Paging-Query based

[Atul Mishra]

All,

I have decided to implement query based paging for my huge report which can return large resultsets and now I need help with implementation.

In Oracle I learned about ROWNUM keyword. And I read articles about using ORDER BY DESC..

have anyone implemented this in Oracle ? If so, can you help ?

Thanks,

[Stan James]

I've used "select ... rownum() where rownum < nnn" in Oracle for this kind of thing. It's been years ... that's probably not exactly the right syntax. Today I had to learn "FETCH FIRST nnn ROWS ONLY" for DB2. ORDER BY might make the database engine fetch all rows and sort them before giving you the count you want so you may not save much.

Here are some notes from prior conversations about paging ...

I. Repeat the query for each new request.
   Algorithm
      Client requests page 1
      Execute the query
      Return rows for page 1
      Client requests page 2
      Execute the query
      Skip rows for page 1
      Return rows for page 2
   Variation
      Most RDBMS vendors support returning a specified range of row numbers.
      Efficiency varies. May not help much when combined with "order by"
   Pro/Con
      Pro: Stateless. Nothing is cached between requests.
      Pro: Simple. Every request is the same.
      Pro: Up to date with any inserts or deletes in the database.
      Con: Database and time intensive. Might be repeating large,
           expensive queries - if users really do ever page fwd.
      Con: Not guaranteed consistent. If other processes are inserting and
           deleting rows, paging fwd or bkwd might skip rows. Fwd then Bkwd
           might not give the same rows.
   Notes   We have used this with mainframe requests where even the nastiest
           queries were fast and the server had no options for storing state.
           We have used row number schemes for client requests, and the highest
           key on a page for page forward only action.
II. Hold the query results on the server
   Algorithm
      Client requests page 1
      Execute query
      Cache complete result set or just PKeys
      Return rows for page 1
      Client requests page 2
      Get rows from cache or get PKeys from cache, rows from database
      Return rows for page 2
   Pros/Cons
      Pro: Does NOT repeat the query
      Pro: Self consistent - fwd then bkwd will give identical results
      Pro: Can share cached results if two users have identical requests
      Con: Big cache in memory or someplace close to it
      Con: Complexity for cache, time-out
III. Hold a scrolling cursor on the server
   Algorithm
      Client requests page 1
      Execute query with a cursor
      Fetch & return rows for page 1
      Client requests page 2
      Fetch & return rows for page 2
   Pro/Con
      Pro: Does not repeat the query
      Pro: Very small state - just a connection & cursor
      Pro: Self consistent
      Con: More open connections. Might hold open one db connection
           per user
      Con: What does this do to the db? Doesn't it cache rows not yet
           fetched?
      Con: Complexity for time-out
IV. Hold results in a temporary table
   Algorithm
      Client requests page 1
      Execute the big query
      For each row of results
         insert the row into a temp table, keyed on session id
      Maybe do this in a stored proc?
      Page through the temp table using I, II or III.
   Pro/Con
      Pro: The big query is only done once
      Con: Initial hit time - adds a lot of inserts
      Con: Need to clean up that temp table some time
   Variation
      Select directly into a temp table if your db allows it