> * I did not see the non-deterministic behavior that Bryan did > with 10.2 trunk--instead, I consistently see 7+ minutes. However, > I only ran the query 3 times, which is probably not enough to have > seen what Bryan saw. Also note that I only ran the DDL one time > and then I re-ran the query the second two times. So given Mike's > previous comment, this may be the reason I didn't see the variance that Bryan did.
Thanks for all the help and good suggestions so far -- they have given me a lot of ideas to pursue. I think that Mike's theory about the source of the non-determinism being due to whether I recreate the tables each time, or re-use the existing tables, is a good one, and I'll see if I can confirm it with more tests of my own. Meanwhile, I'll also investigate the other pointers you passed along. Regarding the comment that the DERBY-2130 test case is a hairy one, and it would be easier to start with a simpler one, I agree. On the other hand, the appeal of the DERBY-2130 test case is that: a) it is complicated enough to vividly show that there is a problem b) it is still pruned down enough so that it runs; some of my other test cases ran for hours and "never" terminated :) Again, thanks for all the help and ideas; they are much appreciated. bryan
