This list of solvers that have worked fairly consistently, try them out locally and if the recovery solver is not activated, you should be safe for longer transient runs:

% biconjugate gradient with block Jacobi preconditioner [Thiago's favorite]
md.toolkits.DefaultAnalysis=bcgslbjacobioptions();

% gnres with block jacobi preconditioner [Martin's favorite]
md.toolkits.DefaultAnalysis=gmresbjacobioptions();
md.toolkits.DefaultAnalysis.ksp_max_it=1;

% ASM, used for mass conservation but may work for other things
md.toolkits.DefaultAnalysis=asmoptions();

% An iterative solver may work for one analysis and not another, you can assign a solver to each, like this:
md.toolkits.StressbalanceAnalysis = gmresbjacobioptions();
md.toolkits.DefaultAnalysis=issmmumpssolver();

Make sure to always keep MUMPS as the recovery solver (it is slow but extremely robust). You probably don't need to do this because MUMPS should already be set as the recovery solver, but just in case:

%Keep recovery mode as MUMPS
md.toolkits.RecoveryAnalysis=mumpsoptions();