Cannot get HSL to work with AmplNLWriter

I have installed HSL_jll.jl as provided by HSL and also installed HSL.jl. With this I can get Ipopt to work with ma27 when I use Ipopt directly:

using JuMP, Ipopt, HSL, HSL_jll
model = Model(Ipopt.Optimizer)
@variable(model, x[1:2]>=0) 
@objective(model, Min, sum(x.^2))
@constraint(model, sin(x[1]) <= -0.5)
set_optimizer_attribute(model, "hsllib", HSL_jll.libhsl_path)
set_optimizer_attribute(model, "linear_solver", "ma27")
optimize!(model)

However, when I try to use AmplNLWriter, it fails:

using JuMP, Ipopt_jll, HSL, HSL_jll, AmplNLWriter
model = Model(() -> AmplNLWriter.Optimizer(Ipopt_jll.amplexe))
@variable(model, x[1:2]>=0) 
@objective(model, Min, sum(x.^2))
@constraint(model, sin(x[1]) <= -0.5)
set_optimizer_attribute(model, "hsllib", HSL_jll.libhsl_path)
set_optimizer_attribute(model, "linear_solver", "ma27")
optimize!(model)
Ipopt 3.14.16: hsllib=/home/anaconda/Notebooks/Joshua/HSL_jll.jl-2023.11.7/override/lib/x86_64-linux-gnu-libgfortran5/libhsl.so
linear_solver=ma27

Exception of type: DYNAMIC_LIBRARY_FAILURE in file "Common/IpLibraryLoader.cpp" at line 86:
 Exception message: libmpifort.so.12: cannot open shared object file: No such file or directory

EXIT: Library loading failure.

I’ve tried a variety of things, but cannot get this to work. I need this for benchmarking study I am doing where I want to include AMPL-based AD.

Hmm this might be a question for @amontoison.

What version of everything?

I am able to reproduce the issue and it’s working if you remove this line:

set_optimizer_attribute(model, "hsllib", HSL_jll.libhsl_path)

Ipopt is looking for a shared library libhsl by default if you set the HSL linear solvers, so if the library is dynamically loaded (which is done by import HSL_jll), Ipopt could find it automatically.

When I wrote the documentation for HSL_jll.jl, I wasn’t sure if it would work on all platforms without it, so as a precaution, I ensured that the path is provided.

I don’t know why it’s not working if we provide the path. I suspect that Ipopt uses other environment variables to load shared libraries. We can see with this exemple that my Ubuntu can’t detect all dependencies of HSL outside Julia because they are not in LD_LIBRARY_PATH.

julia> using MPICH_jll

shell> ldd /home/alexis/Applications/libHSL/lbt_HSL_jll.jl-2023.11.7/HSL_jll.jl-2023.11.7/override/lib/x86_64-linux-gnu-libgfortran5/libhsl.so
	linux-vdso.so.1 (0x00007fff99097000)
	libblastrampoline.so.5 => not found
	libmetis.so => /lib/x86_64-linux-gnu/libmetis.so (0x00007ef56f58a000)
	libmpifort.so.12 => not found
	libgfortran.so.5 => /lib/x86_64-linux-gnu/libgfortran.so.5 (0x00007ef56e000000)
	libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007ef56f4a1000)
	libgomp.so.1 => /lib/x86_64-linux-gnu/libgomp.so.1 (0x00007ef56f457000)
	libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007ef56f437000)
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007ef56dc00000)
	/lib64/ld-linux-x86-64.so.2 (0x00007ef56f60b000)
	libquadmath.so.0 => /lib/x86_64-linux-gnu/libquadmath.so.0 (0x00007ef56e3b8000)

When I remove

set_optimizer_attribute(model, "hsllib", HSL_jll.libhsl_path)

I get:

Ipopt 3.14.16: linear_solver=ma27

Exception of type: DYNAMIC_LIBRARY_FAILURE in file "Common/IpLibraryLoader.cpp" at line 86:
 Exception message: libhsl.so: cannot open shared object file: No such file or directory

EXIT: Library loading failure.

