Optimize collecting dynamic memory usage stats of indices
- At each modification of item though index, measure and cache it's dynamic size, so that it's not necessary to do a walk over the structures at each benchmark interval
- Removed call to
dump
frommeasure
function. This was causing an awful slowdown of the replay, because at every benchmark intervalreplay_benchmark.json
would be written with all the measurements, just to be overwritten at next interval