9 Advanced Materialized Views

Partition Change Tracking

It is possible and advantageous to track freshness to a finer grain than the entire materialized view. The ability to identify which rows in a materialized view are affected by a certain detail table partition, is known as Partition Change Tracking (PCT). When one or more of the detail tables are partitioned, it may be possible to identify the specific rows in the materialized view that correspond to a modified detail partition(s); those rows become stale when a partition is modified while all other rows remain fresh.

You can use PCT to identify which materialized view rows correspond to a particular partition. PCT is also used to support fast refresh after partition maintenance operations on detail tables. For instance, if a detail table partition is truncated or dropped, the affected rows in the materialized view are identified and deleted.

Identifying which materialized view rows are fresh or stale, rather than considering the entire materialized view as stale, allows query rewrite to use those rows that are fresh while in QUERY_REWRITE_INTEGRITY=ENFORCED or TRUSTED modes. Oracle does not rewrite against partial stale materialized view if partition change tracking on the changed table is enabled by the presence of join dependent expression in the materialized view. See "Join Dependent Expression".

To support PCT, a materialized view must satisfy the following requirements:

• At least one of the detail tables referenced by the materialized view must be partitioned.

• Partitioned tables must use either range, list or composite partitioning.

• The top level partition key must consist of only a single column.

• The materialized view must contain either the partition key column or a partition marker or ROWID or join dependent expression of the detail table. See PL/SQL Packages and Types Reference for details regarding the DBMS_MVIEW.PMARKER function.

• If you use a GROUP BY clause, the partition key column or the partition marker or ROWID or join dependent expression must be present in the GROUP BY clause.

• If you use an analytic window function or the MODEL clause, the partition key column or the partition marker or ROWID or join dependent expression must be present in their respective PARTITION BY subclauses.

• Data modifications can only occur on the partitioned table. If PCT refresh is being done for a table which has join dependent expression in the materialized view, then data modifications should not have occurred in any of the join dependent tables.

• The COMPATIBILITY initialization parameter must be a minimum of 9.0.0.0.0.

• PCT is not supported for a materialized view that refers to views, remote tables, or outer joins.

• PCT-based refresh is not supported for UNION ALL materialized views.

CREATE MATERIALIZED VIEW LOG ON SALES WITH ROWID
   (prod_id, time_id, quantity_sold, amount_sold) INCLUDING NEW VALUES;

CREATE MATERIALIZED VIEW LOG ON PRODUCTS WITH ROWID
   (prod_id, prod_name, prod_desc) INCLUDING NEW VALUES;
CREATE MATERIALIZED VIEW LOG ON TIMES WITH ROWID
   (time_id, calendar_month_name, calendar_year) INCLUDING NEW VALUES;

CREATE MATERIALIZED VIEW cust_dly_sales_mv
BUILD DEFERRED REFRESH FAST ON DEMAND
ENABLE QUERY REWRITE AS
SELECT s.time_id, p.prod_id, p.prod_name, COUNT(*),
         SUM(s.quantity_sold), SUM(s.amount_sold),
         COUNT(s.quantity_sold), COUNT(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 s.time_id, p.prod_id, p.prod_name;

SELECT s.time_id, t.calendar_month_name 
FROM sales s, times t WHERE s.time_id = t.time_id;

SELECT s.time_id, y.calendar_year_name
FROM sales s, times_d d, times_m m, times_y y
WHERE s.time_id = d.time_id AND d.day_id = m.day_id AND m.mon_id = y.mon_id;

CREATE MATERIALIZED VIEW cust_mth_sales_mv
BUILD DEFERRED REFRESH FAST ON DEMAND
ENABLE QUERY REWRITE AS
SELECT  t.calendar_month_name, p.prod_id, p.prod_name, COUNT(*),
    SUM(s.quantity_sold), SUM(s.amount_sold),
    COUNT(s.quantity_sold), COUNT(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 t.calendar_month_name, p.prod_id, p.prod_name;

CREATE MATERIALIZED VIEW prod_yr_sales_mv
BUILD DEFERRED
REFRESH FAST ON DEMAND
ENABLE QUERY REWRITE AS
SELECT DBMS_MVIEW.PMARKER(p.rowid), p.prod_category, t.calendar_year, COUNT(*),
       SUM(s.amount_sold), SUM(s.quantity_sold),
       COUNT(s.amount_sold), COUNT(s.quantity_sold)
FROM   sales s, products p, times t
WHERE  s.time_id = t.time_id AND s.prod_id = p.prod_id
GROUP BY  DBMS_MVIEW.PMARKER (p.rowid), p.prod_category, t.calendar_year;

Partitioning a Materialized View

Partitioning a materialized view involves defining the materialized view with the standard Oracle partitioning clauses, as illustrated in the following example. This statement creates a materialized view called part_sales_mv, which uses three partitions, can be fast refreshed, and is eligible for query rewrite.

