18 Advanced Query Rewrite

Cost-Based Optimization

When a query is rewritten, Oracle's cost based optimizer compares the cost of the rewritten query and original query and chooses the cheaper execution plan.

Query rewrite is available with cost-based optimization. Oracle Database optimizes the input query with and without rewrite and selects the least costly alternative. The optimizer rewrites a query by rewriting one or more query blocks, one at a time.

If query rewrite has a choice between several materialized views to rewrite a query block, it will select the ones which can result in reading in the least amount of data. After a materialized view has been selected for a rewrite, the optimizer then tests whether the rewritten query can be rewritten further with other materialized views. This process continues until no further rewrites are possible. Then the rewritten query is optimized and the original query is optimized. The optimizer compares these two optimizations and selects the least costly alternative.

Because optimization is based on cost, it is important to collect statistics both on tables involved in the query and on the tables representing materialized views. Statistics are fundamental measures, such as the number of rows in a table, that are used to calculate the cost of a rewritten query. They are created by using the DBMS_STATS package.

Queries that contain inline or named views are also candidates for query rewrite. When a query contains a named view, the view name is used to do the matching between a materialized view and the query. When a query contains an inline view, the inline view can be merged into the query before matching between a materialized view and the query occurs.

In addition, if the inline view's text definition exactly matches with that of an inline view present in any eligible materialized view, general rewrite may be possible. This is because, whenever a materialized view contains exactly identical inline view text to the one present in a query, query rewrite treats such an inline view as a named view or a table.

Figure 18-1 presents a graphical view of the cost-based approach used during the rewrite process.

Figure 18-1 The Query Rewrite Process

Description of the illustration dwhsg017.gif

General Query Rewrite Methods

The optimizer has a number of different types of query rewrite methods that it can choose from to answer a query. When text match rewrite is not possible, this group of rewrite methods is known as general query rewrite. The advantage of using these more advanced techniques is that one or more materialized views can be used to answer a number of different queries and the query does not always have to match the materialized view exactly for query rewrite to occur.

When using general query rewrite methods, the optimizer uses data relationships on which it can depend, such as primary and foreign key constraints and dimension objects. For example, primary key and foreign key relationships tell the optimizer that each row in the foreign key table joins with at most one row in the primary key table. Furthermore, if there is a NOT NULL constraint on the foreign key, it indicates that each row in the foreign key table must join to exactly one row in the primary key table. A dimension object will describe the relationship between, say, day, months, and year, which can be used to roll up data from the day to the month level.

Data relationships such as these are very important for query rewrite because they tell what type of result is produced by joins, grouping, or aggregation of data. Therefore, to maximize the rewritability of a large set of queries when such data relationships exist in a database, you should declare constraints and dimensions.

Checks Made by Query Rewrite

For query rewrite to occur, there are a number of checks that the data must pass. These checks are:

• Join Compatibility Check

• Data Sufficiency Check

• Grouping Compatibility Check

• Aggregate Computability Check

Join Compatibility Check

In this check, the joins in a query are compared against the joins in a materialized view. In general, this comparison results in the classification of joins into three categories:

• Common joins that occur in both the query and the materialized view. These joins form the common subgraph.

• Delta joins that occur in the query but not in the materialized view. These joins form the query delta subgraph.

• Delta joins that occur in the materialized view but not in the query. These joins form the materialized view delta subgraph.

These can be visualized as shown in Figure 18-2.

Figure 18-2 Query Rewrite Subgraphs

Description of the illustration dwhsg016.gif

Common Joins

The common join pairs between the two must be of the same type, or the join in the query must be derivable from the join in the materialized view. For example, if a materialized view contains an outer join of table A with table B, and a query contains an inner join of table A with table B, the result of the inner join can be derived by filtering the antijoin rows from the result of the outer join. For example, consider the following query:

SELECT p.prod_name, t.week_ending_day, SUM(s.amount_sold)
FROM   sales s, products p, times t
WHERE  s.time_id=t.time_id AND s.prod_id = p.prod_id
AND mv.week_ending_day BETWEEN TO_DATE('01-AUG-1999', 'DD-MON-YYYY')
                          AND TO_DATE('10-AUG-1999', 'DD-MON-YYYY')
GROUP BY p.prod_name, mv.week_ending_day;

The common joins between this query and the materialized view join_sales_time_product_mv are:

s.time_id = t.time_id AND s.prod_id = p.prod_id
 

They match exactly and the query can be rewritten as follows:

SELECT p.prod_name, mv.week_ending_day, SUM(s.amount_sold)
FROM   join_sales_time_product_mv
WHERE  mv.week_ending_day BETWEEN TO_DATE('01-AUG-1999','DD-MON-YYYY')
                         AND   TO_DATE('10-AUG-1999','DD-MON-YYYY')
GROUP BY mv.prod_name, mv.week_ending_day;

The query could also be answered using the join_sales_time_product_oj_mv materialized view where inner joins in the query can be derived from outer joins in the materialized view. The rewritten version will (transparently to the user) filter out the antijoin rows. The rewritten query will have the following structure:

SELECT mv.prod_name, mv.week_ending_day, SUM(mv.amount_sold)
FROM   join_sales_time_product_oj_mv mv
WHERE  mv.week_ending_day BETWEEN TO_DATE('01-AUG-1999','DD-MON-YYYY')
  AND  TO_DATE('10-AUG-1999','DD-MON-YYYY') AND mv.prod_id IS NOT NULL
GROUP BY mv.prod_name, mv.week_ending_day;

In general, if you use an outer join in a materialized view containing only joins, you should put in the materialized view either the primary key or the rowid on the right side of the outer join. For example, in the previous example, join_sales_time_product_oj_mv, there is a primary key on both sales and products.

Another example of when a materialized view containing only joins is used is the case of a semijoin rewrites. That is, a query contains either an EXISTS or an IN subquery with a single table. Consider the following query, which reports the products that had sales greater than $1,000:

SELECT DISTINCT p.prod_name
FROM products p
WHERE EXISTS (SELECT p.prod_id, SUM(s.amount_sold) FROM sales s
              WHERE p.prod_id=s.prod_id HAVING SUM(s.amount_sold) > 1000)
              GROUP BY p.prod_id);

This query could also be represented as:

SELECT DISTINCT p.prod_name
FROM products p WHERE p.prod_id IN (SELECT s.prod_id FROM sales s
                    WHERE s.amount_sold > 1000);

This query contains a semijoin (s.prod_id = p.prod_id) between the products and the sales table.

This query can be rewritten to use either the join_sales_time_product_mv materialized view, if foreign key constraints are active or join_sales_time_product_oj_mv materialized view, if primary keys are active. Observe that both materialized views contain s.prod_id=p.prod_id, which can be used to derive the semijoin in the query. The query is rewritten with join_sales_time_product_mv as follows:

SELECT mv.prod_name
FROM (SELECT DISTINCT mv.prod_name FROM  join_sales_time_product_mv mv
      WHERE mv.amount_sold > 1000);

If the materialized view join_sales_time_product_mv is partitioned by time_id, then this query is likely to be more efficient than the original query because the original join between sales and products has been avoided. The query could be rewritten using join_sales_time_product_oj_mv as follows.

SELECT mv.prod_name 
FROM (SELECT DISTINCT mv.prod_name FROM join_sales_time_product_oj_mv mv
      WHERE mv.amount_sold > 1000 AND mv.prod_id IS NOT NULL);

Rewrites with semi-joins are restricted to materialized views with joins only and are not possible for materialized views with joins and aggregates.

Query Delta Joins

A query delta join is a join that appears in the query but not in the materialized view. Any number and type of delta joins in a query are allowed and they are simply retained when the query is rewritten with a materialized view. In order for the retained join to work, the materialized view must contain the joining key. Upon rewrite, the materialized view is joined to the appropriate tables in the query delta. For example, consider the following query:

SELECT p.prod_name, t.week_ending_day, c.cust_city, SUM(s.amount_sold)
FROM   sales s, products p, times t, customers c
WHERE  s.time_id=t.time_id AND s.prod_id = p.prod_id
AND    s.cust_id = c.cust_id
GROUP BY p.prod_name, t.week_ending_day, c.cust_city;

Using the materialized view join_sales_time_product_mv, common joins are: s.time_id=t.time_id and s.prod_id=p.prod_id. The delta join in the query is s.cust_id=c.cust_id. The rewritten form will then join the join_sales_time_product_mv materialized view with the customers table as follows:

SELECT mv.prod_name, mv.week_ending_day, c.cust_city, SUM(mv.amount_sold)
FROM   join_sales_time_product_mv mv, customers c
WHERE  mv.cust_id = c.cust_id
GROUP BY mv.prod_name, mv.week_ending_day, c.cust_city;

Materialized View Delta Joins

A materialized view delta join is a join that appears in the materialized view but not the query. All delta joins in a materialized view are required to be lossless with respect to the result of common joins. A lossless join guarantees that the result of common joins is not restricted. A lossless join is one where, if two tables called A and B are joined together, rows in table A will always match with rows in table B and no data will be lost, hence the term lossless join. For example, every row with the foreign key matches a row with a primary key provided no nulls are allowed in the foreign key. Therefore, to guarantee a lossless join, it is necessary to have FOREIGN KEY, PRIMARY KEY, and NOT NULL constraints on appropriate join keys. Alternatively, if the join between tables A and B is an outer join (A being the outer table), it is lossless as it preserves all rows of table A.

All delta joins in a materialized view are required to be non-duplicating with respect to the result of common joins. A non-duplicating join guarantees that the result of common joins is not duplicated. For example, a non-duplicating join is one where, if table A and table B are joined together, rows in table A will match with at most one row in table B and no duplication occurs. To guarantee a non-duplicating join, the key in table B must be constrained to unique values by using a primary key or unique constraint.

Consider the following query that joins sales and times:

SELECT t.week_ending_day, SUM(s.amount_sold)
FROM   sales s, times t
WHERE s.time_id = t.time_id AND t.week_ending_day BETWEEN TO_DATE
 ('01-AUG-1999', 'DD-MON-YYYY') AND TO_DATE('10-AUG-1999', 'DD-MON-YYYY')
GROUP BY week_ending_day;

The materialized view join_sales_time_product_mv has an additional join (s.prod_id=p.prod_id) between sales and products. This is the delta join in join_sales_time_product_mv. You can rewrite the query if this join is lossless and non-duplicating. This is the case if s.prod_id is a foreign key to p.prod_id and is not null. The query is therefore rewritten as:

SELECT week_ending_day, SUM(amount_sold)
FROM   join_sales_time_product_mv
WHERE  week_ending_day BETWEEN TO_DATE('01-AUG-1999', 'DD-MON-YYYY')
                       AND     TO_DATE('10-AUG-1999', 'DD-MON-YYYY')
GROUP BY week_ending_day;

The query can also be rewritten with the materialized view join_sales_time_product_mv_oj where foreign key constraints are not needed. This view contains an outer join (s.prod_id=p.prod_id(+)) between sales and products. This makes the join lossless. If p.prod_id is a primary key, then the non-duplicating condition is satisfied as well and optimizer rewrites the query as follows:

SELECT week_ending_day, SUM(amount_sold)
FROM   join_sales_time_product_oj_mv
WHERE  week_ending_day BETWEEN TO_DATE('01-AUG-1999', 'DD-MON-YYYY')
  AND  TO_DATE('10-AUG-1999', 'DD-MON-YYYY')
GROUP BY week_ending_day;

The query can also be rewritten with the materialized view join_sales_time_product_mv_oj where foreign key constraints are not needed. This view contains an outer join (s.prod_id=p.prod_id(+)) between sales and products. This makes the join lossless. If p.prod_id is a primary key, then the non-duplicating condition is satisfied as well and optimizer rewrites the query as follows:

SELECT week_ending_day, SUM(amount_sold)
FROM   join_sales_time_product_oj_mv
WHERE  week_ending_day BETWEEN TO_DATE('01-AUG-1999', 'DD-MON-YYYY')
  AND  TO_DATE('10-AUG-1999', 'DD-MON-YYYY')
GROUP BY week_ending_day;

Note that the outer join in the definition of join_sales_time_product_mv_oj is not necessary because the primary key - foreign key relationship between sales and products in the sh schema is already lossless. It is used for demonstration purposes only, and would be necessary if sales.prod_id were nullable, thus violating the losslessness of the join condition sales.prod_id = products.prod_id.

Current limitations restrict most rewrites with outer joins to materialized views with joins only. There is limited support for rewrites with materialized aggregate views with outer joins, so those materialized views should rely on foreign key constraints to assure losslessness of materialized view delta joins.

Join Equivalence Recognition

Query rewrite is able to make many transformations based upon the recognition of equivalent joins. Query rewrite recognizes the following construct as being equivalent to a join:

WHERE table1.column1 = F(args)   /* sub-expression A */
AND table2.column2 = F(args)     /* sub-expression B */

If F(args) is a PL/SQL function that is declared to be deterministic and the arguments to both invocations of F are the same, then the combination of subexpression A with subexpression B be can be recognized as a join between table1.column1 and table2.column2. That is, the following expression is equivalent to the previous expression:

WHERE table1.column1 = F(args)          /* sub-expression A */
AND table2.column2 = F(args)            /* sub-expression B */
AND table1.column1 = table2.column2     /* join-expression J */

Because join-expression J can be inferred from sub-expression A and subexpression B, the inferred join can be used to match a corresponding join of table1.column1 = table2.column2 in a materialized view.

