Nondeterministic kSolveError #2810
Description
I have a MILP problem (Thermodynamically constrained FBA, computing maximum feasible reaction delta gibbs energy),
which i have implemented though cobrapy + optlang, which below the hood eventuallyh used highpy, and thus HiGHS.
It seems to have a nondeterministic error that comes up with a kSolveError (which is immensely unspecific),
and nothing in the logs.
It seems like it has a probability of occurring maybe 1/1000 solves.
Which us common enough that when computing the ranges for all reaction delta gibbs in ecoli I see it 1 or 2 times.
If i do things like
- force optlang to rebuild the problem by changing the solver to glpk then changing it back to hybrid (highs)
- change the tolerance significantly
- add a linear scaling factor to the objective
It normally goes away (except 1 time in 1000).
But it is very weird. And unclear where it is from
There are a few places in the logs it can be output.
I am not sure if perhaps optlang is corrupting the state or it is a bad initialization or what.
I dumped out the JSON from optlang, but i don't know how easy that is to load.
ecoli_thermo.json.gz