Julia not using the most recent version of Gurobi

Specific Domains Optimization (Mathematical)
1

I have been using Gurobi 9.0.1 in Julia with JuMP. Now I have installed Gurobi 9.1.2, but the output of Gurobi in Julia indicates that it is still using 9.0.1. How do I make it use the most recent version of Gurobi?

I checked ENV["GUROBI_HOME"] in Julia and it is pointing to Gurobi 9.1.2

By the way, after I installed Gurobi 9.1.2, I have not updated or rebulit my Gurobi.jl whose version is Gurobi v0.9.3. Do I need to re-do again something like Pkg.build("Gurobi") or reinstall it?

Thanks.

2

Do I need to re-do again something like Pkg.build("Gurobi") or reinstall it?

Yes, I think this should help. Looking at Gurobi.jl/deps/build.jl, we can see that the build process stores a link to the shared library to be used by Gurobi.jl. Maybe the old deps.jl should be deleted as well, before running build.

1 Like
3

If you update Gurobi, you need to re-run Pkg.build. No need to delete old files.

2 Likes
4

Thank you.

If I want to change between different versions of Gurobi, do I just change the ENV["GUROBI_HOME"] to the corresponding Gurobi version I want to use, and then re-run Pkg.build("Gurobi")?

1 Like
5

Yes

6

Sorry for bringing up this ancient thread, but I feel my issue does not deserve a new one. To fix a Gurobi issue discussed here, I need to switch from 12.0.0 to 12.0.1. ENV["GUROBI_HOME"] points to 12.0.1, and I did Pkg.build("Gurobi"), but checking the version, as described here, I’m still on 12.0.0.

Is 12.0.1 just not supported yet, or is there another issue?

7

Hi Leo, to opt-in to local installations, you know need to additionally set ENV["GUROBI_JL_USE_GUROBI_JLL"] = "false".

See this part of the README Gurobi.jl/README.md at master · jump-dev/Gurobi.jl · GitHub

I’ll see if we can update the automatic binaries: Add support for v12.0.1 by odow · Pull Request #33 · jump-dev/Gurobi_jll.jl · GitHub

1 Like
8

12.0.1 should be automatically installed if you update your packages. No need for GUROBI_HOME.

9

Thanks Oscar! That worked for me.

10

Sorry to go a bit off topic, but is it technically possible without too much work to add support for hot-swapping solver versions without rebuilding and/or updating the packages? The reason I ask is that solver performance is model specific and doesn’t always strictly improve with new versions. So e.g. for some energy models we have in house Gurobi v10 consistently beat Gurobi v11. So it would be nice to be able to choose one version or the other at runtime.

Maybe this could work as a special solver option that gets intercepted by Gurobi.jl and reroutes to the correct binary, or something similar. Not asking the JuMP team to do it, but if you think it’s doable without knowing the full depth of JuMP internals then I might look into it myself.