Query pagination on DB2 - an alternative way
--------------------------------------------

                 Key: OPENJPA-759
                 URL: https://issues.apache.org/jira/browse/OPENJPA-759
             Project: OpenJPA
          Issue Type: Improvement
          Components: sql
    Affects Versions: 2.0.0
         Environment: DB2 UDB
            Reporter: Milosz Tylenda
            Priority: Minor
         Attachments: OPENJPA-759.patch

The attached patch provides an alternative way of doing paging in DB2.
I made it hoping that it would improve the performance of paging queries
but it didn't, at least not when the database and OpenJPA are co-located.
There is probably no point in committing this patch unless someone
proves it gives any advantages.
I am however presenting it here, maybe some users will find it useful
(it would be nice if someone could check timings with remote database).

Attached are:
- a patch which modifies DB2Dictionary and also adds a few tests to 
TestQueryPagination.
- a standalone modified DB2Dictionary class.
- a timer program which was supposed to prove the superior performance but 
failed :)

Some characteristics:
- The idea is to limit the result set returned by SQL query instead of skipping 
rows
  when traversing the result set in OpenJPA. A similar approach exists
  in OracleDictionary.
- If setMaxResulsts and setFirstResult were called on Query, the SQL query
  [QUERY] is modified as follows:

SELECT * FROM (
  SELECT rr.*, ROW_NUMBER() OVER(ORDER BY ORDER OF rr) AS rn FROM (
    [QUERY]
    FETCH FIRST [m] ROWS ONLY
  ) AS rr
) AS r WHERE rn > [n] ORDER BY rn

- The modified SQL query adds one column to the end of column list in the
  result set. Luckily, I couldn't find any side effects of doing this.
- If only setMaxResults was called on Query, only FETCH FIRST [m] ROWS ONLY
  is appended to SQL query - this is how it works currently.
- The new way of paging will be used only if the database is a UDB 8.1 or later
  because of ORDER OF construct and FETCH FIRST [m] ROWS ONLY in a subselect.
  Maybe some other DB2 flavours could also handle it but I have no access.
- User can fall back to the old behaviour by setting supportsSelectStartIndex
  Dictionary property to false.




-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to