Recently, I changed the task distribution so that jobs with more sites would produce more tasks. Whereas this seemed sensible (a rupture affecting more sites should weight more) it has an inconvenient side effect: it easy to have jobs generating too many tasks. On the other hand, with small tiles, it becomes easy to generate too few tasks per job. So we should go back to the past: the weight should depend on the number of ruptures only, not on the number of affected sites. If there are many sites the tasks will be heavier, but that's life. If one tasks is getting ruptures affecting several sites (more than typical for the other tasks) it will be slower. But at least the
rule is simple. The problem of slow tasks must be solved in some other way.
https:/ /github. com/gem/ oq-commonlib/ pull/19