santoshpanda15 / rl-glue-ext

Automatically exported from code.google.com/p/rl-glue-ext
0 stars 0 forks source link

Experiment and Environment CPU 100% of Python agent dies in agent_init #77

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago
Summary: Using the python codec, when an environment crashes in the env_init() 
or env_start() 
methods, it leaves the agent and experiment running, both consuming all 
available CPU.

Background: I am using RL Glue in the context of a large, real-world 
environment.

I start the components like this:

 rl_glue &
 python agent.py &
 python experiment.py &
 python environment1.py &

In the course of debugging, my environment code often crashes in the env_init() 
or env_start() 
methods.  When this happens, the following two processes are left behind, each 
consuming all 
available CPU:

 python agent.py
 python experiment.py

By contrast, when my code crashes in the env_step() method, all of the RL Glue 
processes stop, 
as desired.

The desired behavior is for all of the components to die when one of the 
components dies.  
Ideally, none of my code should ever crash!  But in the course of development, 
testing, and 
debugging of a real system, it is inevitable that crashes will occur.  I am 
hoping that there is a 
possible fix which will allow the other components to behave more gracefully.

Platform:

linux
HOSTTYPE=x86_64
Python 2.5.1
-bash-3.2$ rl_glue --version
RL-Glue Version 3.04, Build 909
python codec

Code to reproduce:

./run1.sh : env dies in env_init(); agent and experiment peg CPU (bug)
./run2.sh : env dies in env_start(): agent and experiment peg CPU (bug)
./run3.sh : env dies in env_step(): agent and experiment die (as desired)

Original issue reported on code.google.com by brian.ta...@gmail.com on 12 Jun 2009 at 9:25

GoogleCodeExporter commented 8 years ago
Above is the original bug report that I got.  The first thing to do will be to 
verify this by reproduction, and then to 
see what happens if we mix and match agents and environments in different 
languages: we need to find out of 
this is a RL-Glue problem or a codec problem.

Here are the files that Jason provided to reproduce the issue.

Original comment by brian.ta...@gmail.com on 12 Jun 2009 at 9:26

Attachments: