SSACSolver freezes until pressing Ctrl+C

[maurosilber]

Hi,

I was running a simulation with the SSACSolver, and noticed that sometimes it gets stuck while reading from stdin in this line:

GillesPy2/gillespy2/solvers/cpp/c_decoder.py

Line 244 in de40702

line = output.read(page_size).decode("ascii")

After pressing Ctrl+C, it resumes, reading and decoding some more lines. Then, it "terminates" early and throws a ValueError: read of closed file from the line shown above.

I'm not really sure if it is terminating because it reached some stopping condition for stochastic simulations or some solver error. But, it is odd that the last time point reaches a different value in subsequent runs with the same seed (for instance, 22672.7, 22762.8, 22882.9 or 23003.0).

The code to reproduce the simulation would be unnecessarily complex to share, as I'm translating a model from other framework to gillespy2, but I can share the generated files (sim-files.zip) from which the model is compiled and run with this line:

GillesPy2/gillespy2/solvers/cpp/c_solver.py

Line 217 in de40702

with subprocess.Popen(sim_args, stdout=subprocess.PIPE, **platform_args) as simulation:

The sim_args are: GillesPy2_Simulation.exe --trajectories 1 --timesteps 1000 --end 30000.0 --interval 1000 --seed 3. Running this outputs several numbers to stdout, freezes, and then outputs some more after Ctrl+C. So, maybe this is enough to reproduce the behaviour?

I'm running this on a Mac M2 in a conda environment with this package versions:

Package    Version  Build               Size       Kind   Source
gillespy2  1.8.2    pyhd8ed1ab_0        250.4 KiB  conda  gillespy2-1.8.2-pyhd8ed1ab_0.conda
python     3.12.5   h30c5eda_0_cpython  12.3 MiB   conda  python-3.12.5-h30c5eda_0_cpython.conda
scons      4.7.0    py312h81bd7bf_0     2.6 MiB    conda  scons-4.7.0-py312h81bd7bf_0.conda

Let me know if you need anything else to reproduce (or I'm doing anything wrong).

Thanks!
Mauro

[jtcooper10]

Hi! It sounds like the simulation isn't frozen, it's running as expected but very slowly. Note the --timesteps 1000 ... --interval 1000 flags; this means "run for 1000 timesteps and dump the results every 1000 timesteps," which is just a long-winded way of saying "wait until the simulation has completed to dump results to Python." This is the default behavior of a C solver, unless live graphing is used.

There are a few options for investigating this:

• Try setting the end time to something much lower and see if it completes.
• Try a more efficient solver, like the TauHybridCSolver -- if your model has high population values during the simulation, the hybrid solver will be far better equipped than the basic SSA direct solver.