Distributed arrays in for loop

General Usage
1

Hi, I’m trying to get distributed arrays to work correctly for my distributed for loop. I’m getting this weird outcome where not all workers do their task always, even though I know they’ve been assigned correctly (using procs). Not sure what’s going on. Here a MWE:

using Distributed

addprocs(4)
@everywhere begin
    using DistributedArrays
    result_D = dzeros(Float64,(2,4), workers()[1:4],[1,4])
    initq = [i*ones(2,1) for i in 1:4]  
    @sync @distributed for ele in initq 
        res_local = localpart(result_D)
        res_local .= ele
    end
end

result = convert(Array,result_D)

When I run this, the results are inconsistent: sometimes the whole array stays empty, sometimes one worker does a job, sometimes a few. What could be the issue?

2

I think you shouldn’t define the distributed array and run the @distributed loop inside the @everywhere block. This might cause the issues you see, since somehow every worker ends up allocating their own array and I am not sure which worker will then write where.

3

Thank you @abraemer ! This indeed did the trick. Here the working version:

using Distributed

addprocs(4)
@everywhere begin

    using Distributed, DistributedArrays
    initq = [i*ones(2,1) for i in 1:4] 
end
result_D = dzeros(Float64,(2,8), workers()[1:4],[1,4])
    
@sync @distributed for ele in initq 
    res_local = localpart(result_D)
    res_local .= ele
end
result = convert(Array,result_D)