Closed btalb closed 4 years ago
There is a errror when executing this command of "./IsaacSimProject "
./IsaacSimProject
unable to read vr path registry from openvrpaths.vrpath
Originally posted by @liucsg in https://github.com/RoboticVisionOrg/benchbot/issues/2#issuecomment-616300070
Hi, btalb, There is collapse after running IsaacSimProject a short time . I don't know where the problem is. Could you export the images of "benchbot/simulator:base" as a compressed file and share with me ? Thank you!
Originally posted by @liucsg in https://github.com/RoboticVisionOrg/benchbot/issues/2#issuecomment-617167398
@liucsg: My understanding is that this issue is caused due to a custom installation process as you weren't able to run benchbot_install
due to network issues.
The behaviour of the simulator crashing is not particularly transparent to the user, as this is expected to be setup correctly through the Docker installer. I am currently working on updates though that will clearly display Simulator errors.
Unfortunately, all I can offer at the moment is a process to check through to try & understand what is causing the crash (I suspect it is Vulkan related):
benchbot_run
a debugging container is created. Attach to this container while BenchBot is running:
docker attach benchbot_debug
In the debug container, confirm X windows are successfully being forwarded to the host (a set of eyes should appear on your screen):
sudo apt install -y x11-apps && xeyes
nvidia-smi
vkcube
If this fails, check their tool for more details:
vulkaninfo
Let me know if these all work, then we can go from there.
@btalb Hi,btalb: I has runned the above process .There is a error when checking Vulkan capabilities of the container.
benchbot@benchbot_debug:/benchbot$ vkcube vkcube: /build/vulkan-tools-1.2.135.0~rc2/cube/cube.c:3488: demo_init_vk_swapchain: Assertion `!err' failed. Aborted (core dumped) benchbot@benchbot_debug:/benchbot$ vulkaninfo error: XDG_RUNTIME_DIR not set in the environment. ERROR at /build/vulkan-tools-1.2.135.0~rc2/vulkaninfo/vulkaninfo.h:240:vkGetPhysicalDeviceSurfacePresentModesKHR failed with ERROR_INITIALIZATION_FAILED
docker run --gpus all -it benchbot/simulator:base vulkaninfo Here's the output of vulkaninfo on a working machine.
'DISPLAY' environment variable not set... skipping surface info error: XDG_RUNTIME_DIR not set in the environment.
VULKANINFO
Vulkan Instance Version: 1.2.135
Instance Extensions: count = 17
VK_EXT_acquire_xlib_display : extension revision 1
VK_EXT_debug_report : extension revision 9
VK_EXT_debug_utils : extension revision 1
VK_EXT_direct_mode_display : extension revision 1
VK_EXT_display_surface_counter : extension revision 1
VK_KHR_device_group_creation : extension revision 1
VK_KHR_display : extension revision 21
VK_KHR_external_fence_capabilities : extension revision 1
VK_KHR_external_memory_capabilities : extension revision 1
VK_KHR_external_semaphore_capabilities : extension revision 1
VK_KHR_get_display_properties2 : extension revision 1
VK_KHR_get_physical_device_properties2 : extension revision 1
VK_KHR_get_surface_capabilities2 : extension revision 1
VK_KHR_surface : extension revision 25
VK_KHR_surface_protected_capabilities : extension revision 1
VK_KHR_xcb_surface : extension revision 6
VK_KHR_xlib_surface : extension revision 6
Layers: count = 6
Okay, this is good. We have identified the issue as broken Vulkan support in your manual installation.
Note that the second result is in a docker container that has been started without UI support (no forwarding of the DISPLAY
variable or mounting of /tmp/.X11-unix
). It is important when doing Vulkan-related tests to first ensure that the container is capable of displaying wiindows (i.e. checking you can make a window popup from within the container like the commands above). That's why I recommend doing all of your testing in the running benchbot_debug
container.
As for the actual Vulkan issue; I don't know how much more help I can give remotely, especially given this installation was done manually / outside of benchbot_install
. Investigating Vulkan issues can be difficult. A lot of work is done in the install process to ensure Vulkan works, it may be worth trying to check through & see what is missing from your installation.
One thing worth confirming is how your system is physically setup. Is it a standard desktop setup (screen plugged directly into Nvidia GPU, with Nvidia GPU as the primary graphics card)? We have had issues getting Vulkan working properly on truly headless machines (no displays attached), over X window forwarding (as the rendering is done on the client side), & when using the Nvidia GPU as a secondary card (like with bumblebee
, primus-select
, etc.).
Hi btalb, I have sovled the problem of vulkan through installing Nvidia-driver and cuda multiple times.
But, there is a problem when running the cmd of "./bazelros run //apps/benchbot_simulator"
2020-04-28 09:15:35.398 WARNING external/com_nvidia_isaac/engine/alice/components/TcpSubscriber.cpp@162: Failed to connect to remote. Will try again in 0.500000 seconds.
Also, it shows as following when running the cmd of "benchbot_submit --containerised
Waiting to establish connection to a running supervisor ... Connected! Waiting to establish connection to a running simulator ...
Hi,btalb,I have solved this problem !
Hi btalb, I canot run simulator. It is always waiting to establish connection to a running simulator ... I excute two commands ,one is benchbot_run, the other is benchbot_submit,as following,
1) benchbot_run --env miniroom:1 --task semantic_slam:active:ground_truth -f 2) benchbot_submit --containerised ./examples/hello_active/ Meanwhile there are two log, one is benchbot_run.log, the other is benchbot_submit.log. benchbot_run.log benchbot_submit.log
Originally posted by @liucsg in https://github.com/RoboticVisionOrg/benchbot/issues/2#issuecomment-615123500