SELECT p.prod_name, s.time_id, t.week_ending_day, SUM(s.amount_sold)
FROM sales s, products p, times t
WHERE s.time_id=t.time_id AND s.prod_id = p.prod_id
GROUP BY p.prod_name, s.time_id, t.week_ending_day;

SELECT prod_name, time_id, week_ending_day, SUM(amount_sold)
FROM join_sales_time_product_mv
GROUP BY prod_name, time_id, week_ending_day;

Rewrite Using Dimensions

This section discusses the following aspects of using dimensions in a rewrite environment:

• Benefits of Using Dimensions

• How to Define Dimensions

CREATE DIMENSION times_dim
LEVEL day IS TIMES.TIME_ID
LEVEL month IS TIMES.CALENDAR_MONTH_DESC
LEVEL quarter IS TIMES.CALENDAR_QUARTER_DESC
LEVEL year IS TIMES.CALENDAR_YEAR
LEVEL fis_week IS TIMES.WEEK_ENDING_DAY
LEVEL fis_month  IS TIMES.FISCAL_MONTH_DESC
LEVEL fis_quarter IS TIMES.FISCAL_QUARTER_DESC
LEVEL fis_year IS TIMES.FISCAL_YEAR
        HIERARCHY cal_rollup
        (day  CHILD OF month CHILD OF quarter CHILD OF year)
        HIERARCHY fis_rollup    
        (day  CHILD OF fis_week CHILD OF fis_month CHILD OF fis_quarter 
        CHILD OF fis_year)
 
        ATTRIBUTE day DETERMINES
        (day_number_in_week, day_name, day_number_in_month,
         calendar_week_number)
 
        ATTRIBUTE month DETERMINES
        (calendar_month_desc, calendar_month_number, calendar_month_name, 
         days_in_cal_month, end_of_cal_month)
 
        ATTRIBUTE quarter DETERMINES 
        (calendar_quarter_desc, calendar_quarter_number,days_in_cal_quarter,
         end_of_cal_quarter)
 
        ATTRIBUTE year DETERMINES
        (calendar_year,  days_in_cal_year, end_of_cal_year)
 
        ATTRIBUTE fis_week DETERMINES
        (week_ending_day, fiscal_week_number);

Text Match Rewrite

The query rewrite engine always initially tries to compare the text of incoming query with the text of the definition of any potential materialized views to rewrite the query. This is because the overhead of doing a simple text comparison is usually negligible comparing to the cost of doing a complex analysis required for the general rewrite.

The query rewrite engine uses two text match methods, full text match rewrite and partial text match rewrite. In full text match the entire text of a query is compared against the entire text of a materialized view definition (that is, the entire SELECT expression), ignoring the white space during text comparison. For example, assume that we have the following materialized view, sum_sales_pscat_month_city_mv:

CREATE MATERIALIZED VIEW sum_sales_pscat_month_city_mv
ENABLE QUERY REWRITE AS
   SELECT p.prod_subcategory, t.calendar_month_desc, c.cust_city,
   SUM(s.amount_sold) AS sum_amount_sold,
   COUNT(s.amount_sold) AS count_amount_sold
   FROM sales s, products p, times t, customers c
   WHERE s.time_id=t.time_id
      AND     s.prod_id=p.prod_id
      AND     s.cust_id=c.cust_id
  GROUP BY p.prod_subcategory, t.calendar_month_desc, c.cust_city;

Consider the following query:

SELECT p.prod_subcategory, t.calendar_month_desc, c.cust_city,
       SUM(s.amount_sold) AS sum_amount_sold,
       COUNT(s.amount_sold) AS count_amount_sold
       FROM sales s, products p, times t, customers c
       WHERE s.time_id=t.time_id
         AND     s.prod_id=p.prod_id
         AND     s.cust_id=c.cust_id
       GROUP BY p.prod_subcategory, t.calendar_month_desc, c.cust_city;
 

This query matches sum_sales_pscat_month_city_mv (white space excluded) and is rewritten as:

SELECT mv.prod_subcategory, mv.calendar_month_desc, mv.cust_city,
       mv.sum_amount_sold, mv.count_amount_sold
FROM   sum_sales_pscat_month_city_mv;

When full text match fails, the optimizer then attempts a partial text match. In this method, the text starting from the FROM clause of a query is compared against the text starting with the FROM clause of a materialized view definition. Therefore, the following query can be rewritten:

SELECT p.prod_subcategory, t.calendar_month_desc, c.cust_city,
       AVG(s.amount_sold)
FROM   sales s, products p, times t, customers c
WHERE  s.time_id=t.time_id AND s.prod_id=p.prod_id
AND    s.cust_id=c.cust_id
GROUP BY p.prod_subcategory, t.calendar_month_desc, c.cust_city;

This query is rewritten as:

SELECT mv.prod_subcategory, mv.calendar_month_desc, mv.cust_city,
       mv.sum_amount_sold/mv.count_amount_sold
FROM   sum_sales_pscat_month_city_mv mv;

Note that, under the partial text match rewrite method, the average of sales aggregate required by the query is computed using the sum of sales and count of sales aggregates stored in the materialized view.

When neither text match succeeds, the optimizer uses a general query rewrite method.

Text match rewrite can distinguish contexts where the difference between uppercase and lowercase is significant and where it is not. For example, the following statements are equivalent:

SELECT X, 'aBc' FROM Y

Select x, 'aBc' From y

Join Back

If some column data requested by a query cannot be obtained from a materialized view, the optimizer further determines if it can be obtained based on a data relationship called a functional dependency. When the data in a column can determine data in another column, such a relationship is called a functional dependency or functional determinance. For example, if a table contains a primary key column called prod_id and another column called prod_name, then, given a prod_id value, it is possible to look up the corresponding prod_name. The opposite is not true, which means a prod_name value need not relate to a unique prod_id.

When the column data required by a query is not available from a materialized view, such column data can still be obtained by joining the materialized view back to the table that contains required column data provided the materialized view contains a key that functionally determines the required column data. For example, consider the following query:

SELECT p.prod_category, t.week_ending_day, SUM(s.amount_sold)
FROM   sales s, products p, times t
WHERE  s.time_id=t.time_id  AND s.prod_id=p.prod_id AND p.prod_category='CD'
GROUP BY p.prod_category, t.week_ending_day;

The materialized view sum_sales_prod_week_mv contains p.prod_id, but not p.prod_category. However, you can join sum_sales_prod_week_mv back to products to retrieve prod_category because prod_id functionally determines prod_category. The optimizer rewrites this query using sum_sales_prod_week_mv as follows:

SELECT p.prod_category, mv.week_ending_day, SUM(mv.sum_amount_sold)
FROM   sum_sales_prod_week_mv mv, products p
WHERE  mv.prod_id=p.prod_id AND p.prod_category='CD'
GROUP BY p.prod_category, mv.week_ending_day;

Here the products table is called a joinback table because it was originally joined in the materialized view but joined again in the rewritten query.

You can declare functional dependency in two ways:

• Using the primary key constraint (as shown in the previous example)

• Using the DETERMINES clause of a dimension

The DETERMINES clause of a dimension definition might be the only way you could declare functional dependency when the column that determines another column cannot be a primary key. For example, the products table is a denormalized dimension table that has columns prod_id, prod_name, and prod_subcategory that functionally determines prod_subcat_desc and prod_category that determines prod_cat_desc.

The first functional dependency can be established by declaring prod_id as the primary key, but not the second functional dependency because the prod_subcategory column contains duplicate values. In this situation, you can use the DETERMINES clause of a dimension to declare the second functional dependency.

The following dimension definition illustrates how functional dependencies are declared:

CREATE DIMENSION products_dim 
        LEVEL product           IS (products.prod_id)
        LEVEL subcategory       IS (products.prod_subcategory) 
        LEVEL category          IS (products.prod_category) 
        HIERARCHY prod_rollup (
                product         CHILD OF 
                subcategory     CHILD OF 
                category
        )
        ATTRIBUTE product DETERMINES products.prod_name 
        ATTRIBUTE product DETERMINES products.prod_desc
        ATTRIBUTE subcategory DETERMINES products.prod_subcat_desc
        ATTRIBUTE category DETERMINES products.prod_cat_desc;

The hierarchy prod_rollup declares hierarchical relationships that are also 1:n functional dependencies. The 1:1 functional dependencies are declared using the DETERMINES clause, as seen when prod_subcategory functionally determines prod_subcat_desc.

If the following materialized view is created:

CREATE MATERIALIZED VIEW sum_sales_pscat_week_mv
ENABLE QUERY REWRITE AS
SELECT p.prod_subcategory, t.week_ending_day,
       SUM(s.amount_sold) AS sum_amount_sole
FROM sales s, products p, times t
WHERE s.time_id = t.time_id AND s.prod_id = p.prod_id
GROUP BY p.prod_subcategory, t.week_ending_day;

Then consider the following query:

SELECT p.prod_subcategory_desc, t.week_ending_day, SUM(s.amount_sold)
FROM   sales s, products p, times t
WHERE  s.time_id=t.time_id AND s.prod_id=p.prod_id
AND    p.prod_subcat_desc LIKE '%Men'
GROUP BY p.prod_subcat_desc, t.week_ending_day;

This can be rewritten by joining sum_sales_pscat_week_mv to the products table so that prod_subcat_desc is available to evaluate the predicate. However, the join will be based on the prod_subcategory column, which is not a primary key in the products table; therefore, it allows duplicates. This is accomplished by using an inline view that selects distinct values and this view is joined to the materialized view as shown in the rewritten query.

SELECT iv.prod_subcat_desc, mv.week_ending_day, SUM(mv.sum_amount_sold)
FROM  sum_sales_pscat_week_mv mv, 
     (SELECT DISTINCT prod_subcategory, prod_subcat_desc
      FROM products) iv
WHERE  mv.prod_subcategory=iv.prod_subcategory 
AND iv.prod_subcat_desc LIKE '%Men'
GROUP BY iv.prod_subcat_desc, mv.week_ending_day;

This type of rewrite is possible because prod_subcategory functionally determines prod_subcategory_desc as declared in the dimension.

Aggregate Computability

Query rewrite can also occur when the optimizer determines if the aggregates requested by a query can be derived or computed from one or more aggregates stored in a materialized view. For example, if a query requests AVG(X) and a materialized view contains SUM(X) and COUNT(X), then AVG(X) can be computed as SUM(X)/COUNT(X).

In addition, if it is determined that the rollup of aggregates stored in a materialized view is required, then, if it is possible, query rewrite also rolls up each aggregate requested by the query using aggregates in the materialized view.

For example, SUM(sales) at the city level can be rolled up to SUM(sales) at the state level by summing all SUM(sales) aggregates in a group with the same state value. However, AVG(sales) cannot be rolled up to a coarser level unless COUNT(sales) or SUM(sales) is also available in the materialized view. Similarly, VARIANCE(sales) or STDDEV(sales) cannot be rolled up unless both COUNT(sales) and SUM(sales) are also available in the materialized view. For example, consider the following query:

ALTER TABLE times MODIFY CONSTRAINT time_pk RELY;
ALTER TABLE customers MODIFY CONSTRAINT customers_pk RELY;
ALTER TABLE sales MODIFY CONSTRAINT sales_time_pk RELY;
ALTER TABLE sales MODIFY CONSTRAINT sales_customer_fk RELY;
SELECT  p.prod_subcategory, AVG(s.amount_sold) AS avg_sales
FROM  sales s, products p WHERE s.prod_id = p.prod_id
GROUP BY p.prod_subcategory;

This statement can be rewritten with materialized view sum_sales_pscat_month_city_mv provided the join between sales and times and sales and customers are lossless and non-duplicating. Further, the query groups by prod_subcategory whereas the materialized view groups by prod_subcategory, calendar_month_desc and cust_city, which means the aggregates stored in the materialized view will have to be rolled up. The optimizer rewrites the query as the following:

SELECT mv.prod_subcategory, SUM(mv.sum_amount_sold)/COUNT(mv.count_amount_sold) 
   AS avg_sales
FROM sum_sales_pscat_month_city_mv mv 
GROUP BY mv.prod_subcategory;

The argument of an aggregate such as SUM can be an arithmetic expression such as A+B. The optimizer tries to match an aggregate SUM(A+B) in a query with an aggregate SUM(A+B) or SUM(B+A) stored in a materialized view. In other words, expression equivalence is used when matching the argument of an aggregate in a query with the argument of a similar aggregate in a materialized view. To accomplish this, Oracle converts the aggregate argument expression into a canonical form such that two different but equivalent expressions convert into the same canonical form. For example, A*(B-C), A*B-C*A, (B-C)*A, and -A*C+A*B all convert into the same canonical form and, therefore, they are successfully matched.

Aggregate Rollup

If the grouping of data requested by a query is at a coarser level than the grouping of data stored in a materialized view, the optimizer can still use the materialized view to rewrite the query. For example, the materialized view sum_sales_pscat_week_mv groups by prod_subcategory and week_ending_day. This query groups by prod_subcategory, a coarser grouping granularity:

ALTER TABLE times MODIFY CONSTRAINT time_pk RELY;
ALTER TABLE sales MODIFY CONSTRAINT sales_time_fk RELY;
SELECT p.prod_subcategory, SUM(s.amount_sold) AS sum_amount
FROM   sales s, products pWHERE  s.prod_id=p.prod_id 
GROUP BY p.prod_subcategory;

Therefore, the optimizer will rewrite this query as:

SELECT mv.prod_subcategory, SUM(mv.sum_amount_sold)
FROM   sum_sales_pscat_week_mv mv
GROUP BY mv.prod_subcategory;

Rollup Using a Dimension

When reporting is required at different levels in a hierarchy, materialized views do not have to be created at each level in the hierarchy provided dimensions have been defined. This is because query rewrite can use the relationship information in the dimension to roll up the data in the materialized view to the required level in the hierarchy.

In the following example, a query requests data grouped by prod_category while a materialized view stores data grouped by prod_subcategory. If prod_subcategory is a CHILD OF prod_category (see the dimension example earlier), the grouped data stored in the materialized view can be further grouped by prod_category when the query is rewritten. In other words, aggregates at prod_subcategory level (finer granularity) stored in a materialized view can be rolled up into aggregates at prod_category level (coarser granularity).

For example, consider the following query:

SELECT p.prod_category, t.week_ending_day, SUM(s.amount_sold) AS sum_amount
FROM   sales s, products p, times t
WHERE  s.time_id=t.time_id AND s.prod_id=p.prod_id
GROUP BY p.prod_category, t.week_ending_day;

Because prod_subcategory functionally determines prod_category, sum_sales_pscat_week_mv can be used with a joinback to products to retrieve prod_category column data, and then aggregates can be rolled up to prod_category level, as shown in the following:

SELECT pv.prod_category, mv.week_ending_day, SUM(mv.sum_amount_sold)
FROM   sum_sales_pscat_week_mv mv,
       (SELECT DISTINCT prod_subcategory, prod_category
        FROM products) pv
WHERE mv.prod_subcategory= pv.prod_subcategory
GROUP BY pv.prod_category, mv.week_ending_day;

When Materialized Views Have Only a Subset of Data

Oracle supports rewriting of queries so that they will use materialized views in which the HAVING or WHERE clause of the materialized view contains a selection of a subset of the data in a table or tables. For example, only those customers who live in New Hampshire. In other words, the WHERE clause in the materialized view will be WHERE state = 'New Hampshire'.

