Outrageously slow on Mac

[schneems]

We accidentally found a reproduction where derailed benchmarks is REALLY slow on a mac and 15x faster on a docker container on the SAME machine. Here's the repro: zombocom/derailed_benchmarks#136 (comment)

It looks like the vast majority of time is spent in get_process_mem

$ sudo --preserve-env rbspy record -- bundle exec derailed bundle:mem
$ rbspy report -f summary -i /Users/rschneeman/.cache/rbspy/records/rbspy-2019-07-04-1zuBautmfg.raw.gz --output summary.txt && cat summary.txt
% self  % total  name
 82.87    83.82  ps_memory - /Users/rschneeman/.gem/ruby/2.5.1/gems/get_process_mem-0.2.3/lib/get_process_mem.rb
 10.19    96.50  block (2 levels) in <top (required)> - /Users/rschneeman/.gem/ruby/2.5.1/gems/derailed_benchmarks-1.3.5/lib/derailed_benchmarks/core_ext/kernel_require.rb

My theory is that the speed difference is due to shelling out to ps on a mac versus with linux we use the /proc file system.

If I replace the ps call with another system call such as echo 1000 then it still takes a long time (0m12.464s) however it is faster than the ps call. Removing the shell out entirely takes it down to 0m1.473s which is still slower than docker, but at least it's fast-ish. It looks like the majority of time comes from the shell out, then after than the ps command actually takes a long time to run. Finally there are some other mac/linux differences here, but this tells me we should focus on trying to remove the shell out command to ps somehow on get_process_mem

Maybe there is a faster way to get this data from mac.

[schneems]

Closed via #32