nre-ga:26.04 renderer fails immediately with `init_fs_encoding` / `source code string cannot contain null bytes

[jinlin1013]

nre-ga:26.04 renderer fails immediately with init_fs_encoding / source code string cannot contain null bytes

Summary

On a fresh local AlpaSim tutorial run (deploy=local topology=1gpu driver=vavam), the renderer-0 container exits immediately. The same failure reproduces outside AlpaSim by running the NRE entrypoint alone, so this does not appear to be a wizard/config issue.

Minimal reproduction

docker run --rm --entrypoint bash nvcr.io/nvidia/nre/nre-ga:26.04 -c '/app/run --help'

Also fails via AlpaSim:

uv run alpasim_wizard deploy=local topology=1gpu driver=vavam wizard.log_dir=$PWD/tutorial

Actual behavior

Fatal Python error: init_fs_encoding: failed to get the Python codec of the filesystem encoding
Python runtime state: core initialized
SyntaxError: source code string cannot contain null bytes

The container exits with code 1. Docker Compose then aborts the rest of the stack (runtime/driver/physics/controller receive SIGTERM / exit 143).

Expected behavior

/app/run --help should print help / start successfully, and the AlpaSim tutorial renderer should come up.

Environment

• Host OS: Ubuntu 22.04.5 LTS (6.8.0-40-generic)
• Docker: 29.6.1
• NVIDIA driver: 580.159.03 (CUDA reported by nvidia-smi: 13.0)
• NVIDIA Container Toolkit: 1.19.1
• GPU: NVIDIA GeForce RTX 3060 12GB
• Image: nvcr.io/nvidia/nre/nre-ga:26.04
• Image digest: sha256:1c3a838440fce96258e055615b146fa8f75ad2a799e131aaa480755592317d1f
• AlpaSim: cloned from https://github.com/NVlabs/alpasim (local package image alpasim-base:0.111.0)

Notes / checks already done

1. Host GPU access in the NRE container works:

docker run --rm --gpus all --entrypoint bash nvcr.io/nvidia/nre/nre-ga:26.04 -c 'nvidia-smi'

2. Re-pull reports the image is already up to date at the same digest; failure persists.
3. Failure happens before any AlpaSim scene/driver workload starts.
4. Secondary unrelated warning during Compose run: Prometheus/dcgm-exporter fails to parse dcgm-counters.csv (wrong number of fields on line 15). This is not what aborts the run; renderer-0 exiting is the primary failure.
5. NuRec hardware docs mention >24GB VRAM. This machine has 12GB, but the crash occurs during Python interpreter initialization, before GPU rendering work.

Question

Is nre-ga:26.04 known to fail on consumer GPUs / this Docker version with the above null bytes error? If so, is there a supported workaround (different NRE tag, env vars, or host requirements)?