SecondaryMultiKeyCreator bug still there?

Question

Hi, I am using the java api for the Berkeley DB, and I ran into trouble using a BTREE to secondary index a RECNO primary database. After searching the forum, it seems like this was fixed here:
http://forums.oracle.com/forums/thread.jspa;jsessionid=8d92079330d60b9511031eb54ca799d8b3cc1d2ddfd8.e38QbNuQbx8Kbi0LbhaMaxqTahj0?messageID=2536190�
However, has this change not propagated into the trunk?
Essentially when I use this multi-key generator, it works fine if i comment out the second add statement, but with two add statements it totally fails to work - reading the secondary key shows a 2 byte secondary key. If you call add 3 times, it will output a 3 byte secondary key.
     private class WordGenerator implements SecondaryMultiKeyCreator {
          public void createSecondaryKeys(...) {
                    Entry e = (Entry) entrybinder.entryToObject(data);
                    DatabaseEntry seckey1 = new DatabaseEntry(), seckey2 = new DatabaseEntry();
                    stringbinder.objectToEntry(e.title.substring(0,2), seckey1);
                    results.add(seckey1);
                    stringbinder.objectToEntry(e.title.substring(2,4), seckey2);
                    results.add(seckey2);
Edit: I am now going to try using a Btree for the primary database, and use a sequence. Hope it works.
Edit2: I have confirmed that the same code works now that I use a sequence on a BTree primary database

Answer 1

Hi,
Probably you already read "[Selecting an access method|http://www.oracle.com/technology/documentation/berkeley-db/db/ref/am_conf/select.html]". There's the place you should look when choosing an access method, not considering how the access methods work with sequences, cursors, etc. So it depends on what kind of data you have and how you'll access it. For example if your data items can be fixed-length and you want to achieve write concurrency, you'll want to consider using the Queue access method, since it can assign IDs faster than a Btree/Recno data structure, and offers higher concurrency for that reason. What I suggest you is to run a performance test, by reproducing your application's behavior under what you expect to be the production use.
Be aware you shouldn't configure the database to support duplicate data items (DB_DUP) when using sequences.
Related doc:
Introduction to sequences: http://www.oracle.com/technology/documentation/berkeley-db/db/ref/sequence/intro.html
Thanks,
Bogdan