WHEN 3
   THEN TO_CHAR(('A_' || T3."CLASS_ID") )
   WHEN 1
   THEN TO_CHAR(('B_' || T3."FAMILY_ID") )
   ELSE TO_CHAR(('C_' || T3."ITEM_ID") )  END)     "PROD",
  T3."CLASS_ID" "D1_PROD_A_ID",
  T3."FAMILY_ID" "D1_PROD_B_ID",
  T3."ITEM_ID" "D1_PROD_C_ID",
  SUM(T2."UNIT_PRICE")     "PRICE",
  COUNT(T2."UNIT_PRICE")  "COUNT_PRICE",
  COUNT(*)  "SYS_COUNT"
FROM
  GLOBAL."PRICE_AND_COST_FACT" T2,
  GLOBAL."PRODUCT_DIM" T3
WHERE
  (T3."ITEM_ID" = T2."ITEM_ID")
GROUP BY
  (T3."CLASS_ID") ,
  ROLLUP ((T3."FAMILY_ID") , (T3."ITEM_ID") )Next I modified the dimension to add a new hierarchy, DUMMY, containing just the leaf level, C. Once I have mapped the new level and re-enabled MVs, I get the following formulation.
SELECT
  GROUPING_ID(T3."CLASS_ID", T3."FAMILY_ID", T3."ITEM_ID")  SYS_GID,
  (CASE GROUPING_ID(T3."CLASS_ID", T3."FAMILY_ID", T3."ITEM_ID")
   WHEN 3
   THEN ('A_' || T3."CLASS_ID")
   WHEN 1
   THEN ('B_' || T3."FAMILY_ID")
   WHEN 0
   THEN ('C_' || T3."ITEM_ID")
   ELSE NULL END)  "PROD",
  T3."CLASS_ID" "D1_PROD_A_ID",
  T3."FAMILY_ID" "D1_PROD_B_ID",
  T3."ITEM_ID" "D1_PROD_C_ID",
  SUM(T2."UNIT_PRICE")     "PRICE",
  COUNT(T2."UNIT_PRICE")  "COUNT_PRICE",
  COUNT(*)  "SYS_COUNT"
FROM
  GLOBAL."PRICE_AND_COST_FACT" T2,
  GLOBAL."PRODUCT_DIM" T3
WHERE
  (T3."ITEM_ID" = T2."ITEM_ID")
GROUP BY
  GROUPING SETS ((T3."CLASS_ID") , (T3."FAMILY_ID", T3."CLASS_ID") , (T3."ITEM_ID", T3."FAMILY_ID", T3."CLASS_ID") )This puts things back the way they were in 11.1.0.7 when the GROUPING SETS clause was used in all cases. Note that the two queries are logically equivalent.