Potential crash when frontend is not fast enough

[edwin0cheng]

I found that in the code below:

generative_agents/reverie/backend_server/reverie.py

Line 138 in 0db6021

init_env_file = f"{sim_folder}/environment/{str(self.step)}.json"

Reverie::__init__ will try to open the current step environment file which is provided by frontend server.

But I saw in

generative_agents/reverie/backend_server/reverie.py

Line 333 in 0db6021

if check_if_file_exists(curr_env_file):

It will wait for the frontend to produce it.

But Reverie::__init__ is called before Reverie::start_server, such that if the frontend is not fast enough, Reverie::__init__ one will crash.

Do I miss something here ?

[LuisESV]

This might be a different issue but when running ./run_backend_automatic.sh -o base_the_ville_isabella_maria_klaus -t test_1 -s 4 --ui False it will stop the simulation every 200 steps (When saving) - Is that expected?

[drudilorenzo]

This might be a different issue but when running ./run_backend_automatic.sh -o base_the_ville_isabella_maria_klaus -t test_1 -s 4 --ui False it will stop the simulation every 200 steps (When saving) - Is that expected?

This is a wanted behavior (by me). Reason:

• The code was full of bugs.
• I really needed something working, and I hadn't time for investigating and solving every problem.
• If the simulation fails after 2000 steps you loose lots of hours. Thus, i decided to divide the simulation in chunk by saving every 200 steps. In this case, even if it fails, it is possible to restart from a checkpoint not too far in time.

Does it make sense?

[LuisESV]

That makes sense. Having said that, It should continue automatically right? OR is it expected for us to reload the browser?

[drudilorenzo]

That makes sense. Having said that, It should continue automatically right? OR is it expected for us to reload the browser?

It should automatically restart a new simulation. Here the code the make it happen: https://github.com/drudilorenzo/generative_agents/blob/fix-and-improve/reverie/backend_server/automatic_execution.py#L93