We could regularly track malloc_info and store that in the heaptrack file to visualize it later.
This would add the following information:
- per-heap (i.e. per thread) information
- internal small/large bins
- fragmentation
- in general, useful to double-check against overhead estimation
The bad is that this is potentially costly and generates relatively large XML data streams... Maybe worth doing only every 1s or so.
See also: