broker start timeout when Mesos agent start with gpu/nvidia isolation

[rhhliu]

When mesos agent start with --isolation=cgroups/devices,filesystem/linux,gpu/nvidia
For example:
sudo ./bin/mesos-agent.sh --ip=192.168.0.251 --no-hostname_lookup --master=192.168.0.215:5050 --work_dir=/work/data/mesos-slavels2 --no-switch_user --isolation=cgroups/devices,filesystem/linux,gpu/nvidia

then the broker start will failed with timeout.
$ ./kafka-mesos.sh broker start 1
Error: broker start timeout

we found the gpu/nvidia isolation is the real cause. Without gpu/nvidia, everything works fine.