JULIA_EXCLUSIVE and affinity of sub-processes

[tkf]

If I start Julia with JULIA_EXCLUSIVE=1 julia1.7 -t4 I get

julia> results = Vector{String}(undef, Threads.nthreads())
       Threads.@threads for i in 1:Threads.nthreads()
           results[i] = read(`sh -c 'taskset -p $$'`, String)
       end
       results
4-element Vector{String}:
 "pid 27240's current affinity mask: 1\n"
 "pid 27237's current affinity mask: 2\n"
 "pid 27239's current affinity mask: 4\n"
 "pid 27238's current affinity mask: 8\n"

This seems to be non-ideal. Programmers ideally don't need to care which tasks run on which thread. So, it's not possible to control the CPU affinity of sub-processes in idiomatic Julia code. If two processes that can run in parallel is happened to be spawned from the same worker thread, they cannot fully utilize CPU resources.

How about saving the affinity at an early phase of the startup (so that configuration of taskset etc. propagates) and then set it in uv_process_options_t when calling uv_spawn?