Ole Solberg wrote:
I rebuilt from the same source, but now with 'debug=false' and ran derbyall on the same host: predicatePushdown did NOT fail. So this seems to be the same instability as in e.g. the Wisconsin test, and triggered by larger/slower(?) code causing an unexpected query plan to be chosen.
I seem to recall than in a discussion on optimizer timeout it was said that the optimizer will terminate if it has run for longer than the so-far lowest estimate for running a query. I guess that could result in a less optimal plan on a slow computer. Maybe this is what is occurring here?
The above strategy makes sense for one-time execution of queries. In the case where one will execute the same prepared query very many times, it may be desirably to spend more time optimizing a query than the time it takes to execute the query once.
-- Øystein Grøvlen
