Slurm: SharedArrays fails on ClusterManagers.jl

[juandarias]

The documentation example for SharedArrays in parallel computing fails when using a SlurmManager:
using ClusterManagers, Distributed
addprocs(SlurmManager(8), topology=:all_to_all)
using SharedArrays

a = SharedArray{Float64}(100)
@distributed for i = 1:100
a[i] = i
end

Returns an empty array. The following code will work:

using ClusterManagers, Distributed
addprocs(8, topology=:all_to_all)
using SharedArrays

a = SharedArray{Float64}(100)
@distributed for i = 1:100
a[i] = i
end

Packages:
Julia 1.1.0
ClusterManagers 0.3.2

[juandarias]

Any idea?

[vchuravy]

I am not sure what should happen here SharedArray only works when all processes are on the same machine. Something that is not guaranteed with SLURM.

[juandarias]

Shall I raise this as an issue or feature request for SharedArray?

[vchuravy]

As I said it is not something that SharedArray can really support, but maybe it could throw an error.