I tried copying the Triton libraries from the DS 6 triton
image to the base
image, but I’m getting the following error:
libtriton_tensorflow1.so: /lib/x86_64-linux-gnu/libm.so.6: version
GLIBC_2.29’ not found`
I tracked it down to being caused by Triton being compiled on Focal with a higher libc6
version, while the devel
and base
images run on Bionic with libc6
version 2.27. First, I don’t understand why. Given the enormous size of the triton image, it’s impossible to use it in production, but that’s fine since there is a base image. Then, my question is why on earth is the triton
image not based on Bionic, or why not finally upgrade to Focal for devel
and base
? Does this mean that the dependencies listed on the DeepStream Quick Start page are meaningless? Could Nvidia perhaps share the Dockerfile for the triton
image? If so, everyone could first of all compile it with the right libc6
version and then just customize it with the needed backends.