XMLQuery starts using XMLSEQUENCEFROMXMLTYPE when adding filter to query

Question

Hi,
I'm creating a single XML view on 3 relational tables. These table represent:
1. a dataset description (table: refs)
2. a keyset (table: keyset, a table with between refs )
3. a data set (table: data, consisting of reference to key, value, and reference to a dataset)
Per dataset I can have multiple keys, per key I have multiple values
(1:N for dataset:keys, 1:N keys:values)
The definition is given below:
DROP TABLE data;
DROP TABLE refs;
DROP TABLE keyset;
CREATE TABLE data (ref int, key int, value float);
CREATE TABLE refs (ref int);
CREATE TABLE keyset (key int, ref int);
CREATE INDEX data_krv ON data (key,ref,value);
CREATE INDEX keyset_kr ON keyset (key,ref);
INSERT INTO refs VALUES (1);
INSERT INTO refs VALUES (2);
INSERT INTO data VALUES (1,1,1.5);
INSERT INTO data VALUES (1,1,2.5);
INSERT INTO data VALUES (1,2,3.5);
INSERT INTO data VALUES (1,2,4.5);
INSERT INTO data VALUES (2,1,5.5);
INSERT INTO data VALUES (2,1,6.5);
INSERT INTO data VALUES (2,2,7.5);
INSERT INTO data VALUES (2,2,8.5);
INSERT INTO keyset SELECT DISTINCT key, ref FROM data;
CREATE OR REPLACE VIEW drk_xml_view OF XMLType
with OBJECT ID
  extract(object_value,'/ref').getnumberval()
AS
SELECT xmlElement(
     "ref",
     xmlElement("ref_id", refs.ref),
     (SELECT xmlAgg(
                xmlElement("key",
                  xmlElement("key_id",keyset.key),
                  xmlElement("values",
                      (SELECT xmlAgg(xmlElement("value",data.value))
                       FROM data
                       WHERE data.key=keyset.key AND data.ref=keyset.ref)
      FROM keyset WHERE refs.ref = keyset.ref
) FROM refs
/When I do a query like:
SELECT xmlQuery('for $i in /ref return max($i/key/values/value)' passing object_value returning content) from drk_xml_view;the explain plan looks as expected:
| Id  | Operation              | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT       |           |     2 |    26 |     3   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE        |           |     1 |    65 |            |          |
|   2 |   NESTED LOOPS         |           |     1 |    65 |     6   (0)| 00:00:01 |
|*  3 |    INDEX FAST FULL SCAN| KEYSET_KR |     1 |    26 |     3   (0)| 00:00:01 |
|*  4 |    INDEX RANGE SCAN    | DATA_KRV  |     1 |    39 |     3   (0)| 00:00:01 |
|   5 |  SORT AGGREGATE        |           |     1 |       |            |          |
|   6 |   FAST DUAL            |           |     1 |       |     2   (0)| 00:00:01 |
|   7 |  TABLE ACCESS FULL     | REFS      |     2 |    26 |     3   (0)| 00:00:01 |
Predicate Information (identified by operation id):
   3 - filter("KEYSET"."REF"=:B1)
   4 - access("DATA"."KEY"="KEYSET"."KEY" AND "DATA"."REF"=:B1)
       filter("DATA"."REF"="KEYSET"."REF")
Note
   - dynamic sampling used for this statementThis is very nicely optimized.
But now I do this one:
SELECT xmlQuery('for $i in /ref[./ref_id<2] return max($i/key/values/value)' passing object_value returning content) from drk_xml_view;(where I only added a constraint on the /ref/ref_id) the following occurs:
| Id  | Operation                          | Name                   | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT                   |                        |     2 |    26 |     3   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE                    |                        |     1 |     2 |            |          |
|   2 |   COLLECTION ITERATOR PICKLER FETCH| XMLSEQUENCEFROMXMLTYPE |       |       |            |          |
|   3 |  SORT AGGREGATE                    |                        |     1 |       |            |          |
|*  4 |   FILTER                           |                        |       |       |            |          |
|   5 |    FAST DUAL                       |                        |     1 |       |     2   (0)| 00:00:01 |
|   6 |  TABLE ACCESS FULL                 | REFS                   |     2 |    26 |     3   (0)| 00:00:01 |
Predicate Information (identified by operation id):
   4 - filter(:B1<2)
Note
   - dynamic sampling used for this statementSo, it doesn't use any index any more and creates a XMLSequenceFromXMLType instead. This one is then parsed which results in really bad performance.
I've tried to add extra indexes to the tables, but that didn't result in an optimilization. From a SQL point of view this seems quite easy, but using the view it gives me problems.
Is there someone who can explain me what is happening here? Or give me some pointers?
Please let me know if you need to know more, or if something is unclear.
Best regards

Answer 1

SQL> SELECT xmlQuery('for $i in /ref where $i/ref_id < 2 return max($i/key/values/value) ' passing object_value returning content)
  2    from drk_xml_view
  3  /
XMLQUERY('FOR$IIN/REFWHERE$I/REF_ID<2RETURNMAX($I/KEY/VALUES/VALUE)'PASSINGOBJECT_VALUERETURNINGCONTENT)
4.5
Execution Plan
Plan hash value: 2754328746
| Id  | Operation           | Name   | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT    |        |     2 |    26 |     2   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE     |        |     1 |    65 |            |          |
|*  2 |   HASH JOIN         |        |     1 |    65 |     5  (20)| 00:00:01 |
|*  3 |    TABLE ACCESS FULL| KEYSET |     1 |    26 |     2   (0)| 00:00:01 |
|*  4 |    TABLE ACCESS FULL| DATA   |     1 |    39 |     2   (0)| 00:00:01 |
|   5 |  SORT AGGREGATE     |        |     1 |       |            |          |
|*  6 |   FILTER            |        |       |       |            |          |
|   7 |    FAST DUAL        |        |     1 |       |     2   (0)| 00:00:01 |
|   8 |  TABLE ACCESS FULL  | REFS   |     2 |    26 |     2   (0)| 00:00:01 |
Predicate Information (identified by operation id):
   2 - access("DATA"."KEY"="KEYSET"."KEY" AND
              "DATA"."REF"="KEYSET"."REF")
   3 - filter("KEYSET"."REF"=:B1)
   4 - filter("DATA"."REF"=:B1)
   6 - filter(:B1<2)
Note
   - dynamic sampling used for this statement
SQL>