Any thoughts on how to get this to recognize the right path?

My package environment is as follows on an Ubuntu machine:

(@v1.11) pkg> st
Status `~/.julia/environments/v1.11/Project.toml`
  [7c4d4715] AmplNLWriter v1.2.2
  [052768ef] CUDA v5.5.2
  [45b445bb] CUDSS v0.3.2
  [1037b233] ExaModels v0.8.0
  [34c5aeac] HSL v0.4.3
  [ffdf776e] InfiniteExaModels v0.1.0 `https://github.com/infiniteopt/InfiniteExaModels.jl#main`
  [20393b10] InfiniteOpt v0.5.9 `https://github.com/infiniteopt/InfiniteOpt.jl#master`
  [b6b21f68] Ipopt v1.6.7
  [4076af6c] JuMP v1.23.4
  [2621e9c9] MadNLP v0.8.4
  [d72a61cc] MadNLPGPU v0.7.3
  [309f4015] MathOptSymbolicAD v0.2.1
  [f4238b75] NLPModelsIpopt v0.10.2
  [c36e90e8] PowerModels v0.21.3
  [017b0a0e] HSL_jll v2023.11.7+0 `../../../Notebooks/Joshua/HSL_jll.jl-2023.11.7`
  [9cc047cb] Ipopt_jll v300.1400.1600+0

I have an idea on how to fix it.
The paths of all libraries visible to Julia can be recovered with:

using Libdl
Libdl.dllist()

We probably need to add an environment variable, just like we did with LBT in the past, so that ampl_exe can detect all dependencies.
We need to update AmplNLWriter.jl with @odow for this.

using Libdl

libs = Libdl.dllist()
dirs = unique(dirname(path) for path in libs)
separator = Sys.iswindows() ? ";" : ":"
library_paths = join(dirs, separator)

if Sys.iswindows()
  println("PATH=" * library_paths)
elseif Sys.isapple()
  println("DYLD_LIBRARY_PATH=" * library_paths)
else
  println("LD_LIBRARY_PATH=" * library_paths)
end

I tested by setting the environment variable before starting Julia.
It should be the same result if we create it automatically for the user in AmplNLWriter.jl before that we execute ampl_exe:

alexis@HP-Spectre:~/Bureau/git$ LD_LIBRARY_PATH=:/lib/x86_64-linux-gnu:/home/alexis/julia/julia-1.11.1/bin/../lib:/lib64:/home/alexis/julia/julia-1.11.1/bin/../lib/julia:/home/alexis/julia/julia-1.11.1/lib/julia:/home/alexis/Applications/OpenBLAS/lib:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Base64:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Markdown:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/InteractiveUtils:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Serialization:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Logging:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Test:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/SuiteSparse_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/SparseArrays:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/StyledStrings:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Unicode:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/REPL:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Printf:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Dates:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/TOML:/home/alexis/.julia/compiled/v1.11/Preferences:/home/alexis/.julia/compiled/v1.11/JLLWrappers:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/CompilerSupportLibraries_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/NetworkOptions:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/MbedTLS_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/LibSSH2_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/LibGit2_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/LibGit2:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/ArgTools:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/nghttp2_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/LibCURL_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/MozillaCACerts_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/LibCURL:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Downloads:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Tar:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/p7zip_jll:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/UUIDs:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/Pkg:/home/alexis/julia/julia-1.11.1/share/julia/compiled/v1.11/LazyArtifacts:/home/alexis/.julia/compiled/v1.11/MPIPreferences:/home/alexis/.julia/compiled/v1.11/Hwloc_jll:/home/alexis/.julia/artifacts/a610bcf72542cca9e649774fb393a74dc28fd353/lib:/home/alexis/.julia/compiled/v1.11/MPICH_jll:/home/alexis/.julia/artifacts/e85c0a68e07fee0ee7b19c2abc210b1af2f4771a/lib:/home/alexis/.julia/compiled/v1.11/METIS_jll:/home/alexis/.julia/artifacts/e5244f73466343373a87ed1345efac5e3440fa96/lib:/home/alexis/.julia/artifacts/e5244f73466343373a87ed1345efac5e3440fa96/lib/metis/metis_Int32_Real64/lib:/home/alexis/.julia/artifacts/e5244f73466343373a87ed1345efac5e3440fa96/lib/metis/metis_Int64_Real32/lib:/home/alexis/.julia/artifacts/e5244f73466343373a87ed1345efac5e3440fa96/lib/metis/metis_Int64_Real64/lib:/home/alexis/.julia/compiled/v1.11/HSL_jll:/home/alexis/Applications/libHSL/lbt_HSL_jll.jl-2023.11.7/HSL_jll.jl-2023.11.7/override/lib/x86_64-linux-gnu-libgfortran5