CREATE MATERIALIZED VIEW part_sales_mv
PARALLEL PARTITION BY RANGE (time_id)
(PARTITION month1
      VALUES LESS THAN (TO_DATE('31-12-1998', 'DD-MM-YYYY'))
      PCTFREE 0 PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
      TABLESPACE sf1,
 PARTITION month2
      VALUES LESS THAN (TO_DATE('31-12-1999', 'DD-MM-YYYY'))
      PCTFREE 0 PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
      TABLESPACE sf2,
 PARTITION month3
      VALUES LESS THAN (TO_DATE('31-12-2000', 'DD-MM-YYYY'))
      PCTFREE 0 PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
      TABLESPACE sf3)
BUILD DEFERRED
REFRESH FAST
ENABLE QUERY REWRITE AS
SELECT s.cust_id, s.time_id,
  SUM(s.amount_sold) AS sum_dol_sales, SUM(s.quantity_sold) AS sum_unit_sales
  FROM sales s GROUP BY s.time_id, s.cust_id;

Partitioning a Prebuilt Table

Alternatively, a materialized view can be registered to a partitioned prebuilt table as illustrated in the following example:

CREATE TABLE part_sales_tab(time_id, cust_id, sum_dollar_sales, sum_unit_sale)
PARALLEL PARTITION BY RANGE (time_id)
(PARTITION month1
      VALUES LESS THAN (TO_DATE('31-12-1998', 'DD-MM-YYYY'))
      PCTFREE 0 PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
      TABLESPACE sf1,
 PARTITION month2
      VALUES LESS THAN (TO_DATE('31-12-1999', 'DD-MM-YYYY'))
      PCTFREE 0 PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
      TABLESPACE sf2,
PARTITION month3
      VALUES LESS THAN (TO_DATE('31-12-2000', 'DD-MM-YYYY'))
      PCTFREE 0 PCTUSED 99
      STORAGE (INITIAL 64k NEXT 16k PCTINCREASE 0)
      TABLESPACE sf3)  AS
SELECT s.time_id, s.cust_id, SUM(s.amount_sold) AS sum_dollar_sales, 
  SUM(s.quantity_sold) AS sum_unit_sales
FROM sales s GROUP BY s.time_id, s.cust_id;

CREATE MATERIALIZED VIEW part_sales_tab_mv
ON PREBUILT TABLE
ENABLE QUERY REWRITE AS 
SELECT s.time_id,  s.cust_id, SUM(s.amount_sold) AS sum_dollar_sales, 
     SUM(s.quantity_sold) AS sum_unit_sales 
FROM sales s GROUP BY s.time_id, s.cust_id;

In this example, the table part_sales_tab has been partitioned over three months and then the materialized view was registered to use the prebuilt table. This materialized view is eligible for query rewrite because the ENABLE QUERY REWRITE clause has been included.

Rolling Materialized Views

When a data warehouse or data mart contains a time dimension, it is often desirable to archive the oldest information and then reuse the storage for new information. This is called the rolling window scenario. If the fact tables or materialized views include a time dimension and are horizontally partitioned by the time attribute, then management of rolling materialized views can be reduced to a few fast partition maintenance operations provided the unit of data that is rolled out equals, or is at least aligned with, the range partitions.

If you plan to have rolling materialized views in your data warehouse, you should determine how frequently you plan to perform partition maintenance operations, and you should plan to partition fact tables and materialized views to reduce the amount of system administration overhead required when old data is aged out. An additional consideration is that you might want to use data compression on your infrequently updated partitions.

You are not restricted to using range partitions. For example, a composite partition using both a time value and a key value could result in a good partition solution for your data.

See Chapter 15, " Maintaining the Data Warehouse" for further details regarding CONSIDER FRESH and for details regarding compression.

OLAP Cubes

While data warehouse environments typically view data in the form of a star schema, OLAP environments view data in the form of a hierarchical cube. A hierarchical cube includes the data aggregated along the rollup hierarchy of each of its dimensions and these aggregations are combined across dimensions. It includes the typical set of aggregations needed for business intelligence queries.

• Time, which contains (all times), year, quarter, and month.

• Product, which contains (all products), division, brand, and item.

This means there are 16 aggregate groups in the hierarchical cube. This is because the four levels of time are multiplied by four levels of product to produce the cube. Table 9-1 shows the four levels of each dimension.

Note that as you increase the number of dimensions and levels, the number of groups to calculate increases dramatically. This example involves 16 groups, but if you were to add just two more dimensions with the same number of levels, you would have 4 x 4 x 4 x 4 = 256 different groups. Also, consider that a similar increase in groups occurs if you have multiple hierarchies in your dimensions. For example, the time dimension might have an additional hierarchy of fiscal month rolling up to fiscal quarter and then fiscal year. Handling the explosion of groups has historically been the major challenge in data storage for OLAP systems.

