Christophe Courtois

I can reproduce at will: without parallelization: https://explain.dalibo.com/plan/4J with: https://explain.dalibo.com/plan/5gW 178 Mbytes read in both cases (from OS cache), but its 500 ms or 5 s. I suspect that the...

According to @ioguix, adding the workers IO timings may be normal. If this is the case, I think that pev2 should add an hint, perhaps the average per worker (beware...

just in case, 2 examples from our workshop of the same query with and without async: https://explain.dalibo.com/plan/QEC https://explain.dalibo.com/plan/j6V (merci @fljdin )

I'd love too. I suppose that Pierre would love to have a few examples to test some algorithms and how we would like the differences to show up. BTW :...

Ideas of I/O intensive operations, as just discussed: * temp files due to sorting ( https://explain.dalibo.com/plan/pES ) or hash joins ( https://explain.dalibo.com/plan/f0S ) * temporary tables ( "local" buffers, see...

ping @gleu

Must be an option The raw code must always be available

Different warnings could appear if Generation or Inlining or Optimization are a big part of the total query time. These are different `jit_*_above_cost` parameters.

Note that the JIT was the real problem ; I'm not sure that you can detect the problem without `VERBOSE`

+1