Saving SharedArrays

General Usage
1

If I generate data as:

using Distributed
using SharedArrays
using JLD2
n_trials = 10^2;

results = SharedArray{Float64}(n_trials);

@sync @distributed for i in 1:n_trials
    results[i] = i;
end
jldsave("test.jld2"; results)

when I then go to load it:

using JLD2
data = jldopen("test.jld2")
results = data["results"]

I get the error:

ERROR: ConcurrencyViolationError("setfield!: atomic field cannot be written non-atomically")
Stacktrace:
  [1] macro expansion
    @ ~/.julia/packages/JLD2/AilrO/src/data/reconstructing_datatypes.jl:563 [inlined]
  [2] jlconvert(#unused#::JLD2.ReadRepresentation{Distributed.Future, JLD2.OnDiskRepresentation{(0, 8, 16, 24, 32), Tuple{Int64, Int64, Int64, ReentrantLock, Union{Nothing, Some{Any}}}, Tuple{Int64, Int64, Int64, JLD2.RelOffset, JLD2.RelOffset}}()}, f::JLD2.JLDFile{JLD2.MmapIO}, ptr::Ptr{Nothing}, header_offset::JLD2.RelOffset)
    @ JLD2 ~/.julia/packages/JLD2/AilrO/src/data/reconstructing_datatypes.jl:508
  [3] read_scalar
# and it continues for quite a bit

Is this a bug? Or is there an alternative way to save SharedArrays?

2

I assume if you jldsave("test.jld2"; results = results.s), loading it works?

3

Yes, that works.