To perform this type of query rewrite, Oracle must determine if the data requested in the query is contained in, or is a subset of, the data stored in the materialized view. The following sections detail the conditions where Oracle can solve this problem and thus rewrite a query to use a materialized view that contains a filtered portion of the data in the detail table.

To determine if query rewrite can occur on filtered data, a selection computability check is performed when both the query and the materialized view contain selections (non-joins) and the check is done on the WHERE as well as the HAVING clause. If the materialized view contains selections and the query does not, then the selection compatibility check fails because the materialized view is more restrictive than the query. If the query has selections and the materialized view does not, then the selection compatibility check is not needed.

A materialized view's WHERE or HAVING clause can contain a join, a selection, or both, and still be used to rewrite a query. Predicate clauses containing expressions, or selecting rows based on the values of particular columns, are examples of non-join predicates.

expression relational operator constant

WHERE prod_id = 102

WHERE prod_id BETWEEN 0 AND 200

WHERE prod_id > 10 AND prod_id < 50

WHERE prod_id BETWEEN 0 AND 200

WHERE (sales.amount_sold * .07) BETWEEN 1.00 AND 100.00

WHERE (sales.amount_sold * .07) BETWEEN 0.0 AND 200.00

WHERE (cost.unit_price * 0.95) > (cost_unit_cost  * 1.25)

WHERE (cost.unit_price * 0.95) > (cost_unit_cost  * 1.25)

WHERE prod_name = 'Shorts' AND prod_category = 'Men'

WHERE prod_category = 'Men'

WHERE prod_category = 'Men'

WHERE prod_name = 'Shorts' AND prod_category = 'Men'

WHERE (prod_id, cust_id) IN ((1022, 1000), (1033, 2000))

WHERE prod_id IN (1022,1033) AND cust_id IN (1000, 2000)

WHERE prod_id = 1022 AND cust_id IN (1000, 2000)

WHERE (prod_id, cust_id) IN ((1022, 1000), (1022, 2000))

WHERE (city_population > 15000 AND city_population < 25000 
   AND state_name = 'New Hampshire')

WHERE (city_population < 5000 AND state_name = 'New York') OR 
   (city_population BETWEEN 10000 AND 50000 AND state_name = 'New Hampshire')

CREATE MATERIALIZED VIEW product_sales_mv
BUILD IMMEDIATE
REFRESH FORCE
ENABLE QUERY REWRITE AS
SELECT p.prod_name, SUM(s.amount_sold) AS dollar_sales
FROM products p, sales s
WHERE p.prod_id = s.prod_id
GROUP BY prod_name
HAVING SUM(s.amount_sold) BETWEEN 5000 AND 50000;

SELECT p.prod_name, SUM(s.amount_sold) AS dollar_sales
FROM products p, sales s WHERE p.prod_id = s.prod_id
GROUP BY prod_name
HAVING SUM(s.amount_sold) BETWEEN 10000 AND 20000;

SELECT mv.prod_name, mv.dollar_sales FROM product_sales_mv mv
WHERE mv.dollar_sales BETWEEN 10000 AND 20000;

CREATE MATERIALIZED VIEW popular_promo_sales_mv
BUILD IMMEDIATE
REFRESH FORCE
ENABLE QUERY REWRITE AS 
SELECT p.promo_name, SUM(s.amount_sold) AS sum_amount_sold
FROM  promotions p, sales s
WHERE s.promo_id = p.promo_id
AND p.promo_name IN ('coupon', 'premium', 'giveaway')
GROUP BY promo_name;

SELECT p.promo_name, SUM(s.amount_sold)
FROM  promotions p, sales s
WHERE s.promo_id = p.promo_id AND p.promo_name IN ('coupon', 'premium')
GROUP BY p.promo_name;

SELECT * FROM popular_promo_sales_mv mv
WHERE mv.promo_name IN ('coupon', 'premium');

Partition Change Tracking (PCT) Rewrite

PCT rewrite enables the optimizer to accurately rewrite queries with fresh data using materialized views that are only partially fresh. To do so, Oracle keeps track of which partitions in the detail tables have been updated. Oracle then tracks which rows in the materialized view originate from the affected partitions in the detail tables. The optimizer is then able to use those portions of the materialized view that are known to be fresh.

The optimizer uses PCT rewrite in QUERY_REWRITE_INTEGRITY = ENFORCED and TRUSTED modes. The optimizer does not use PCT rewrite in STALE_TOLERATED mode because data freshness is not considered in that mode.

You can use PCT rewrite with partitioning, but hash partitioning is not supported. The following sections discuss aspects of using PCT:

• PCT Rewrite Based on Range Partitioned Tables

• PCT Rewrite Based on Range-List Partitioned Tables

• PCT Rewrite Based on List Partitioned Tables

• PCT Rewrite and PMARKER

• PCT Rewrite Using Rowid as PMARKER

CREATE TABLE part_sales_by_time (time_id, prod_id, amount_sold,
       quantity_sold)
  PARTITION BY RANGE (time_id)
  (
    PARTITION old_data
      VALUES LESS THAN (TO_DATE('01-01-1999', 'DD-MM-YYYY'))
      PCTFREE 0  PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0),
    PARTITION quarter1
      VALUES LESS THAN (TO_DATE('01-04-1999', 'DD-MM-YYYY'))
      PCTFREE 0  PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0),
    PARTITION quarter2
      VALUES LESS THAN (TO_DATE('01-07-1999', 'DD-MM-YYYY'))
      PCTFREE 0  PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0),
    PARTITION quarter3
      VALUES LESS THAN (TO_DATE('01-10-1999', 'DD-MM-YYYY'))
      PCTFREE 0  PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0),
    PARTITION quarter4
      VALUES LESS THAN (TO_DATE('01-01-2000', 'DD-MM-YYYY'))
      PCTFREE 0  PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0),
    PARTITION max_partition
      VALUES LESS THAN (MAXVALUE)
      PCTFREE 0  PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
  )
  AS
  SELECT s.time_id, s.prod_id, s.amount_sold, s.quantity_sold
  FROM sales s;
 

CREATE MATERIALIZED VIEW  sales_in_1999_mv
  BUILD IMMEDIATE
  REFRESH FORCE ON DEMAND
  ENABLE QUERY REWRITE
  AS
  SELECT s.time_id, s.prod_id, p.prod_name, SUM(quantity_sold)
  FROM part_sales_by_time s, products p
  WHERE p.prod_id = s.prod_id
    AND s.time_id BETWEEN TO_DATE('01-01-1999', 'DD-MM-YYYY')
    AND TO_DATE('31-12-1999', 'DD-MM-YYYY')
  GROUP BY s.time_id, s.prod_id, p.prod_name;

SELECT s.time_id, p.prod_name, SUM(quantity_sold)
  FROM part_sales_by_time s, products p
  WHERE p.prod_id = s.prod_id
    AND s.time_id < TO_DATE(''01-02-1999'', ''DD-MM-YYYY'')
    AND s.time_id >= TO_DATE(''01-01-1999'', ''DD-MM-YYYY'')
  GROUP BY s.time_id, p.prod_name');

INSERT INTO part_sales_by_time 
  VALUES (TO_DATE('26-12-1999', 'DD-MM-YYYY'),38920,2500, 20);
 
commit;

SELECT s.time_id, p.prod_name, SUM(quantity_sold)
  FROM part_sales_by_time s, products p
  WHERE p.prod_id = s.prod_id
  AND s.time_id < TO_DATE(''01-07-1999'', ''DD-MM-YYYY'')
  AND s.time_id >= TO_DATE(''01-03-1999'', ''DD-MM-YYYY'')
  GROUP BY s.time_id, p.prod_name');

SELECT s.time_id, p.prod_name, SUM(quantity_sold)
  FROM part_sales_by_time s, products p
  WHERE p.prod_id = s.prod_id
  AND s.time_id < TO_DATE(''31-10-1999'', ''DD-MM-YYYY'') AND
       s.time_id > TO_DATE(''01-07-1999'', ''DD-MM-YYYY'')
  GROUP BY s.time_id, p.prod_name');

PCT Rewrite Based on Range-List Partitioned Tables

If the detail table is range-list partitioned, a materialized view that depends on this detail table can support PCT at both the partitioning and subpartitioning levels. If both the partition and subpartition keys are present in the materialized view, PCT can be done at a finer granularity; materialized view refreshes can be done to smaller portions of the materialized view and more queries could be rewritten with a stale materialized view. Alternatively, if only the partition key is present in the materialized view, PCT can be done with courser granularity.

Consider the following range-list partitioned table:

CREATE TABLE sales_par_range_list
 (calendar_year, calendar_month_number, day_number_in_month,
  country_name, prod_id, prod_name, quantity_sold, amount_sold)
PARTITION BY RANGE (calendar_month_number)
SUBPARTITION BY LIST (country_name)
 (PARTITION q1 VALUES LESS THAN (4)
 (SUBPARTITION q1_America VALUES
 ('United States of America', 'Argentina'),
   SUBPARTITION q1_Asia VALUES ('Japan', 'India'),
   SUBPARTITION q1_Europe VALUES ('France', 'Spain', 'Ireland')),
   PARTITION q2 VALUES LESS THAN (7)
  (SUBPARTITION q2_America VALUES
   ('United States of America', 'Argentina'),
   SUBPARTITION q2_Asia VALUES ('Japan', 'India'),
   SUBPARTITION q2_Europe VALUES ('France', 'Spain', 'Ireland')),
     PARTITION q3 VALUES LESS THAN (10)
  (SUBPARTITION q3_America VALUES
   ('United States of America', 'Argentina'),
   SUBPARTITION q3_Asia VALUES ('Japan', 'India'),
   SUBPARTITION q3_Europe VALUES ('France', 'Spain', 'Ireland')),
     PARTITION q4 VALUES LESS THAN (13)
  (SUBPARTITION q4_America VALUES
   ('United States of America', 'Argentina'),
   SUBPARTITION q4_Asia VALUES ('Japan', 'India'),
   SUBPARTITION q4_Europe VALUES ('France', 'Spain', 'Ireland')))
  AS SELECT t.calendar_year, t.calendar_month_number,
     t.day_number_in_month, c1.country_name, s.prod_id,
     p.prod_name, s.quantity_sold, s.amount_sold
  FROM times t, countries c1, products p, sales s, customers c2
  WHERE s.time_id = t.time_id AND s.prod_id = p.prod_id AND
        s.cust_id = c2.cust_id AND c2.country_id = c1.country_id AND
        c1.country_name IN ('United States of America', 'Argentina',
          'Japan', 'India', 'France', 'Spain', 'Ireland');

Then consider the following materialized view sum_sales_per_year_month_mv, which has the total amount of products sold each month of each year:

CREATE MATERIALIZED VIEW  sum_sales_per_year_month_mv
BUILD IMMEDIATE
REFRESH FORCE ON DEMAND
ENABLE QUERY REWRITE AS
SELECT s.calendar_year, s.calendar_month_number,
         SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_range_list s WHERE s.calendar_year > 1990
GROUP BY s.calendar_year, s.calendar_month_number;

sales_per_country_mv supports PCT against sales_par_range_list at the range partitioning level as its range partition key calendar_month_number is in its SELECT and GROUP BY list:

INSERT INTO sales_par_range_list
   VALUES (2001, 3, 25, 'Spain', 20, 'PROD20',  300,  20.50);

This statement inserts a row with calendar_month_number = 3 and country_name = 'Spain'. This row is inserted into partition q1 subpartition Europe. After this INSERT statement, sum_sales_per_year_month_mv is stale with respect to partition q1 of sales_par_range_list. So any incoming query that accesses data from this partition in sales_par_range_list cannot be rewritten, for example, the following statement:

Note that the following query accesses data from partitions q1 and q2. Because q1 was updated, the materialized view is stale with respect to q1 so PCT rewrite is unavailable.

SELECT s.calendar_year, SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_range_list s
WHERE s.calendar_year = 2000 
  AND s.calendar_month_number BETWEEN 2 AND 6
GROUP BY s.calendar_year;

An example of a statement that does rewrite after the INSERT statement is the following, because it accesses fresh material:

SELECT s.calendar_year, SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_range_list s
WHERE s.calendar_year = 2000 AND s.calendar_month_number BETWEEN 5 AND 9
GROUP BY s.calendar_year;

Figure 18-3 offers a graphical illustration of what is stale and what is fresh.

Figure 18-3 PCT Rewrite and Range-List Partitioning

Description of the illustration dwhsg115.gif

PCT Rewrite Based on List Partitioned Tables

If the LIST partitioning key is present in the materialized view's SELECT and GROUP BY, then PCT will be supported by the materialized view. Regardless of the supported partitioning type, if the partition marker or rowid of the detail table is present in the materialized view then PCT is supported by the materialized view on that specific detail table.

CREATE TABLE sales_par_list
(calendar_year, calendar_month_number, day_number_in_month,
 country_name, prod_id, quantity_sold, amount_sold)
 PARTITION BY LIST (country_name)
 (PARTITION America
      VALUES ('United States of America', 'Argentina'),
  PARTITION Asia
      VALUES ('Japan', 'India'),
  PARTITION Europe
      VALUES ('France', 'Spain', 'Ireland'))
  AS SELECT t.calendar_year, t.calendar_month_number, 
         t.day_number_in_month, c1.country_name, s.prod_id, 
         s.quantity_sold, s.amount_sold
  FROM times t, countries c1, sales s, customers c2
  WHERE s.time_id = t.time_id and s.cust_id = c2.cust_id and 
        c2.country_id = c1.country_id and
        c1.country_name IN ('United States of America', 'Argentina',
       'Japan', 'India', 'France', 'Spain', 'Ireland');

If a materialized view is created on the table sales_par_list, which has a list partitioning key, PCT rewrite will use that materialized view for potential rewrites.

