Cannot use "threads" parameter without crashing

[burchds]

I have been using JuMP and Cbc for a couple of years now. It's fantastic. However, I recently upgraded to the latest version of Julia and latest version of JuMP and Cbc. After upgrading my models would not run. In fact, Julia would just crash every time.

I am using the following packages on Julia 1.7.1 (current stable release). {edit: Hardware is AMD 3900x CPU, mentioning in case it's a problem that isn't seen on Intel processors.}
[9961bab8] Cbc v0.9.1
[60bf3e95] GLPK v0.15.2
[4076af6c] JuMP v0.22.2

I've spent some time troubleshooting the problem. It looks like I cannot use the "threads" parameter. If I comment out that line, then the code runs.

Here is a working example of what I am testing to isolate the problem. As you can see, I have commented out the line that attempts to set the "threads" parameter.

This code does work:

using JuMP
using Cbc

function main()
		n = 15
		m = 5
		vecCost = [1115, 1042, 1052, 1059, 1071, 1075, 1089, 1034, 1021, 1027, 1009, 996, 1024, 1011, 907]
        storesMatrix = Bool[1 1 0 0 0;
		                    1 0 1 0 0;
		                    1 0 0 1 0;
		                    1 0 0 0 1;
		                    1 0 0 0 0;
		                    0 1 1 0 0;
		                    0 1 0 1 0;
		                    0 1 0 0 1;
		                    0 1 0 0 0;
		                    0 0 1 1 0;
		                    0 0 1 0 1;
		                    0 0 1 0 0;
		                    0 0 0 1 1;
		                    0 0 0 1 0;
		                    0 0 0 0 1]
		model = Model(Cbc.Optimizer)
		set_optimizer_attribute(model, "seconds", 60)
		set_optimizer_attribute(model, "loglevel", 0)
		set_optimizer_attribute(model, "ratioGap", 0.0001)
		
		#set_optimizer_attribute(model, "threads", 4)
        
		@variable(model, x[1:n], Bin)
        @objective(model, Min, sum(vecCost[i] * x[i] for i in 1:n))
        @constraint(model, column_con1[j in 1:m], sum(storesMatrix[i, j] * x[i] for i in 1:n) == 1)
        JuMP.optimize!(model)
		for i in 1:n
            println(JuMP.value(x[i]))
        end
end

main()

This code does not work:

using JuMP
using Cbc

function main()
		n = 15
		m = 5
		vecCost = [1115, 1042, 1052, 1059, 1071, 1075, 1089, 1034, 1021, 1027, 1009, 996, 1024, 1011, 907]
        storesMatrix = Bool[1 1 0 0 0;
		                    1 0 1 0 0;
		                    1 0 0 1 0;
		                    1 0 0 0 1;
		                    1 0 0 0 0;
		                    0 1 1 0 0;
		                    0 1 0 1 0;
		                    0 1 0 0 1;
		                    0 1 0 0 0;
		                    0 0 1 1 0;
		                    0 0 1 0 1;
		                    0 0 1 0 0;
		                    0 0 0 1 1;
		                    0 0 0 1 0;
		                    0 0 0 0 1]
		model = Model(Cbc.Optimizer)
		set_optimizer_attribute(model, "seconds", 60)
		set_optimizer_attribute(model, "loglevel", 0)
		set_optimizer_attribute(model, "ratioGap", 0.0001)
		
		set_optimizer_attribute(model, "threads", 4)
        
		@variable(model, x[1:n], Bin)
        @objective(model, Min, sum(vecCost[i] * x[i] for i in 1:n))
        @constraint(model, column_con1[j in 1:m], sum(storesMatrix[i, j] * x[i] for i in 1:n) == 1)
        JuMP.optimize!(model)
		for i in 1:n
            println(JuMP.value(x[i]))
        end
end

main()

I have also tried setting it to 1 thread. It just doesn't seem to like the parameter being used at all.

[odow]

I can't reproduce this:

julia> using JuMP

julia> using Cbc

julia> function main()
                       n = 15
                       m = 5
                       vecCost = [1115, 1042, 1052, 1059, 1071, 1075, 1089, 1034, 1021, 1027, 1009, 996, 1024, 1011, 907]
               storesMatrix = Bool[1 1 0 0 0;
                                           1 0 1 0 0;
                                           1 0 0 1 0;
                                           1 0 0 0 1;
                                           1 0 0 0 0;
                                           0 1 1 0 0;
                                           0 1 0 1 0;
                                           0 1 0 0 1;
                                           0 1 0 0 0;
                                           0 0 1 1 0;
                                           0 0 1 0 1;
                                           0 0 1 0 0;
                                           0 0 0 1 1;
                                           0 0 0 1 0;
                                           0 0 0 0 1]
                       model = Model(Cbc.Optimizer)
                       set_optimizer_attribute(model, "seconds", 60)
                       set_optimizer_attribute(model, "loglevel", 0)
                       set_optimizer_attribute(model, "ratioGap", 0.0001)
                       
                       set_optimizer_attribute(model, "threads", 4)
               
                       @variable(model, x[1:n], Bin)
               @objective(model, Min, sum(vecCost[i] * x[i] for i in 1:n))
               @constraint(model, column_con1[j in 1:m], sum(storesMatrix[i, j] * x[i] for i in 1:n) == 1)
               JuMP.optimize!(model)
                       for i in 1:n
                   println(JuMP.value(x[i]))
               end
       end
main (generic function with 1 method)

julia> main()
0.0
0.0
1.0
0.0
0.0
1.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1.0

julia> 

(cbc) pkg> st
      Status `/private/tmp/cbc/Project.toml`
  [9961bab8] Cbc v0.9.1
  [4076af6c] JuMP v0.22.3

How are you starting Julia? What is versioninfo()? Are you running any things in parallel?

[GregPlowman]

I can reproduce on Windows.

Seems it might be Windows specific, see https://discourse.julialang.org/t/cbc-solver-crashes-julia-with-multithreading/73859

[odow]

I don't have a Windows machine to debug. At this point I would encourage you instead to use HiGHS.jl. It's actively developed, and has similar or better performance to Cbc.

[GregPlowman]

Sure. Thanks! I’m using HiGHS now.
(I was really just providing another data point for others)

[odow]

I've confirmed that this is Windows-specific: #190. But this is unlikely to be resolved. If you are on Windows, don't use the threads parameter.

[mlubin]

Let's print a warning if we see windows users using threads?

[odow]

Done here: #192

[odow]

Closing because this is unlikely to be fixed and Cbc.jl disables threading on Windows.