Describe the bug
We currently have the problem, when running Mujoco on a compute cluster without GPU and without Display, so everything offscreen, that we get an "ERROR: Could not allocate offscreen framebuffer" error frequently, that kills our process.
This Error is also displayed in the MUJOCO_LOG.txt
We found that having more than one instance of mujoco open increases the probability of the error, however, we should have far more than enough memory left during the error.
To Reproduce
Difficult to do, since we were only able to reproduce it on our compute cluster
Desktop (please complete the following information):
output of: echo $HOME: the home folder of the user
output of: echo $USER: the appropriate user
output of: echo $LD_PRELOAD: /usr/lib64/libEGL.so
Additional context
We do want to render an instance of a virtual camera after a rollout, however, often the error happens far before that.
We understand that this error is probably very dependent on our setting and cluster, however, we hope that somebody here can point us toward the right direction to get to a solution
Describe the bug We currently have the problem, when running Mujoco on a compute cluster without GPU and without Display, so everything offscreen, that we get an "ERROR: Could not allocate offscreen framebuffer" error frequently, that kills our process.
This Error is also displayed in the MUJOCO_LOG.txt
We found that having more than one instance of mujoco open increases the probability of the error, however, we should have far more than enough memory left during the error.
To Reproduce Difficult to do, since we were only able to reproduce it on our compute cluster
Desktop (please complete the following information):
Environment
echo $LD_LIBRARY_PATH
:[home]/.mujoco/mujoco210/bin:/usr/lib/nvidia
echo $HOME
: the home folder of the userecho $USER
: the appropriate userecho $LD_PRELOAD
:/usr/lib64/libEGL.so
Additional context We do want to render an instance of a virtual camera after a rollout, however, often the error happens far before that.
We understand that this error is probably very dependent on our setting and cluster, however, we hope that somebody here can point us toward the right direction to get to a solution