To illustrate this feature, the following example creates a materialized view that has the total amounts sold of every product in each country for each year. The view depends on detail tables sales_par_list and products.

CREATE MATERIALIZED VIEW sales_per_country_mv
BUILD IMMEDIATE
REFRESH FORCE ON DEMAND
ENABLE QUERY REWRITE AS
SELECT s.calendar_year AS calendar_year, s.country_name AS country_name,
 p.prod_name AS prod_name, SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_list s, products p
WHERE s.prod_id = p.prod_id AND s.calendar_year <= 2000
GROUP BY s.calendar_year, s.country_name, prod_name;

sales_per_country_mv supports PCT against sales_par_list as its list partition key country_name is in its SELECT and GROUP BY list. Table products is not partitioned, so sales_per_country_mv does not support PCT against this table.

A query could be rewritten (in ENFORCED or TRUSTED modes) in terms of sales_per_country_mv even if sales_per_country_mv is stale if the incoming query accesses only fresh parts of the materialized view. You can determine which parts of the materialized view are FRESH only if the updated tables are PCT enabled in the materialized view. If non-PCT enabled tables have been updated, then the rewrite is not possible with fresh data from that specific materialized view as you cannot identify the FRESH portions of the materialized view.

sales_per_country_mv supports PCT on sales_par_list and does not support PCT on table product. If table products is updated, then PCT rewrite is not possible with sales_per_country_mv as you cannot tell which portions of the materialized view are FRESH.

The following updates sales_par_list as follows:

INSERT INTO sales_par_list VALUES (2000, 10, 22, 'France', 900, 20, 200.99);

This statement inserted a row into partition Europe in table sales_par_list. Now sales_per_country_mv is stale, but PCT rewrite (in ENFORCED and TRUSTED modes) is possible as this materialized view supports PCT against table sales_par_list. The fresh and stale areas of the materialized view are identified based on the partitioned detail table sales_par_list.

Figure 18-4 illustrates what is fresh and what is stale in this example.

Figure 18-4 PCT Rewrite and List Partitioning

Description of the illustration dwhsg114.gif

Consider the following query:

SELECT s.country_name, p.prod_name, SUM(s.amount_sold) AS sum_sales, 
   COUNT(*) AS cnt
FROM sales_par_list s, products p
WHERE s.prod_id = p.prod_id AND s.calendar_year = 2000
  AND s.country_name IN ('United States of America', 'Japan')
GROUP BY s.country_name, p.prod_name;

This query accesses partitions America and Asia in sales_par_list; these partition have not been updated so rewrite is possible with stale materialized view sales_per_country_mv as this query will access only FRESH portions of the materialized view.

The query is rewritten in terms of sales_per_country_mv as follows:

SELECT country_name, prod_name, SUM(sum_sales) AS sum_slaes, SUM(cnt) AS cnt
FROM sales_per_country_mv WHERE calendar_year = 2000 
  AND country_name IN ('United States of America', 'Japan')
GROUP BY country_name, prod_name;

Now consider the following query:

SELECT s.country_name,  p.prod_name,
 SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_list s, products p
WHERE  s.prod_id = p.prod_id AND s.calendar_year = 1999
  AND s.country_name IN ('Japan', 'India', 'Spain')
GROUP BY s.country_name, p.prod_name;

This query accesses partitions Europe and Asia in sales_par_list. Partition Europe has been updated, so this query cannot be rewritten in terms of sales_per_country_mv as the required data from the materialized view is stale.

You will be able to rewrite after any kinds of updates to sales_par_list, that is DMLs, direct loads and Partition Maintenance Operations (PMOPs) if the incoming query accesses FRESH parts of the materialized view.

CREATE MATERIALIZED VIEW sales_per_dt_partition_mv
BUILD IMMEDIATE
REFRESH FORCE ON DEMAND
ENABLE QUERY REWRITE AS
SELECT s.calendar_year AS calendar_year, p.prod_name AS prod_name,
       DBMS_MVIEW.PMARKER(s.rowid) pmarker,
       SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_list s, products p
WHERE s.prod_id = p.prod_id AND s.calendar_year > 2000
GROUP BY s.calendar_year, DBMS_MVIEW.PMARKER(s.rowid), p.prod_name;

DELETE FROM sales_par_list WHERE country_name = 'India';

SELECT p.prod_name, SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_list s, products p
WHERE  s.prod_id = p.prod_id AND s.calendar_year = 2001 AND
       s.country_name IN ('United States of America', 'Argentina')
GROUP BY p.prod_name;

SELECT prod_name, SUM(sum_sales) AS sum_sales, SUM(cnt) AS cnt
FROM sales_per_dt_partition_mv
WHERE calendar_year = 2001 AND pmarker = 1000
GROUP BY prod_name;

CREATE TABLE product_par_list
(prod_id, prod_name, prod_category,
 prod_subcategory, prod_list_price)
 PARTITION BY LIST (prod_category)
 (PARTITION prod_cat1
      VALUES ('Boys', 'Men'),
  PARTITION prod_cat2
      VALUES ('Girls', 'Women'))
 AS
   SELECT prod_id, prod_name, prod_category, prod_subcategory, prod_list_price
   FROM products;

CREATE MATERIALIZED VIEW  sum_sales_per_category_mv
BUILD IMMEDIATE
REFRESH FORCE ON DEMAND
ENABLE QUERY REWRITE AS
SELECT p.rowid prid, p.prod_category, 
       SUM (s.amount_sold) sum_sales, COUNT(*) cnt
FROM sales_par_list s, product_par_list p
WHERE s.prod_id = p.prod_id and s.calendar_year <= 2000
GROUP BY p.rowid, p.prod_category;

product_par_list       pmarker value
----------------       -------------
prod_cat1                      1000
prod_cat2                      1001
prod_cat3                      1002

DELETE FROM product_par_list WHERE prod_name = 'MEN';

SELECT p.prod_category, SUM(s.amount_sold) AS sum_sales, COUNT(*) AS cnt
FROM sales_par_list s, product_par_list p
WHERE  s.prod_id = p.prod_id AND p.prod_category IN
       ('Girls', 'Women') AND s.calendar_year <= 2000
GROUP BY p.prod_category;

SELECT prod_category, sum_sales, cnt
FROM sum_sales_per_category_mv WHERE DBMS_MVIEW.PMARKER(srid) IN (1000)
GROUP BY prod_category;

Multiple Materialized Views

Query rewrite has been extended to enable the rewrite of a query using multiple materialized views. If query rewrite determines that there is no set of materialized views that will return all of the data, then query rewrite retrieves the remaining data from the base tables.

Query rewrite using multiple materialized views can take advantage of many different types and combinations of rewrite, such as using PCT and IN-lists. The following examples illustrate some of the queries where query rewrite is now possible.

Consider the following two materialized views, cust_avg_credit_mv1 and cust_avg_credit_mv2. cust_avg_credit_mv1 asks for all customers average credit limit for each postal code that were born between the years 1940 and 1950. cust_avg_credit_mv2 asks for customers average credit limit for each postal code that were born after 1950 and before or on 1970.

The materialized views' definitions for this example are as follows:

CREATE MATERIALIZED VIEW cust_avg_credit_mv1
ENABLE QUERY REWRITE
AS SELECT cust_postal_code, cust_year_of_birth, 
       SUM(cust_credit_limit) AS sum_credit,
       COUNT(cust_credit_limit) AS count_credit
FROM customers
WHERE cust_year_of_birth BETWEEN 1940 AND 1950
GROUP BY cust_postal_code, cust_year_of_birth;

CREATE MATERIALIZED VIEW cust_avg_credit_mv2
ENABLE QUERY REWRITE
AS SELECT cust_postal_code, cust_year_of_birth,
       SUM(cust_credit_limit) AS sum_credit, 
       COUNT(cust_credit_limit) AS count_credit
FROM customers 
WHERE cust_year_of_birth > 1950 AND cust_year_of_birth <= 1970
GROUP BY cust_postal_code, cust_year_of_birth;

Query 1: One Matched Interval in Materialized View and Query

Consider a query that asks for all customers average credit limit for each postal code who were born between 1940 and 1970. This query is matched by the interval BETWEEN on cust_year_of_birth.

SELECT cust_postal_code, AVG(cust_credit_limit) AS avg_credit
FROM customers c
WHERE cust_year_of_birth BETWEEN 1940 AND 1970
GROUP BY cust_postal_code;

The preceding query can be rewritten in terms of these two materialized views to get all the data as follows:

SELECT v1.cust_postal_code, 
SUM(v1.sum_credit)/SUM(v1.count_credit) AS avg_credit
FROM (SELECT cust_postal_code, sum_credit, count_credit
 FROM cust_avg_credit_mv1 
 GROUP BY cust_postal_code
 UNION ALL
 SELECT cust_postal_code, sum_credit, count_credit
 FROM cust_avg_credit_mv2
 GROUP BY cust_postal_code) v1
 GROUP BY v1.cust_postal_code;

Note that the UNION ALL query is used in an inline view because of the re-aggregation that needs to take place. Note also how query rewrite was the count aggregate to perform this rollup.

Query 2: Query Outside of Data Contained in Materialized View

When the materialized view goes beyond the range asked by the query, a filter (also called selection) is added to the rewritten query to drop out the unneeded rows returned by the materialized view. This case is illustrated in the following query:

SELECT cust_postal_code, SUM(cust_credit_limit) AS sum_credit
FROM customers c 
WHERE cust_year_of_birth BETWEEN 1945 AND 1955
GROUP BY cust_postal_code;

Query 2 is rewritten as:

SELECT v1.cust_postal_code, SUM(v1.sum_credit)
FROM
(SELECT cust_postal_code, SUM(sum_credit) AS sum_credit
FROM cust_avg_credit_mv1 
WHERE cust_year_of_birth BETWEEN 1945 AND 1950
GROUP BY cust_postal_code
UNION ALL
SELECT cust_postal_code, SUM(sum_credit) AS sum_credit
FROM cust_birth_mv2
WHERE cust_year_of_birth > 1950 AND cust_year_of_birth <= 1955
GROUP BY cust_postal_code) v1
GROUP BY v1.cust_postal_code;

Query 3: Requesting More Data Than is in the Materialized View

What if a query asks for more data than is contained in the two materialized views? It will still rewrite using both materialized views and the data in the base table. In the following example, a new set of materialized views without aggregates is defined It will still rewrite using both materialized views and the data in the base table.

CREATE MATERIALIZED VIEW cust_birth_mv1
ENABLE QUERY REWRITE
AS SELECT cust_last_name, cust_first_name, cust_year_of_birth
FROM customers WHERE cust_year_of_birth BETWEEN 1940 AND 1950;

CREATE MATERIALIZED VIEW cust_avg_credit_mv2
ENABLE QUERY REWRITE
AS SELECT cust_last_name, cust_first_name, cust_year_of_birth
FROM customers 
WHERE cust_year_of_birth > 1950 AND cust_year_of_birth <= 1970;

Our queries now require all customers born between 1940 and 1990.

SELECT cust_last_name, cust_first_name 
FROM customers c WHERE cust_year_of_birth BETWEEN 1940 AND 1990;

Query rewrite needs to access the base table to access the customers that were born after 1970 and before or on 1990. Therefore, Query 3 is rewritten as the following:

SELECT cust_last_name, cust_first_name 
FROM cust_birth_mv1 
UNION ALL
SELECT cust_last_name, cust_first_name 
FROM cust_birth_mv2
UNION ALL
SELECT cust_last_name, cust_first_name 
FROM customers c 
WHERE cust_year_of_birth > 1970 AND cust_year_of_birth <= 1990;

Query 4: Requesting Data on More than One Selection Column

Consider the following query, which asks for all customers who have a credit limit between 1,000 and 10,000 and were born between the years 1945 and 1960. This query is a multi-selection query because it is asking for data on more than one selection column.

SELECT cust_last_name, cust_first_name
FROM customers WHERE cust_year_of_birth BETWEEN 1945 AND 1960 AND
   cust_credit_limit BETWEEN 1000 AND 10000;

Figure 18-5 shows a two-selection query, which can be rewritten with the two-selection materialized views described in the following section.

Figure 18-5 Query Rewrite Using Multiple Materialized Views

Description of the illustration dwhsg111.gif

The graph in Figure 18-5 illustrates the materialized views that can be used to satisfy this query. credit_mv1 asks for customers that have credit limits between 1,000 and 5,000 and were born between 1945 and 1950. credit_mv2 asks for customers that have credit limits > 5,000 and <= 10,000 and were born between 1945 and 1960. credit_mv3 asks for customers that have credit limits between 1,000 and 5,000 and were born after 1950 and before or on 1955.

The materialized views' definitions for this case are as follows:

CREATE MATERIALIZED VIEW credit_mv1
ENABLE QUERY REWRITE
AS SELECT cust_last_name, cust_first_name, cust_credit_limit, cust_year_of_birth
FROM customers
WHERE cust_credit_limit BETWEEN 1000 AND 5000 
AND cust_year_of_birth BETWEEN 1945 AND 1950;

CREATE MATERIALIZED VIEW credit_mv2
ENABLE QUERY REWRITE
AS SELECT cust_last_name, cust_first_name, cust_credit_limit, cust_year_of_birth
FROM customers
WHERE cust_credit_limit > 5000 
   AND cust_credit_limit <= 10000 AND cust_year_of_birth BETWEEN 1945 AND 1960;

CREATE MATERIALIZED VIEW credit_mv3
ENABLE QUERY REWRITE AS
SELECT cust_last_name, cust_first_name, cust_credit_limit, cust_year_of_birth
FROM customers
WHERE cust_credit_limit BETWEEN 1000 AND 5000 
  AND cust_year_of_birth > 1950 AND cust_year_of_birth <= 1955;

Query 4 can be rewritten by using all three materialized views to access most of the data. However, because not all the data can be obtained from these three materialized views, query rewrite will also access the base tables to retrieve the data for customers who have credit limits between 1,000 and 5,000 and were born between 1955 and 1960. It is rewritten as follows:

SELECT cust_last_name, cust_first_name
FROM credit_mv1
UNION ALL
SELECT cust_last_name, cust_first_name
FROM credit_mv2
UNION ALL
SELECT cust_last_name, cust_first_name
FROM credit_mv3
UNION ALL
SELECT cust_last_name, cust_first_name
FROM customers
WHERE cust_credit_limit BETWEEN 1000 AND 5000 
  AND cust_year_of_birth > 1955 AND cust_year_of_birth <= 1960;

This example illustrates how a multi-selection query can be rewritten with multiple materialized views. The example was simplified to show no overlapping data among the three materialized views. However, query rewrite can perform similar rewrites.