Typical OLAP queries slice and dice different parts of the cube comparing aggregations from one level to aggregation from another level. For instance, a query might find sales of the grocery division for the month of January, 2002 and compare them with total sales of the grocery division for all of 2001.

Partitioning Materialized Views for OLAP

Materialized views with multiple aggregate groups will give their best performance for refresh and query rewrite when partitioned appropriately.

PCT refresh in a rolling window scenario requires partitioning at the top level on some level from the time dimension. And, partition pruning for queries rewritten against this materialized view requires partitioning on GROUPING_ID column. Hence, the most effective partitioning scheme for these materialized views is to use composite partitioning (range-list on (time, GROUPING_ID) columns). By partitioning the materialized views this way, you enable:

• PCT refresh, thereby improving refresh performance.

• Partition pruning: only relevant aggregate groups will be accessed, thereby greatly reducing the query processing cost.

If you do not want to use PCT refresh, you can just partition by list on GROUPING_ID column.

Compressing Materialized Views for OLAP

You should consider data compression when using highly redundant data, such as tables with many foreign keys. In particular, materialized views created with the ROLLUP clause are likely candidates. See Oracle Database SQL Reference for data compression syntax and restrictions and "Storage And Table Compression" for details regarding compression.

Materialized Views with Set Operators

Oracle provides support for materialized views whose defining query involves set operators. Materialized views with set operators can now be created enabled for query rewrite. You can refresh the materialized view using either ON COMMIT or ON DEMAND refresh.

Fast refresh is supported if the defining query has the UNION ALL operator at the top level and each query block in the UNION ALL, meets the requirements of a materialized view with aggregates or materialized view with joins only. Further, the materialized view must include a constant column (known as a UNION ALL marker) that has a distinct value in each query block, which, in the following example, is columns 1 marker and 2 marker.

See "Restrictions on Fast Refresh on Materialized Views with UNION ALL" for detailed restrictions on fast refresh for materialized views with UNION ALL.

CREATE MATERIALIZED VIEW LOG ON sales WITH ROWID;
CREATE MATERIALIZED VIEW LOG ON customers WITH ROWID;

CREATE MATERIALIZED VIEW unionall_sales_cust_joins_mv
REFRESH FAST ON COMMIT 
ENABLE QUERY REWRITE AS
(SELECT c.rowid crid, s.rowid srid, c.cust_id, s.amount_sold, 1 marker
FROM sales s, customers c
WHERE s.cust_id = c.cust_id AND c.cust_last_name = 'Smith')
UNION ALL
(SELECT c.rowid crid, s.rowid srid, c.cust_id, s.amount_sold, 2 marker
FROM sales s, customers c
WHERE s.cust_id = c.cust_id AND c.cust_last_name = 'Brown');

CREATE MATERIALIZED VIEW LOG ON sales WITH ROWID, SEQUENCE
(amount_sold, time_id)
INCLUDING NEW VALUES;

CREATE MATERIALIZED VIEW LOG ON times WITH ROWID, SEQUENCE 
  (time_id, fiscal_year, fiscal_quarter_number, day_number_in_week) 
INCLUDING NEW VALUES;

CREATE MATERIALIZED VIEW unionall_sales_mix_mv
REFRESH FAST ON DEMAND AS
(SELECT 'Year' umarker, NULL, NULL, t.fiscal_year, 
        SUM(s.amount_sold) amt, COUNT(s.amount_sold), COUNT(*)
 FROM sales s, times t
 WHERE s.time_id = t.time_id 
 GROUP BY t.fiscal_year)
UNION ALL
(SELECT 'Quarter' umarker, NULL, NULL, t.fiscal_quarter_number, 
        SUM(s.amount_sold) amt, COUNT(s.amount_sold), COUNT(*)
FROM sales s, times t 
WHERE s.time_id = t.time_id and t.fiscal_year = 2001
GROUP BY t.fiscal_quarter_number)
UNION ALL
(SELECT 'Daily' umarker, s.rowid rid, t.rowid rid2, t.day_number_in_week,
        s.amount_sold amt, 1, 1
FROM sales s, times t
WHERE s.time_id = t.time_id 
AND t.time_id between '01-Jan-01' AND '01-Dec-31');

Querying Materialized Views with Virtual Private Database

For all security concerns, a materialized view serves as a view that happens to be materialized when you are directly querying the materialized view. When creating a view or materialized view, the owner needs to have the necessary permissions to access the underlying base relations of the view or materialized view that they are creating. With these permissions, the owner can publish a view or materialized view that other users can access, assuming they have been granted access to the view or materialized view.

Using materialized views with Virtual Private Database (VPD) is similar. When you create a materialized view, there must not be any VPD policies in effect against the base relations of the materialized view for the owner of the materialized view. However, the owner of the materialized view may establish a VPD policy on the new materialized view. Users who access the materialized view are subject to the VPD policy on the materialized view. However, they are not additionally subject to the VPD policies of the underlying base relations of the materialized view, since security processing of the underlying base relations is performed against the owner of the materialized view.