I'm stress testing 4.1.25 of BDB using the Perl API. I'm coming up with some interesting results that are rather confusing for me. My testing is rather simple, I create a Btree DB with no duplication and no enviornment, then write in 32 Million unique records. After completion of this, I shutdown the application doing the test and run db_verify on the database. My expectation was that it would fail. It did in fact fail with:

db_verify: Page 709826: invalid next_pgno 709844
db_verify: Page 709827: invalid next_pgno 709843

However, rerunning db_verify again produced a different set of invald next_pgno's and then another run of db_verify came back passing with no errors.

So my question is, is this strangeness caused by the corruption of the db file or is this an issue with the verfication tool?