Query 5: Intervals and Constrained Intervals

This example illustrates how a multi-selection query can be rewritten using a single selection materialized view. In this example, there are two intervals in the query and one constrained interval in the materialized view. It asks for customers that have credit limits between 1,000 and 10,000 and were born between 1945 and 1960. But suppose that credit_mv1 asks for just customers that have credit limits between 1,000 and 5,000. credit_mv1 is not constrained by a selection in cust_year_of_birth, therefore covering the entire range of birth year values for the query.

Figure 18-6 Constrained Materialized View Selections

Description of the illustration dwhsg113.gif

The area between the lines in Figure 18-6 represents the data credit1_mv.

The new credit_mv1 is defined as follows:

CREATE MATERIALIZED VIEW credit_mv1
ENABLE QUERY REWRITE
AS SELECT cust_last_name, cust_first_name, cust_credit_limit, cust_year_of_birth
FROM customers WHERE cust_credit_limit BETWEEN 1000 AND 5000;

The query is as follows:

SELECT cust_last_name, cust_first_name
FROM customers WHERE cust_year_of_birth BETWEEN 1945 AND 1960
  AND cust_credit_limit BETWEEN 1000 AND 10000;

And finally the rewritten query is as follows:

SELECT cust_last_name, cust_first_name
FROM credit_mv1 WHERE cust_year_of_birth BETWEEN 1945 AND 1960
UNION ALL
SELECT cust_last_name, cust_first_name
FROM customers WHERE cust_year_of_brith BETWEEN 1945 AND 1960
  AND cust_credit_limit > 5000 AND cust_credit_limit <= 10000;

Query 6: Query has Single Column IN-List and Materialized Views have Single Column Intervals

Multiple materialized view query rewrite can process an IN-list in the incoming query and rewrite the query in terms of materialized views that have intervals on the same selection column. Given that an IN-list represents discrete values in an interval, this rewrite capability is a natural extension to the intervals only scenario described earlier.

The following is an example of a one column IN-list selection in the query and one column interval selection in the materialized views. Consider a query that asks for the number of customers for each country who were born in any of the following year: 1945, 1950, 1955, 1960, 1965, 1970 or 1975. This query is constrained by an IN-list on cust_year_of_birth.

SELECT c2.country_name, count(c1.country_id)
FROM customers c1, countries c2
WHERE c1.country_id = c2.country_id AND 
   c1.cust_year_of_birth IN (1945, 1950, 1955, 1960, 1965, 1970, 1975)
GROUP BY c2.country_name;

Consider the following two materialized views. cust_country_birth_mv1 asks for the number of customers for each country that were born between the years 1940 and 1950. cust_country_birth_mv2 asks for the number of customers for each country that were born after 1950 and before or on 1970. The preceding query can be rewritten in terms of these two materialized views to get the total number of customers for each country born in 1945, 1950, 1955, 1960, 1965 and 1970. The base table access is required to obtain the number of customers that were born in 1975.

The materialized views' definitions for this example are as follows:

CREATE MATERIALIZED VIEW cust_country_birth_mv1
ENABLE QUERY REWRITE
AS SELECT c2.country_name, c1.cust_year_of_birth,
 COUNT(c1.country_id) AS count_customers
FROM customers c1, countries c2
WHERE c1.country_id = c2.country_id AND 
      cust_year_of_birth BETWEEN 1940 AND 1950
GROUP BY c2.country_name, c1.cust_year_of_birth;

CREATE MATERIALIZED VIEW cust_country_birth_mv2
ENABLE QUERY REWRITE
AS SELECT c2.country_name, c1.cust_year_of_birth, 
 COUNT(c1.country_id) AS count_customers
FROM customers c1, countries c2
WHERE c1.country_id = c2.country_id AND cust_year_of_birth > 1950 
AND cust_year_of_birth <= 1970
GROUP BY c2.country_name, c1.cust_year_of_birth;

So, Query 6 is rewritten as:

SELECT v1.country_name, SUM(v1.count_customers)
FROM (SELECT country_name, SUM(count_customers) AS count_customers
FROM cust_country_birth_mv1 
WHERE cust_year_of_birth IN (1945, 1950)
GROUP BY country_name
UNION ALL
SELECT country_name, SUM(count_customers) AS count_customers
FROM cust_country_birth_mv2
WHERE cust_year_of_birth IN (1955, 1960, 1965, 1970)
GROUP BY country_name
UNION ALL
SELECT c2.country_name, COUNT(c1.country_id) AS count_customers
FROM customers c1, countries c2
WHERE c1.country_id = c2.country_id AND cust_year_of_birth IN (1975)
GROUP BY c2.country_name) v1
GROUP BY v1.country_name;

Query 7: PCT Rewrite with Multiple Materialized Views

Rewrite with multiple materialized views can also take advantage of PCT rewrite. PCT rewrite refers to the capability of rewriting a query with only the fresh portions of a materialized view when the materialized view is stale. This feature is used in ENFORCED or TRUSTED integrity modes, and with multiple materialized view rewrite, it can use the fresh portions of the materialized view to get the fresh data from it, and go to the base table to get the stale data. So the rewritten query will UNION ALL only the fresh data from one or more materialized views and obtain the rest of the data from the base tables to answer the query. Therefore, all the PCT rules and conditions apply here as well. The materialized view should be PCT enabled and the changes made to the base table should be such that the fresh and stale portions of the materialized view can be clearly identified.

This example assumes you have a query that asks for customers who have credit limits between 1,000 and 10,000 and were born between 1945 and 1964. Also, the customer table is partitioned by cust_date_of_birth and there is a PCT-enabled materialized view called credit_mv1 that also asks for customers who have a credit limit between 1,000 and 10,000 and were born between 1945 and 1964.

SELECT cust_last_name, cust_first_name
FROM customers WHERE cust_credit_limit BETWEEN 1000 AND 10000;

In Figure 18-7, the diagram illustrates those regions of the materialized view that are fresh (dark) and stale (light) with respect to the base table partitions p1-p6.

Figure 18-7 PCT and Multiple Materialized View Rewrite

Description of the illustration dwhsg112.gif

Let us say that we are in ENFORCED mode and that p1, p2, p3, p5, and p6 of the customer table are fresh and partition p4 is stale. This means that all partitions of credit_mv1 cannot be used to answer the query. The rewritten query must get the results for customer partition p4 from some other materialized view or as shown in this example, from the base table. Below, we can see part of the table definition for the customers table showing how the table is partitioned:

CREATE TABLE customers
(PARTITION BY RANGE (cust_year_of_birth)
 PARTITION p1 VALUES LESS THAN (1945),
 PARTITION p2 VALUES LESS THAN (1950), 
 PARTITION p3 VALUES LESS THAN (1955),
 PARTITION p4 VALUES LESS THAN (1960),
 PARTITION p5 VALUES LESS THAN (1965),
 PARTITION p6 VALUES LESS THAN (1970);

The materialized view definition for the preceding example is as follows:

CREATE MATERIALIZED VIEW credit_mv1
ENABLE QUERY REWRITE
AS SELECT cust_last_name, cust_first_name, cust_credit_limit, cust_year_of_birth
FROM customers
WHERE cust_credit_limit BETWEEN 1000 AND 10000 
AND cust_year_of_birth BETWEEN 1945 AND 1964;

Note that this materialized view is PCT enabled with respect to table customers.

The rewritten query is as follows:

SELECT cust_last_name, cust_first_name FROM credit_mv1
WHERE cust_credit_limit BETWEEN 1000 AND 10000 AND
   (cust_year_of_birth >= 1945 AND cust_year_of_birth < 1955 OR
    cust_year_of_birth BETWEEN 1945 AND 1964)
UNION ALL
SELECT cust_last_name, cust_first_name
FROM customers WHERE cust_credit_limit BETWEEN 1000 AND 10000 
  AND cust_year_of_birth < 1960 AND cust_year_of_birth >= 1955;

Query Rewrite Using Nested Materialized Views

Query rewrite attempts to iteratively take advantage of nested materialized views. Oracle Database first tries to rewrite a query with materialized views having aggregates and joins, then with a materialized view containing only joins. If any of the rewrites succeeds, Oracle repeats that process again until no rewrites are found. For example, assume that you had created materialized views join_sales_time_product_mv and sum_sales_time_product_mv as in the following:

CREATE MATERIALIZED VIEW join_sales_time_product_mv
ENABLE QUERY REWRITE AS
SELECT p.prod_id, p.prod_name, t.time_id, t.week_ending_day,
       s.channel_id, s.promo_id, s.cust_id, s.amount_sold
FROM   sales s, products p, times t
WHERE  s.time_id=t.time_id AND s.prod_id = p.prod_id;

CREATE MATERIALIZED VIEW sum_sales_time_product_mv
ENABLE QUERY REWRITE AS
SELECT mv.prod_name, mv.week_ending_day, COUNT(*) cnt_all, 
       SUM(mv.amount_sold) sum_amount_sold, 
       COUNT(mv.amount_sold) cnt_amount_sold
FROM join_sales_time_product_mv mv
GROUP BY mv.prod_name, mv.week_ending_day;

Then consider the following query:

SELECT p.prod_name, t.week_ending_day, SUM(s.amount_sold)
FROM sales s, products p, times t
WHERE s.time_id=t.time_id AND s.prod_id=p.prod_id
GROUP BY p.prod_name, t.week_ending_day;

Oracle finds that join_sales_time_product_mv is eligible for rewrite. The rewritten query has this form:

SELECT mv.prod_name, mv.week_ending_day, SUM(mv.amount_sold)
FROM join_sales_time_product_mv mv
GROUP BY mv.prod_name, mv.week_ending_day;

Because a rewrite occurred, Oracle tries the process again. This time, the query can be rewritten with single-table aggregate materialized view sum_sales_store_time into the following form:

SELECT mv.prod_name, mv.week_ending_day, mv.sum_amount_sold
FROM sum_sales_time_product_mv mv;

Query Rewrite in the Presence of Inline Views

Oracle Database supports general query rewrite when the user query contains an inline view, or a subquery in the FROM list. Query rewrite matches inline views in the materialized view with inline views in the request query when the text of the two inline views exactly match. In this case, rewrite treats the matching inline view as it would a named view, and general rewrite processing is possible.

The following is an example where the materialized view contains an inline view, and the query has the same inline view, but the aliases for these views are different:

CREATE MATERIALIZED VIEW inline_example
ENABLE QUERY REWRITE AS
SELECT t.calendar_month_name, t.calendar_year,  p.prod_category, 
       SUM(V1.revenue) AS sum_revenue
FROM times t, products p, 
     (SELECT time_id, prod_id, amount_sold*0.2 AS revenue FROM sales) V1
WHERE t.time_id = V1.time_id AND p.prod_id = V1.prod_id
GROUP BY calendar_month_name, calendar_year, prod_category;

And here is the query that will be rewritten to use the materialized view:

SELECT t.calendar_month_name, t.calendar_year, p.prod_category,
       SUM(X1.revenue) AS sum_revenue
FROM times t, products p,
     (SELECT time_id, prod_id, amount_sold*0.2 AS revenue FROM sales) X1
WHERE t.time_id = X1.time_id AND p.prod_id = X1.prod_id
GROUP BY calendar_month_name, calendar_year, prod_category;

Query Rewrite in the Presence of Duplicate Tables

Oracle accomplishes query rewrite of queries that contain multiple references to the same tables, or selfjoins by employing two different strategies. Using the first strategy, you need to ensure that the query and the materialized view definitions have the same aliases for the multiple references to a table. If you do not provide a matching alias, Oracle will try the second strategy, where the joins in the query and the materialized view are compared to match the multiple references in the query to the multiple references in the materialized view.

The following is an example of a materialized view and a query. In this example, the query is missing a reference to a column in a table so an exact text match will not work. General query rewrite can occur, however, because the aliases for the table references match.

To demonstrate the self-join rewriting possibility with the sh sample schema, the following addition is assumed to include the actual shipping and payment date in the fact table, referencing the same dimension table times. This is for demonstration purposes only and will not return any results:

ALTER TABLE sales ADD (time_id_ship DATE);
ALTER TABLE sales ADD (CONSTRAINT time_id_book_fk FOREIGN key (time_id_ship)
 REFERENCES times(time_id) ENABLE NOVALIDATE);
ALTER TABLE sales MODIFY CONSTRAINT time_id_book_fk RELY;
ALTER TABLE sales ADD (time_id_paid DATE);
ALTER TABLE sales ADD (CONSTRAINT time_id_paid_fk FOREIGN KEY (time_id_paid)
 REFERENCES times(time_id) ENABLE NOVALIDATE);
ALTER TABLE sales MODIFY CONSTRAINT time_id_paid_fk RELY;

Now, you can define a materialized view as follows:

CREATE MATERIALIZED VIEW sales_shipping_lag_mv
ENABLE QUERY REWRITE AS
SELECT t1.fiscal_week_number, s.prod_id,
         t2.fiscal_week_number - t1.fiscal_week_number AS lag
FROM times t1, sales s, times t2 
WHERE t1.time_id = s.time_id AND t2.time_id = s.time_id_ship;

The following query fails the exact text match test but is rewritten because the aliases for the table references match:

SELECT s.prod_id, t2.fiscal_week_number - t1.fiscal_week_number AS lag
FROM times t1, sales s, times t2 
WHERE t1.time_id = s.time_id AND t2.time_id = s.time_id_ship;

Note that Oracle performs other checks to ensure the correct match of an instance of a multiply instanced table in the request query with the corresponding table instance in the materialized view. For instance, in the following example, Oracle correctly determines that the matching alias names used for the multiple instances of table times does not establish a match between the multiple instances of table times in the materialized view.