alexis@HP-Spectre:~/Bureau/git$ julia
               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.11.1 (2024-10-16)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> using JuMP, Ipopt_jll, HSL, HSL_jll, AmplNLWriter

julia> model = Model(() -> AmplNLWriter.Optimizer(Ipopt_jll.amplexe))
A JuMP Model
├ solver: AmplNLWriter
├ objective_sense: FEASIBILITY_SENSE
├ num_variables: 0
├ num_constraints: 0
└ Names registered in the model: none

julia> @variable(model, x[1:2]>=0)
2-element Vector{VariableRef}:
 x[1]
 x[2]

julia> @objective(model, Min, sum(x.^2))
x[1]² + x[2]²

julia> @constraint(model, sin(x[1]) <= -0.5)
sin(x[1]) - -0.5 ≤ 0

julia> set_optimizer_attribute(model, "hsllib", HSL_jll.libhsl_path)

julia> set_optimizer_attribute(model, "linear_solver", "ma27")

julia> optimize!(model)
Ipopt 3.14.16: hsllib=/home/alexis/Applications/libHSL/lbt_HSL_jll.jl-2023.11.7/HSL_jll.jl-2023.11.7/override/lib/x86_64-linux-gnu-libgfortran5/libhsl.so
linear_solver=ma27


******************************************************************************
This program contains Ipopt, a library for large-scale nonlinear optimization.
 Ipopt is released as open source code under the Eclipse Public License (EPL).
         For more information visit https://github.com/coin-or/Ipopt
******************************************************************************

This is Ipopt version 3.14.16, running with linear solver ma27.

Number of nonzeros in equality constraint Jacobian...:        0
Number of nonzeros in inequality constraint Jacobian.:        1
Number of nonzeros in Lagrangian Hessian.............:        2

Total number of variables............................:        2
                     variables with only lower bounds:        2
                variables with lower and upper bounds:        0
                     variables with only upper bounds:        0
Total number of equality constraints.................:        0
Total number of inequality constraints...............:        1
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        1

