Don't save state for crashing MPI simulations

CliMA · Dec 22, 2023 · 6971880 · 6971880
1 parent 573f6fd
commit 6971880
Showing 1 changed file with 7 additions and 2 deletions.
diff --git a/src/solver/solve.jl b/src/solver/solve.jl
@@ -68,8 +68,13 @@ function solve_atmos!(simulation)
             return AtmosSolveResults(sol, :success, walltime)
         end
     catch ret_code
-        CA.save_restart_func(integrator, simulation.output_dir)
-        CA.save_to_disk_func(integrator, simulation.output_dir)
+        if !CA.is_distributed(comms_ctx)
+            # We can only save when not distributed because we don't have a way to sync the
+            # MPI processes (maybe just one MPI rank crashes, leading to a hanginging
+            # simulation)
+            CA.save_restart_func(integrator, simulation.output_dir)
+            CA.save_to_disk_func(integrator, simulation.output_dir)
+        end
         @error "ClimaAtmos simulation crashed. Stacktrace for failed simulation" exception =
             (ret_code, catch_backtrace())
         return AtmosSolveResults(nothing, :simulation_crashed, nothing)