The following query cannot be rewritten using sales_shipping_lag_mv, even though the alias names of the multiply instanced table time match because the joins are not compatible between the instances of time aliased by t2:

SELECT s.prod_id, t2.fiscal_week_number - t1.fiscal_week_number AS lag
FROM times t1, sales s, times t2 
WHERE t1.time_id = s.time_id AND t2.time_id = s.time_id_paid;

This request query joins the instance of the time table aliased by t2 on the s.time_id_paid column, while the materialized views joins the instance of the times table aliased by t2 on the s.time_id_ship column. Because the join conditions differ, Oracle correctly determines that rewrite cannot occur.

The following query does not have any matching alias in the materialized view, sales_shipping_lag_mv, for the table, times. But query rewrite will now compare the joins between the query and the materialized view and correctly match the multiple instances of times.

SELECT s.prod_id, x2.fiscal_week_number - x1.fiscal_week_number AS lag
FROM times x1, sales s, times x2
WHERE x1.time_id = s.time_id AND x2.time_id = s.time_id_ship;

Query Rewrite Using Date Folding

Date folding rewrite is a specific form of expression matching rewrite. In this type of rewrite, a date range in a query is folded into an equivalent date range representing higher date granules. The resulting expressions representing higher date granules in the folded date range are matched with equivalent expressions in a materialized view. The folding of date range into higher date granules such as months, quarters, or years is done when the underlying datatype of the column is an Oracle DATE. The expression matching is done based on the use of canonical forms for the expressions.

DATE is a built-in datatype which represents ordered time units such as seconds, days, and months, and incorporates a time hierarchy (second -> minute -> hour -> day -> month -> quarter -> year). This hard-coded knowledge about DATE is used in folding date ranges from lower-date granules to higher-date granules. Specifically, folding a date value to the beginning of a month, quarter, year, or to the end of a month, quarter, year is supported. For example, the date value 1-jan-1999 can be folded into the beginning of either year 1999 or quarter 1999-1 or month 1999-01. And, the date value 30-sep-1999 can be folded into the end of either quarter 1999-03 or month 1999-09.

Because date values are ordered, any range predicate specified on date columns can be folded from lower level granules into higher level granules provided the date range represents an integral number of higher level granules. For example, the range predicate date_col >= '1-jan-1999' AND date_col < '30-jun-1999' can be folded into either a month range or a quarter range using the TO_CHAR function, which extracts specific date components from a date value.

The advantage of aggregating data by folded date values is the compression of data achieved. Without date folding, the data is aggregated at the lowest granularity level, resulting in increased disk space for storage and increased I/O to scan the materialized view.

Consider a query that asks for the sum of sales by product types for the year 1998:

SELECT p.prod_category, SUM(s.amount_sold)
FROM sales s, products p
WHERE s.prod_id=p.prod_id AND s.time_id >= TO_DATE('01-jan-1998', 'dd-mon-yyyy')
  AND s.time_id <   TO_DATE('01-jan-1999', 'dd-mon-yyyy')
GROUP BY p.prod_category;

CREATE MATERIALIZED VIEW sum_sales_pcat_monthly_mv
ENABLE QUERY REWRITE AS
SELECT p.prod_category, TO_CHAR(s.time_id,'YYYY-MM') AS month, 
       SUM(s.amount_sold) AS sum_amount
FROM sales s, products p
WHERE s.prod_id=p.prod_id
GROUP BY p.prod_category, TO_CHAR(s.time_id, 'YYYY-MM');

SELECT p.prod_category, SUM(s.amount_sold)
FROM sales s, products p
WHERE s.prod_id=p.prod_id
AND TO_CHAR(s.time_id, 'YYYY-MM') >= '01-jan-1998'
AND TO_CHAR(s.time_id, 'YYYY-MM') < '01-jan-1999'
GROUP BY p.prod_category;

SELECT mv.prod_category, mv.sum_amount
FROM sum_sales_pcat_monthly_mv mv
WHERE month >= '01-jan-1998' AND month < '01-jan-1999';

The range specified in the query represents an integral number of years, quarters, or months. Assume that there is a materialized view mv3 that contains pre-summarized sales by prod_type and is defined as follows:

CREATE MATERIALIZED VIEW mv3
ENABLE QUERY REWRITE AS
SELECT prod_name, TO_CHAR(sales.time_id,'yyyy-mm') AS month, SUM(amount_sold) AS sum_sales
FROM sales, products WHERE sales.prod_id = products.prod_id
GROUP BY prod_name, TO_CHAR(sales_time_id, 'yyyy-mm');

The query can be rewritten by first folding the date range into the month range and then matching the expressions representing the months with the month expression in mv3. This rewrite is shown in two steps (first folding the date range followed by the actual rewrite).

SELECT prod_name, SUM(amount_sold) AS sum_sales
FROM sales, products
WHERE sales.prod_id = products.prod_id AND TO_CHAR(sales.time_id, 'yyyy-mm') >= 
      TO_CHAR('01-jan-1998', 'yyyy-mm') AND TO_CHAR(sales.time_id, '01-jan-1999',
 'yyyy-mm') < TO_CHAR(TO_DATE(''01-jan-1999'', ''dd-mon-yyyy''), ''yyyy-mm'')
GROUP BY prod_name;

SELECT prod_name, sum_sales
FROM mv3 WHERE month >=
      TO_CHAR(TO_DATE('01-jan-1998', 'dd-mon-yyyy'), 'yyyy-mm')
  AND month < TO_CHAR(TO_DATE('01-jan-1999', 'dd-mon-yyyy'), 'yyyy-mm');

If mv3 had pre-summarized sales by prod_name and year instead of prod_name and month, the query could still be rewritten by folding the date range into year range and then matching the year expressions.

Query Rewrite Using View Constraints

Data warehouse applications recognize multi-dimensional cubes in the database by identifying integrity constraints in the relational schema. Integrity constraints represent primary and foreign key relationships between fact and dimension tables. By querying the data dictionary, applications can recognize integrity constraints and hence the cubes in the database. However, this does not work in an environment where database administrators, for schema complexity or security reasons, define views on fact and dimension tables. In such environments, applications cannot identify the cubes properly. By allowing constraint definitions between views, you can propagate base table constraints to the views, thereby allowing applications to recognize cubes even in a restricted environment.

View constraint definitions are declarative in nature, but operations on views are subject to the integrity constraints defined on the underlying base tables, and constraints on views can be enforced through constraints on base tables. Defining constraints on base tables is necessary, not only for data correctness and cleanliness, but also for materialized view query rewrite purposes using the original base objects.

Materialized view rewrite extensively uses constraints for query rewrite. They are used for determining lossless joins, which, in turn, determine if joins in the materialized view are compatible with joins in the query and thus if rewrite is possible.

DISABLE NOVALIDATE is the only valid state for a view constraint. However, you can choose RELY or NORELY as the view constraint state to enable more sophisticated query rewrites. For example, a view constraint in the RELY state allows query rewrite to occur when the query integrity level is set to TRUSTED. Table 18-3 illustrates when view constraints are used for determining lossless joins.

Note that view constraints cannot be used for query rewrite integrity level ENFORCED. This level enforces the highest degree of constraint enforcement ENABLE VALIDATE.

CREATE VIEW time_view AS
SELECT time_id, TO_NUMBER(TO_CHAR(time_id, 'ddd')) AS day_in_year FROM times;

ALTER VIEW time_view ADD (CONSTRAINT time_view_pk 
   PRIMARY KEY (time_id) DISABLE NOVALIDATE);
ALTER VIEW time_view MODIFY CONSTRAINT time_view_pk RELY;
ALTER TABLE sales ADD (CONSTRAINT time_view_fk FOREIGN KEY (time_id)
   REFERENCES time_view(time_id) DISABLE NOVALIDATE);
ALTER TABLE sales MODIFY CONSTRAINT time_view_fk RELY;

CREATE MATERIALIZED VIEW sales_pcat_cal_day_mv 
ENABLE QUERY REWRITE AS
SELECT p.prod_category, t.day_in_year, SUM(s.amount_sold) AS sum_amount_sold
FROM time_view t, sales s, products p 
WHERE t.time_id = s.time_id AND p.prod_id = s.prod_id
GROUP BY p.prod_category, t.day_in_year;

SELECT t.day_in_year, SUM(s.amount_sold) AS sum_amount_sold
FROM time_view t, sales s WHERE t.time_id = s.time_id
GROUP BY t.day_in_year;

SELECT p.prod_category, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p WHERE p.prod_id = s.prod_id
GROUP BY p.prod_category;

ALTER TABLE sales DROP CONSTRAINT time_view_fk;
DROP VIEW time_view;

Query Rewrite Using Set Operator Materialized Views

You can use query rewrite with materialized views that contain set operators. In this case, the query and materialized view do not have to match textually for rewrite to occur. As an example, consider the following materialized view, which uses the postal codes for male customers from San Francisco or Los Angeles:

CREATE MATERIALIZED VIEW cust_male_postal_mv
ENABLE QUERY REWRITE AS
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_gender = 'M' AND c.cust_city = 'San Francisco'
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_gender = 'M' AND c.cust_city = 'Los Angeles';

If you have the following query, which displays the postal codes for male customers from San Francisco or Los Angeles:

SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'Los Angeles' AND c.cust_gender = 'M'
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'San Francisco' AND c.cust_gender = 'M';

The rewritten query will be the following:

SELECT mv.cust_city, mv.cust_postal_code
FROM cust_male_postal_mv mv;

The rewritten query has dropped the UNION ALL and replaced it with the materialized view. Normally, query rewrite has to use the existing set of general eligibility rules to determine if the SELECT subselections under the UNION ALL are equivalent in the query and the materialized view.

If, for example, you have a query that retrieves the postal codes for male customers from San Francisco, Palmdale, or Los Angeles, the same rewrite can occur as in the previous example but query rewrite must keep the UNION ALL with the base tables, as in the following:

SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city= 'Palmdale' AND c.cust_gender ='M'
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'Los Angeles' AND c.cust_gender = 'M'
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'San Francisco' AND c.cust_gender = 'M';

The rewritten query will be:

SELECT mv.cust_city, mv.cust_postal_code
FROM cust_male_postal_mv mv
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'Palmdale' AND c.cust_gender = 'M';

So query rewrite will detect the case where a subset of the UNION ALL can be rewritten using the materialized view cust_male_postal_mv.

UNION, UNION ALL, and INTERSECT are commutative, so query rewrite can rewrite regardless of the order the subselects are found in the query or materialized view. However, MINUS is not commutative. A MINUS B is not equivalent to B MINUS A. Therefore, the order in which the subselects appear under the MINUS operator in the query and the materialized view must be in the same order for rewrite to happen. As an example, consider the case where there exists an old version of the customer table called customer_old and you want to find the difference between the old one and the current customer table only for male customers who live in London. That is, you want to find those customers in the current one that were not in the old one. The following example shows how this is done using a MINUS operator:

SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city= 'Los Angeles' AND c.cust_gender = 'M'
MINUS
SELECT c.cust_city, c.cust_postal_code
FROM customers_old c
WHERE c.cust_city = 'Los Angeles' AND c.cust_gender = 'M';

Switching the subselects would yield a different answer. This illustrates that MINUS is not commutative.

SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_gender = 'M' and c.cust_city = 'San Francisco'
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_gender = 'F' and c.cust_city = 'Los Angeles';

CREATE MATERIALIZED VIEW cust_postal_mv
ENABLE QUERY REWRITE AS
SELECT 1 AS marker, c.cust_gender, c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'Los Angeles'
UNION ALL
SELECT 2 AS marker, c.cust_gender, c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city = 'San Francisco';

SELECT mv.cust_city, mv.cust_postal_code
FROM cust_postal_mv mv
WHERE mv.marker = 2 AND mv.cust_gender = 'M'
UNION ALL
SELECT mv.cust_city, mv.cust_postal_code
FROM cust_postal_mv mv
WHERE mv.marker = 1 AND mv.cust_gender = 'F';

SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city= 'Palmdale' AND c.cust_gender ='M'
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_gender = 'M' and c.cust_city = 'San Francisco'
UNION ALL
SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_gender = 'F' and c.cust_city = 'Los Angeles';

SELECT c.cust_city, c.cust_postal_code
FROM customers c
WHERE c.cust_city= 'Palmdale' AND c.cust_gender ='M'
UNION ALL
SELECT mv.cust_city, mv.cust_postal_code

FROM cust_postal_mv mv
WHERE mv.marker = 2 AND mv.cust_gender = 'M'
UNION ALL
  SELECT mv.cust_city, mv.cust_postal_code
  FROM cust_postal_mv mv
  WHERE mv.marker = 1 AND mv.cust_gender = 'F';

Query Rewrite in the Presence of Grouping Sets

This section discusses various considerations for using query rewrite with grouping sets.