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0  1.9999960e-04 5.10e-01 9.80e-01  -1.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1  1.1526117e-04 5.00e-01 2.62e+01  -1.7 2.62e-01    -  1.00e+00 3.77e-02h  1
   2  1.1923173e-04 5.00e-01 3.62e+04  -1.7 3.73e-01    -  1.00e+00 7.60e-04h  1
   3  1.1924119e-04 5.00e-01 3.05e+09  -1.7 2.59e-01    -  1.00e+00 1.18e-05h  1
   4r 1.1924119e-04 5.00e-01 1.00e+03  -0.3 0.00e+00    -  0.00e+00 5.89e-08R  2
   5r 2.3520927e+01 5.00e-01 1.30e-04  -0.3 4.84e+00    -  1.00e+00 1.00e+00f  1
   6r 5.9645822e+00 5.00e-01 9.50e-01  -1.0 2.41e+00    -  1.00e+00 1.00e+00f  1
   7r 2.8785403e+00 5.00e-01 2.95e-01  -1.7 7.46e-01    -  1.00e+00 1.00e+00f  1
   8r 1.3854380e+00 5.00e-01 1.03e-01  -2.5 5.20e-01    -  1.00e+00 1.00e+00f  1
   9r 7.3420281e-01 5.00e-01 3.46e-02  -3.8 3.20e-01    -  1.00e+00 1.00e+00f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10r 4.8913422e-01 5.00e-01 1.50e-09  -3.8 1.57e-01    -  1.00e+00 1.00e+00f  1
  11r 1.2429185e-01 5.00e-01 3.73e-03  -5.7 3.47e-01    -  1.00e+00 1.00e+00f  1
  12r 1.0106452e-01 5.00e-01 1.84e-11  -5.7 3.46e-02    -  1.00e+00 1.00e+00f  1
  13r 2.6157831e-02 5.00e-01 1.97e-04  -8.6 1.56e-01    -  1.00e+00 1.00e+00f  1
  14r 2.4339605e-02 5.00e-01 3.88e-14  -8.6 5.72e-03    -  1.00e+00 1.00e+00f  1
  15r 6.5611573e-03 5.00e-01 3.88e-14  -8.6 7.50e-02    -  1.00e+00 1.00e+00f  1

Number of Iterations....: 15

                                   (scaled)                 (unscaled)
Objective...............:   1.9148563642281853e-03    1.9148563642281853e-03
Dual infeasibility......:   8.7516499683580481e-02    8.7516499683580481e-02
Constraint violation....:   4.9999998000250589e-01    4.9999998000250589e-01
Variable bound violation:   9.9974940954384683e-09    9.9974940954384683e-09
Complementarity.........:   7.1935703593449981e-08    7.1935703593449981e-08
Overall NLP error.......:   4.9999998000250589e-01    4.9999998000250589e-01


Number of objective function evaluations             = 19
Number of objective gradient evaluations             = 6
Number of equality constraint evaluations            = 0
Number of inequality constraint evaluations          = 19
Number of equality constraint Jacobian evaluations   = 0
Number of inequality constraint Jacobian evaluations = 18
Number of Lagrangian Hessian evaluations             = 16
Total seconds in IPOPT                               = 0.010

EXIT: Converged to a point of local infeasibility. Problem may be infeasible.

Thanks for all you help thus far!

I tried using this, but my issue persists. I could try setting the environment variable before starting Julia like you did, but I am not sure how to collect all the paths. How did you generate that string? Did you join all the strings listed in Libdl.dllist()?

I just copy-pasted the output of this small script, but I forgot to mention that I had loaded HSL_jll.jl before running it:

using Libdl, HSL_jll

libs = Libdl.dllist()
dirs = unique(dirname(path) for path in libs)
separator = Sys.iswindows() ? ";" : ":"
library_paths = join(dirs, separator)

if Sys.iswindows()
    println("PATH=" * library_paths)
elseif Sys.isapple()
    println("DYLD_LIBRARY_PATH=" * library_paths)
else
    println("LD_LIBRARY_PATH=" * library_paths)
end

Then I closed Julia, copy-pasted the new environment variable LD_LIBRARY_PATH into my shell, launched Julia, and ran your code:

shell> LD=LIBRARY_PATH=...
shell> julia
using JuMP, Ipopt_jll, HSL, HSL_jll, AmplNLWriter

model = Model(() -> AmplNLWriter.Optimizer(Ipopt_jll.amplexe))
@variable(model, x[1:2] >= 0)
@objective(model, Min, sum(x.^2))
@constraint(model, sin(x[1]) <= -0.5)

set_optimizer_attribute(model, "hsllib", HSL_jll.libhsl_path)
set_optimizer_attribute(model, "linear_solver", "ma27")

optimize!(model)

For your question about how I generate that string, I just joined all the strings listed in Libdl.dllist() with a sepator : (on Linux) between them.

2 Likes

That did the trick, thanks!