Re: [sumo-user] Assorted errors when using threads (1 new simulation per

[Marcelo Andrade Rodrigues D Almeida <md@xxxxxxxxx>]

Hi Jakob

Thank you for your response

"As far as I can tell you are starting a single sumo simulation and are trying to use it from multiple threads."

Maybe the example is not as simple as I wanted. I'm actually doing this. I open a simulation then I create two threads that start their own simulation (signalized by the "env.reset(execution_name)" command).

"The traci operations are not thread safe. This is what works

- starting a different sumo simulation for each thread and using them as described in https://sumo.dlr.de/docs/TraCI/Interfacing_TraCI_from_Python.html#controlling_parallel_simulations_from_the_same_traci_script"

I'm already doing this. I give them labels (with a unique execution_name)

"- using your own synchronization so that only one thread is talking to sumo at the same time"

I used it around the start command. After starting, they should be thread-safe, since I'm using the labels with getConnection(), right? Or do I need to add synchronization for every use?

Sincerely,

Marcelo d'Almeida

On Mon, Apr 26, 2021 at 7:00 AM Jakob Erdmann <namdre.sumo@xxxxxxxxx> wrote:

As far as I can tell you are starting a single sumo simulation and are trying to use it from multiple threads.

The traci operations are not thread safe. This is what works

- starting a different sumo simulation for each thread and using them as described in https://sumo.dlr.de/docs/TraCI/Interfacing_TraCI_from_Python.html#controlling_parallel_simulations_from_the_same_traci_script

- using your own synchronization so that only one thread is talking to sumo at the same time

regards,

Jakob

Am So., 25. Apr. 2021 um 22:46 Uhr schrieb Marcelo Andrade Rodrigues D Almeida <md@xxxxxxxxx>:

Hi everyone

I'm having trouble using multiple simulations in different threads

The difference can be perceived by changing max_workers to 1 and 2.

"        with ThreadPoolExecutor(max_workers=len(possible_actions)) as executor:" (planning_only_agent.py, line 100)

[See attached example (a very simplified version of what I'm actually using)]

With max_workers=1, it works just fine.

With 2, it gives all sort of errors:

"bad char in struct format"

"unpack requires a buffer of [x] bytes"

A lot of times it just hangs

"Received answer 0 for command 211."

The easiest way I found to reproduce is to add a breakpoint into a subscribe call and hit continue several times. It is likely to crash when threads stop at the same command, although it occurs eventually without breaking points.

Tested in version 1.9.0 (specifically v1_9_0+0210-fa53508669)

Thank you in advance

Sincerely,

Marcelo d'Almeida

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user