SELECT p.prod_subcategory, t.calendar_month_desc, c.cust_city, 
  SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, customers c, products p, times t
WHERE s.time_id=t.time_id AND s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY GROUPING SETS ((p.prod_subcategory, t.calendar_month_desc),
   (c.cust_city, p.prod_subcategory));

SELECT p.prod_subcategory, t.calendar_month_desc, c.cust_city, 
  SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, customers c, products p, times t
WHERE s.time_id=t.time_id AND s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY GROUPING SETS 
((p.prod_subcategory, t.calendar_month_desc), 
   (c.cust_city, p.prod_subcategory));

SELECT mv.prod_subcategory, mv.calendar_month_desc, mv.cust_city,
  SUM(mv.sum_amount_sold) AS sum_amount_sold
FROM sum_sales_pscat_month_city_mv mv
GROUP BY GROUPING SETS 
((mv.prod_subcategory, mv.calendar_month_desc), 
  (mv.cust_city, mv.prod_subcategory));

CREATE MATERIALIZED VIEW sum_grouping_set_mv
ENABLE QUERY REWRITE AS
SELECT p.prod_category, p.prod_subcategory, c.cust_state_province, c.cust_city,
  GROUPING_ID(p.prod_category,p.prod_subcategory,
              c.cust_state_province,c.cust_city) AS gid,
  SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY GROUPING SETS 
((p.prod_category, p.prod_subcategory, c.cust_city), 
  (p.prod_category, p.prod_subcategory, c.cust_state_province, c.cust_city),
  (p.prod_category, p.prod_subcategory));

SELECT p.prod_subcategory, c.cust_city, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY p.prod_subcategory, c.cust_city;

SELECT prod_subcategory, cust_city, SUM(sum_amount_sold) AS sum_amount_sold
FROM sum_grouping_set_mv
WHERE gid = grouping identifier of (prod_category,prod_subcategory, cust_city)
GROUP BY prod_subcategory, cust_city;

SELECT p.prod_category, p.prod_subcategory, c.cust_city,
       SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY GROUPING SETS 
((p.prod_category, p.prod_subcategory, c.cust_city), 
  (p.prod_category, p.prod_subcategory));

SELECT prod_subcategory, cust_city, sum_amount_sold
FROM sum_grouping_set_mv
WHERE gid = grouping identifier of (prod_category,prod_subcategory, cust_city)
   OR gid = grouping identifier of (prod_category,prod_subcategory)

SELECT p.prod_category, p.prod_subcategory, c.cust_state_province, 
  t.calendar_month_desc, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c, times t
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY GROUPING SETS 
((p.prod_subcategory, t.calendar_month_desc), 
  (t.calendar_month_desc), 
  (p.prod_category, p.prod_subcategory, c.cust_state_province), 
  (p.prod_category, p.prod_subcategory));

SELECT null, p.prod_subcategory, null,
    t.calendar_month_desc, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c, times t
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY p.prod_subcategory, t.calendar_month_desc
UNION ALL
  SELECT null, null, null,
    t.calendar_month_desc, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c, times t
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY t.calendar_month_desc
UNION ALL
SELECT p.prod_category, p.prod_subcategory, c.cust_state_province,
    null, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c, times t
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY p.prod_category, p.prod_subcategory, c.cust_state_province
UNION ALL 
  SELECT p.prod_category, p.prod_subcategory, null,
    null, SUM(s.amount_sold) AS sum_amount_sold
  FROM sales s, products p, customers c, times t
  WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
  GROUP BY p.prod_category, p.prod_subcategory;

SELECT null, p.prod_subcategory, null,
    t.calendar_month_desc, SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c, times t
WHERE s.prod_id = p.prod_id AND s.cust_id = c.cust_id
GROUP BY GROUPING SETS
  ((p.prod_subcategory, t.calendar_month_desc),
    (t.calendar_month_desc),)
UNION ALL
  SELECT prod_category, prod_subcategory, cust_state_province,
    null, SUM(sum_amount_sold) AS sum_amount_sold
  FROM sum_grouping_set_mv
  WHERE gid = <grouping id of (prod_category,prod_subcategory, cust_city)>
  GROUP BY p.prod_category, p.prod_subcategory, c.cust_state_province
UNION ALL 
  SELECT prod_category, prod_subcategory, null,
    null, sum_amount_sold
  FROM sum_grouping_set_mv 
  WHERE gid = <grouping id of (prod_category,prod_subcategory)>

Query Rewrite in the Presence of Window Functions

Window functions are used to compute cumulative, moving and centered aggregates. These functions work with the following aggregates: SUM, AVG, MIN/MAX., COUNT, VARIANCE, STDDEV, FIRST_VALUE, and LAST_VALUE. A query with window function can be rewritten using exact text match rewrite. This requires that the materialized view definition also matches the query exactly. When there is no window function on the materialized view, then a query with a window function can be rewritten provided the aggregate in the query is found in the materialized view and all other eligibility checks such as the join computability checks are successful. A window function on the query is compared to the window function in the materialized view using its canonical form format. This enables query rewrite to rewrite even complex window functions.

When a query with a window function requires rollup during query rewrite, query rewrite will, whenever possible, split the query into an inner query with the aggregate and an outer query with the windowing function. This will permit query rewrite to rewrite the aggregate in the inner query before applying the window function. One exception is when the query has both a window function and grouping sets. In this case presence of the grouping set will prevent query rewrite from splitting the query so query rewrite will not take place in this case.

Query Rewrite and Expression Matching

An expression that appears in a query can be replaced with a simple column in a materialized view provided the materialized view column represents a precomputed expression that matches with the expression in the query. If a query can be rewritten to use a materialized view, it will be faster. This is because materialized views contain precomputed calculations and do not need to perform expression computation.

The expression matching is done by first converting the expressions into canonical forms and then comparing them for equality. Therefore, two different expressions will generally be matched as long as they are equivalent to each other. Further, if the entire expression in a query fails to match with an expression in a materialized view, then subexpressions of it are tried to find a match. The subexpressions are tried in a top-down order to get maximal expression matching.

Consider a query that asks for sum of sales by age brackets (1-10, 11-20, 21-30, and so on).

CREATE MATERIALIZED VIEW sales_by_age_bracket_mv
ENABLE QUERY REWRITE AS
SELECT TO_CHAR((2000-c.cust_year_of_birth)/10-0.5,999) AS age_bracket,
       SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, customers c WHERE s.cust_id=c.cust_id
GROUP BY TO_CHAR((2000-c.cust_year_of_birth)/10-0.5,999);

The following query rewrites, using expression matching:

SELECT TO_CHAR(((2000-c.cust_year_of_birth)/10)-0.5,999), SUM(s.amount_sold) 
FROM sales s, customers c WHERE s.cust_id=c.cust_id
GROUP BY TO_CHAR((2000-c.cust_year_of_birth)/10-0.5,999);

This query is rewritten in terms of sales_by_age_bracket_mv based on the matching of the canonical forms of the age bracket expressions (that is, 2000 - c.cust_year_of_birth)/10-0.5), as follows:

SELECT age_bracket, sum_amount_sold FROM sales_by_age_bracket_mv;

PARTITION BY RANGE (time_id)
(PARTITION SALES_Q1_1998 
           VALUES LESS THAN (TO_DATE('01-APR-1998', 'DD-MON-YYYY')),
 PARTITION SALES_Q2_1998 
           VALUES LESS THAN (TO_DATE('01-JUL-1998', 'DD-MON-YYYY')),
 PARTITION SALES_Q3_1998 
           VALUES LESS THAN (TO_DATE('01-OCT-1998', 'DD-MON-YYYY')), 
...

CREATE MATERIALIZED VIEW sum_sales_per_city_mv
ENABLE QUERY REWRITE AS
SELECT s.time_id, p.prod_subcategory, c.cust_city,
       SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c
WHERE s.cust_id = c.cust_id AND s.prod_id = p.prod_id
GROUP BY time_id, prod_subcategory, cust_city;

INSERT INTO SALES VALUES(17, 10, '01-DEC-2000', 4, 380, 123.45, 54321);

SELECT s.time_id, p.prod_subcategory, c.cust_city,
       SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c
WHERE s.cust_id = c.cust_id AND s.prod_id = p.prod_id
AND   s.time_id < TO_DATE('01-OCT-2000','DD-MON-YYYY') 
OR s.time_id >= TO_DATE('01-OCT-2001','DD-MON-YYYY'))
GROUP BY time_id, prod_subcategory, cust_city;

SELECT s.time_id, p.prod_subcategory, c.cust_city,
       SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c
WHERE s.cust_id = c.cust_id AND s.prod_id = p.prod_id
AND   s.time_id BETWEEN TO_DATE('01-JAN-2000', 'DD-MON-YYYY')
AND TO_DATE('01-JUL-2000', 'DD-MON-YYYY')
GROUP BY time_id, prod_subcategory, cust_city;

SELECT time_id, prod_subcategory, cust_city, sum_amount_sold
FROM sum_sales_per_city_mv
WHERE time_id BETWEEN TO_DATE('01-JAN-2000', 'DD-MON-YYYY')
AND TO_DATE('01-JUL-2000', 'DD-MON-YYYY');

CREATE MATERIALIZED VIEW sum_sales_per_city_2_mv
ENABLE QUERY REWRITE AS
SELECT DBMS_MVIEW.PMARKER(s.rowid) AS pmarker,
       t.fiscal_quarter_desc, p.prod_subcategory, c.cust_city,
       SUM(s.amount_sold) AS sum_amount_sold
FROM sales s, products p, customers c, times t
WHERE s.cust_id = c.cust_id AND s.prod_id = p.prod_id
AND   s.time_id = t.time_id
GROUP BY DBMS_MVIEW.PMARKER(s.rowid),
    p.prod_subcategory, c.cust_city, t.fiscal_quarter_desc;

INSERT INTO SALES VALUES(17, 10, '01-DEC-2000', 4, 380, 123.45, 54321);

SELECT p.prod_subcategory, c.cust_city, SUM(s.amount_sold) AS sum_amount_sold 
FROM sales s, products p, customers c, times t
WHERE s.cust_id = c.cust_id AND s.prod_id = p.prod_id AND s.time_id = t.time_id
AND c.cust_city= 'Nuernberg' 
AND s.time_id >=TO_DATE('01-JAN-2000','dd-mon-yyyy')
AND s.time_id <  TO_DATE('01-APR-2000','dd-mon-yyyy') 
GROUP BY prod_subcategory, cust_city;

SELECT mv.prod_subcategory, mv.cust_city, SUM(mv.sum_amount_sold) 
FROM sum_sales_per_city_2_mv mv 
WHERE mv.pmarker = 31070 AND mv.cust_city= 'Nuernberg'
GROUP BY prod_subcategory, cust_city; 

Cursor Sharing and Bind Variables

Query rewrite is supported when the query contains user bind variables as long as the actual bind values are not required during query rewrite. If the actual values of the bind variables are required during query rewrite, then we say that query rewrite is dependent on the bind values. Because the user bind variables are not available during query rewrite time, if query rewrite is dependent on the bind values, it is not possible to rewrite the query. For example, consider the following materialized view, customer_mv, which has the predicate, (customer_id >= 1000), in the WHERE clause:

CREATE MATERIALIZED VIEW customer_mv
ENABLE QUERY REWRITE AS
SELECT cust_id, prod_id,  SUM(amount_sold) AS total_amount
FROM sales WHERE cust_id >= 1000
GROUP BY cust_id, prod_id;

Consider the following query, which has a user bind variable, :user_id, in its WHERE clause:

SELECT cust_id, prod_id, SUM(amount_sold) AS sum_amount
FROM sales WHERE cust_id > :user_id
GROUP BY cust_id, prod_id;

Because the materialized view, customer_mv, has a selection in its WHERE clause, query rewrite is dependent on the actual value of the user bind variable, user_id, to compute the containment. Because user_id is not available during query rewrite time and query rewrite is dependent on the bind value of user_id, this query cannot be rewritten.

Even though the preceding example has a user bind variable in the WHERE clause, the same is true regardless of where the user bind variable appears in the query. In other words, irrespective of where a user bind variable appears in a query, if query rewrite is dependent on its value, then the query cannot be rewritten.

Now consider the following query which has a user bind variable, :user_id, in its SELECT list:

SELECT cust_id + :user_id, prod_id, SUM(amount_sold) AS total_amount
FROM sales WHERE cust_id >= 2000
GROUP BY cust_id, prod_id;

Because the value of the user bind variable, user_id, is not required during query rewrite time, the preceding query will rewrite.

SELECT cust_id + :user_id, prod_id, total_amount
FROM customer_mv;

Handling Expressions in Query Rewrite

Rewrite with some expressions is also supported when the expression evaluates to a constant, such as TO_DATE('12-SEP-1999','DD-Mon-YYYY'). For example, if an existing materialized view is defined as:

CREATE MATERIALIZED VIEW sales_on_valentines_day_99_mv
BUILD IMMEDIATE
REFRESH FORCE
ENABLE QUERY REWRITE AS
SELECT s.prod_id, s.cust_id, s.amount_sold
FROM times t, sales s WHERE s.time_id = t.time_id
AND t.time_id = TO_DATE('14-FEB-1999', 'DD-MON-YYYY');

Then the following query can be rewritten:

SELECT s.prod_id, s.cust_id, s.amount_sold
FROM sales s, times t WHERE s.time_id = t.time_id
AND t.time_id = TO_DATE('14-FEB-1999', 'DD-MON-YYYY');

This query would be rewritten as follows:

SELECT * FROM sales_on_valentines_day_99_mv;

Using EXPLAIN PLAN with Query Rewrite

The EXPLAIN PLAN facility is used as described in Oracle Database SQL Reference. For query rewrite, all you need to check is that the operation shows MAT_VIEW REWRITE ACCESS. If it does, then query rewrite has occurred. An example is the following, which creates the materialized view cal_month_sales_mv:

CREATE MATERIALIZED VIEW cal_month_sales_mv
ENABLE QUERY REWRITE AS
SELECT  t.calendar_month_desc, SUM(s.amount_sold) AS dollars
FROM sales s, times t WHERE s.time_id = t.time_id
GROUP BY t.calendar_month_desc;

If EXPLAIN PLAN is used on the following SQL statement, the results are placed in the default table PLAN_TABLE. However, PLAN_TABLE must first be created using the utlxplan.sql script. Note that EXPLAIN PLAN does not actually execute the query.

EXPLAIN PLAN FOR
SELECT  t.calendar_month_desc, SUM(s.amount_sold)
FROM  sales s, times t WHERE s.time_id = t.time_id
GROUP BY t.calendar_month_desc; 

For the purposes of query rewrite, the only information of interest from PLAN_TABLE is the operation OBJECT_NAME, which identifies the method used to execute this query. Therefore, you would expect to see the operation MAT_VIEW REWRITE ACCESS in the output as illustrated in the following:

SELECT OPERATION, OBJECT_NAME FROM PLAN_TABLE;

OPERATION                  OBJECT_NAME
--------------------       -----------
SELECT STATEMENT
MAT_VIEW REWRITE ACCESS    CALENDAR_MONTH_SALES_MV

Using the EXPLAIN_REWRITE Procedure with Query Rewrite

It can be difficult to understand why a query did not rewrite. The rules governing query rewrite eligibility are quite complex, involving various factors such as constraints, dimensions, query rewrite integrity modes, freshness of the materialized views, and the types of queries themselves. In addition, you may want to know why query rewrite chose a particular materialized view instead of another. To help with this matter, Oracle provides the DBMS_MVIEW.EXPLAIN_REWRITE procedure to advise you when a query can be rewritten and, if not, why not. Using the results from DBMS_MVIEW.EXPLAIN_REWRITE, you can take the appropriate action needed to make a query rewrite if at all possible.

