Administered by:
@esmi am still quite concerned that a good amount of memory isn't being returned back to the system on the esmbot image natives
@esmhere's a graph of memory usage during a stress test
it mostly sticks at around 100mb, with occasional spikes from very heavy workloads
the RSS at the end was 3GB
@esm@wetdry.worldjemalloc, mimalloc, etc. mitigate this very slightly, but do not resolve the issue
@esmgave it a try on freebsd because i was curious, it seems to be a similar situation there as well
@esminterestingly when i tested on mac, the activity monitor reported memory usage typically being below 100mb - but other tools on the same system reported the total usage being much higher
@esmi'll probably try and convert the stress test script into an actual c++ program, just to take some possible factors out of the equation
@esmi have made a c++ version of the script
memory handling appears to be the same at first glance
@esm would this be fragmentation? seems like it but this one isn't experienced so might be wrong..
if it is then there are a few malloc impls that automatically defrag memory, however this one can't find the one it remembers reading about specifically 