Comment 5 for bug 622668

Ignoring lag, or not using DBLoopTuner at all, will not help. The overhead per chunk is higher than the target time, so the chunk size will decrease to the minimum chunk size. This is currently set to 1.

A quick fix would seem to be to increase the minimum chunk size. However, we are trying to limit our transactions to <2s to avoid contention issues.

We need to fix the per-chunk overhead.

Maybe calculating everything upfront into a temporary table could help? I'd like to avoid temporary tables, but maybe we really need them.

Maybe it is just an slow query to fix.

Maybe we need a better index.