Note that the query specified in the EXPLAIN_REWRITE statement does not actually execute.

DBMS_MVIEW.EXPLAIN_REWRITE (
    query           VARCHAR2,
    mv              VARCHAR2(30),
    statement_id    VARCHAR2(30));

DBMS_MVIEW.EXPLAIN_REWRITE (
    query           [VARCHAR2 | CLOB],
    mv               VARCHAR2(30),
    output_array     SYS.RewriteArrayType);

CREATE TABLE REWRITE_TABLE(
  statement_id          VARCHAR2(30),   -- id for the query
  mv_owner              VARCHAR2(30),   -- owner of the MV
  mv_name               VARCHAR2(30),   -- name of the MV
  sequence              INTEGER,        -- sequence no of the msg
  query                 VARCHAR2(2000), -- user query
  query_block_no        INTEGER,        -- block no of the current subquery
  rewritten_txt         VARCHAR2(2000), -- rewritten query
  message               VARCHAR2(512),  -- EXPLAIN_REWRITE msg
  pass                  VARCHAR2(3),    -- rewrite pass no
  mv_in_msg             VARCHAR2(30),   -- MV in current message 
  measure_in_msg        VARCHAR2(30),   -- Measure in current message 
  join_back_tbl         VARCHAR2(30),   -- Join back table in message 
  join_back_col         VARCHAR2(30),   -- Join back column in message
  original_cost         INTEGER,        -- Cost of original query
  rewritten_cost        INTEGER,        -- Cost of rewritten query
  flags                 INTEGER,        -- associated flags
  reserved1             INTEGER,        -- currently not used 
  reerved2              VARCHAR2(10))   -- currently not used;

EXECUTE DBMS_MVIEW.EXPLAIN_REWRITE -
('SELECT p.prod_name, SUM(amount_sold) ' || -
'FROM sales s, products p ' || -
'WHERE s.prod_id = p.prod_id ' || -
' AND prod_name > ''B%'' ' || -
' AND prod_name < ''C%'' ' || -
'GROUP BY prod_name', -
'TestXRW.PRODUCT_SALES_MV', -
'SH');

SELECT message FROM rewrite_table ORDER BY sequence;
MESSAGE                                                                         
--------------------------------------------------------------------------------
QSM-01033: query rewritten with materialized view, PRODUCT_SALES_MV             
1 row selected.

DECLARE
  qrytext VARCHAR2(500)  :='SELECT cust_first_name, cust_last_name,
 SUM(amount_sold) AS dollar_sales FROM sales s, customers c WHERE s.cust_id=
 c.cust_id GROUP BY cust_first_name, cust_last_name';
    idno    VARCHAR2(30) :='ID1';
BEGIN
  DBMS_MVIEW.EXPLAIN_REWRITE(qrytext, '', idno);
END;
/
SELECT message FROM rewrite_table ORDER BY sequence;

SQL> MESSAGE                                      
--------------------------------------------------------------------------------
QSM-01082: Joining materialized view, CAL_MONTH_SALES_MV, with table, SALES, not possible
QSM-01022: a more optimal materialized view than PRODUCT_SALES_MV was used to rewrite
QSM-01022: a more optimal materialized view than FWEEK_PSCAT_SALES_MV was used to rewrite
QSM-01033: query rewritten with materialized view, SALES_MV

TYPE RewriteMessage IS OBJECT(
  mv_owner        VARCHAR2(30),   -- MV's schema
  mv_name         VARCHAR2(30),   -- Name of the MV
  sequence        NUMBER(3),      -- sequence no of the msg 
  query_text      VARCHAR2(2000), -- User query
  query_block_no  NUMBER(3),      -- block no of the current subquery
  rewritten_text  VARCHAR2(2000), -- rewritten query text 
  message         VARCHAR2(512),  -- EXPLAIN_REWRITE error msg
  pass            VARCHAR2(3),    -- Query rewrite pass no 
  mv_in_msg       VARCHAR2(30),   -- MV in current message 
  measure_in_msg  VARCHAR2(30),   -- Measure in current message 
  join_back_tbl   VARCHAR2(30),   -- Join back table in current msg 
  join_back_col   VARCHAR2(30),   -- Join back column in current msg 
  original_cost   NUMBER(10),     -- Cost of original query 
  rewritten_cost  NUMBER(10),     -- Cost rewritten query 
  flags           NUMBER,         -- Associated flags
  reserved1       NUMBER,         -- For future use
  reserved2       VARCHAR2(10)    -- For future use
);

CREATE MATERIALIZED VIEW avg_sales_city_state_mv
ENABLE QUERY REWRITE AS
SELECT c.cust_city, c.cust_state_province, AVG(s.amount_sold)
FROM sales s, customers c WHERE s.cust_id = c.cust_id
GROUP BY c.cust_city, c.cust_state_province;

SELECT c.cust_state_province, AVG(s.amount_sold)
FROM sales s, customers c WHERE s.cust_id = c.cust_id
GROUP BY c.cust_state_province;

SET SERVEROUTPUT ON
DECLARE
  Rewrite_Array SYS.RewriteArrayType := SYS.RewriteArrayType();
  querytxt VARCHAR2(1500) := 'SELECT c.cust_state_province,
  AVG(s.amount_sold)
   FROM sales s, customers c WHERE s.cust_id = c.cust_id
   GROUP BY c.cust_state_province';
  i NUMBER;
BEGIN
  DBMS_MVIEW.EXPLAIN_REWRITE(querytxt, 'AVG_SALES_CITY_STATE_MV',
  Rewrite_Array);
  FOR i IN 1..Rewrite_Array.count
  LOOP
    DBMS_OUTPUT.PUT_LINE(Rewrite_Array(i).message);
  END LOOP;
END;
/

QSM-01065: materialized view, AVG_SALES_CITY_STATE_MV, cannot compute
  measure, AVG, in the query
QSM-01101: rollup(s) took place on mv, AVG_SALES_CITY_STATE_MV
QSM-01053: NORELY referential integrity constraint on table, CUSTOMERS,
  in TRUSTED/STALE TOLERATED integrity mode
PL/SQL procedure successfully completed.

DBMS_MVIEW.EXPLAIN_REWRITE(
   query          IN CLOB,
   mv             IN VARCHAR2,
   statement_id   IN VARCHAR2);

DBMS_MVIEW.EXPLAIN_REWRITE(
   query           IN CLOB,
   mv              IN VARCHAR2,
   msg_array       IN OUT SYS.RewriteArrayType);
 

DBMS_MVIEW.EXPLAIN_REWRITE(query_txt, 'mv1, mv2, mv3')

QSM-01127: query rewritten with materialized view(s), mv1, mv2, and mv3.

SYS.XRW(list_of_mvs, list_of_commands, query_text),

QUERY_TXT:      User query text
REWRITTEN_TXT:  Rewritten query text
QUERY_BLOCK_NO: Query block number to identify each query blocks in
                case the query has subqueries or inline views
PASS:           Pass indicates whether a given message was generated
                before or after the view merging process of query rewrite.
COSTS:          Costs indicates the estimated execution cost of the
                original query and the rewritten query

DROP MATERIALIZED VIEW month_sales_mv;
 
CREATE MATERIALIZED VIEW month_sales_mv
  ENABLE QUERY REWRITE
  AS
  SELECT t.calendar_month_number, SUM(s.amount_sold) AS sum_dollars
  FROM sales s, times t
  WHERE s.time_id = t.time_id
GROUP BY t.calendar_month_number;
 
SET SERVEROUTPUT ON
DECLARE
  querytxt VARCHAR2(1500) := 'SELECT t.calendar_month_number,
         SUM(s.amount_sold) AS sum_dollars FROM sales s, times t
  WHERE s.time_id = t.time_id GROUP BY t.calendar_month_number';
BEGIN
  SYS.XRW('MONTH_SALES_MV', 'COSTS, PASS, REWRITTEN_TXT, QUERY_BLOCK_NO', querytxt);
END;
/

============================================================================
>> MESSAGE  : QSM-01151: query was rewritten
>> RW QUERY : SELECT MONTH_SALES_MV.CALENDAR_MONTH_NUMBER CALENDAR_MONTH_NUMBER,
MONTH_SALES_MV.SUM_DOLLARS SUM_DOLLARS FROM SH.MONTH_SALES_MV MONTH_SALES_MV
>> ORIG COST: 19.952763130792                  RW COST: 1.80687108
============================================================================
>>
------------------------- ANALYSIS OF QUERY REWRITE -------------------------
>>
>> QRY BLK #: 0
>> MESSAGE  : QSM-01209: query rewritten with materialized view, 
   MONTH_SALES_MV, using text match algorithm
>> RW QUERY : SELECT MONTH_SALES_MV.CALENDAR_MONTH_NUMBER CALENDAR_MONTH_NUMBER,
   MONTH_SALES_MV.SUM_DOLLARS SUM_DOLLARS FROM SH.MONTH_SALES_MV MONTH_SALES_MV
>> ORIG COST: 19.952763130792                  RW COST: 1.80687108
>> MESSAGE OUTPUT BEFORE VIEW MERGING...
============================ END OF MESSAGES ===============================
PL/SQL procedure successfully completed.

Query Rewrite Considerations: Constraints

Make sure all inner joins referred to in a materialized view have referential integrity (foreign key/primary key constraints) with additional NOT NULL constraints on the foreign key columns. Since constraints tend to impose a large overhead, you could make them NO VALIDATE and RELY and set the parameter QUERY_REWRITE_INTEGRITY to STALE_TOLERATED or TRUSTED. However, if you set QUERY_REWRITE_INTEGRITY to ENFORCED, all constraints must be enabled, enforced, and validated to get maximum rewritability.

You should avoid using the ON DELETE clause as it can lead to unexpected results.

Query Rewrite Considerations: Dimensions

You can express the hierarchical relationships and functional dependencies in normalized or denormalized dimension tables using the HIERARCHY and DETERMINES clauses of a dimension. Dimensions can express intra-table relationships which cannot be expressed by constraints. Set the parameter QUERY_REWRITE_INTEGRITY to TRUSTED or STALE_TOLERATED for query rewrite to take advantage of the relationships declared in dimensions.

Query Rewrite Considerations: Outer Joins

Another way of avoiding constraints is to use outer joins in the materialized view. Query rewrite will be able to derive an inner join in the query, such as (A.a=B.b), from an outer join in the materialized view (A.a = B.b(+)), as long as the rowid of B or column B.b is available in the materialized view. Most of the support for rewrites with outer joins is provided for materialized views with joins only. To exploit it, a materialized view with outer joins should store the rowid or primary key of the inner table of an outer join. For example, the materialized view join_sales_time_product_mv_oj stores the primary keys prod_id and time_id of the inner tables of outer joins.

Query Rewrite Considerations: Text Match

If you need to speed up an extremely complex, long-running query, you could create a materialized view with the exact text of the query. Then the materialized view would contain the query results, thus eliminating the time required to perform any complex joins and search through all the data for that which is required.

Query Rewrite Considerations: Aggregates

To get the maximum benefit from query rewrite, make sure that all aggregates which are needed to compute ones in the targeted set of queries are present in the materialized view. The conditions on aggregates are quite similar to those for incremental refresh. For instance, if AVG(x) is in the query, then you should store COUNT(x) and AVG(x) or store SUM(x) and COUNT(x) in the materialized view. See "General Restrictions on Fast Refresh" for fast refresh requirements.

Query Rewrite Considerations: Grouping Conditions

Aggregating data at lower levels in the hierarchy is better than aggregating at higher levels because lower levels can be used to rewrite more queries. Note, however, that doing so will also take up more space. For example, instead of grouping on state, group on city (unless space constraints prohibit it).

Instead of creating multiple materialized views with overlapping or hierarchically related GROUP BY columns, create a single materialized view with all those GROUP BY columns. For example, instead of using a materialized view that groups by city and another materialized view that groups by month, use a single materialized view that groups by city and month.

Use GROUP BY on columns that correspond to levels in a dimension but not on columns that are functionally dependent, because query rewrite will be able to use the functional dependencies automatically based on the DETERMINES clause in a dimension. For example, instead of grouping on prod_name, group on prod_id (as long as there is a dimension which indicates that the attribute prod_id determines prod_name, you will enable the rewrite of a query involving prod_name).

Query Rewrite Considerations: Expression Matching

If several queries share the same common subselect, it is advantageous to create a materialized view with the common subselect as one of its SELECT columns. This way, the performance benefit due to precomputation of the common subselect can be obtained across several queries.

Query Rewrite Considerations: Date Folding

When creating a materialized view that aggregates data by folded date granules such as months or quarters or years, always use the year component as the prefix but not as the suffix. For example, TO_CHAR(date_col, 'yyyy-q') folds the date into quarters, which collate in year order, whereas TO_CHAR(date_col, 'q-yyyy') folds the date into quarters, which collate in quarter order. The former preserves the ordering while the latter does not. For this reason, any materialized view created without a year prefix will not be eligible for date folding rewrite.

Query Rewrite Considerations: Statistics

Optimization with materialized views is based on cost and the optimizer needs statistics of both the materialized view and the tables in the query to make a cost-based choice. Materialized views should thus have statistics collected using the DBMS_STATS package.

Query Rewrite Considerations: Hints

This section discusses the following considerations:

• REWRITE and NOREWRITE Hints

• REWRITE_OR_ERROR Hint

• Multiple Materialized View Rewrite Hints

• EXPAND_GSET_TO_UNION Hint

SELECT /*+ NOREWRITE */ p.prod_subcategory, SUM(s.amount_sold)
FROM   sales s, products p WHERE  s.prod_id = p.prod_id
GROUP BY p.prod_subcategory;

SELECT /*+ REWRITE (sum_sales_pscat_week_mv) */ 
       p.prod_subcategory,  SUM(s.amount_sold)
FROM   sales s, products p WHERE  s.prod_id=p.prod_id
GROUP BY p.prod_subcategory;

ORA-30393: a query block in the statement did not rewrite

SELECT /*+ REWRITE_OR_ERROR */ p.prod_subcategory, SUM(s.amount_sold)
FROM sales s, products p WHERE s.prod_id = p.prod_id
GROUP BY p.prod_subcategory;