NVIDIA / NVIDIA/NemoClaw

[Ubuntu 24.04][Onboard] Brave-preset sandbox onboard times out: "Sandbox … did not become ready within 180s" after 500+s gateway-image upload

Open
#3,344 7 comments 0 reactions 1 assignee Claimed by @cjagwani View on GitHub
area: onboarding area: sandbox integration: brave NV QA platform: ubuntu
Dominant language
TypeScript
Stars
22.5k
Forks
3.1k
Avg merge
1d 1h
Merged PRs (30d)
715

Description

## Description

Description

On NemoClaw v0.0.38 (Ubuntu 24.04, RTX 6000 Ada), running `nemoclaw onboard --fresh --name brave-test` with Ollama (qwen2.5:7b) provider and `BRAVE_API_KEY` set in the shell so the Brave web-search preset is enabled, the wizard exports the sandbox image successfully and uploads it to the OpenShell gateway, but then the in-gateway sandbox-create step takes much longer than the wizard's 180s readiness check — the wizard concludes "Sandbox 'brave-test' was created but did not become ready within 180s. The orphaned sandbox has been removed — you can safely retry." and the resulting registry has no sandbox.

The image-upload phase alone logged "Still uploading image into OpenShell gateway... (525s elapsed)" before "Image openshell/sandbox-from:1778494022 is available in the gateway." — so the cumulative onboard time on this host is roughly 9 min image-upload + 3 min sandbox-ready wait. The 180s timeout is hit even though the sandbox was just created seconds earlier.

Net effect: onboard with the Brave preset enabled fails on Ubuntu 24.04 + RTX 6000 Ada with the latest installer despite a healthy gateway, Ollama running, and a valid Brave API key — making the Web-Search day0 path (T560826) un-passable without a manual retry / longer timeout.

Environment
Device:        Ubuntu workstation 2u1g-x570-1795 (10.63.136.90)

OS: Ubuntu 24.04.4 LTS (kernel 6.17.0-19-generic)
Architecture: x86_64
GPU: NVIDIA RTX 6000 Ada Generation, 46068 MiB
Node.js: v22.22.2
npm: 10.9.7
Docker: 29.4.3 (build 055a478)
OpenShell CLI: openshell 0.0.36
NemoClaw: v0.0.38
OpenClaw: 2026.4.24 (cbcfdf6)
Provider: Local Ollama, model qwen2.5:7b (4.7 GB)
Preset: brave + npm/pypi/huggingface/brew/local-inference auto-applied
BRAVE_API_KEY: 31-char BSA-prefixed key (provided in shell)
Steps to Reproduce
1. Fresh-ish host with NemoClaw v0.0.38 installed; gateway 'nemoclaw'

already running healthy; Ollama installed and running with qwen2.5:7b.
2. export BRAVE_API_KEY=BSA...
3. Run: nemoclaw onboard --fresh --name brave-test
4. Wizard prompts (this run):
[3/8] inference: Choose [1]: 7 (Local Ollama)
Choose model [1]: (default qwen2.5:7b)
"Apply this configuration? [Y/n]:" → Enter
"Enable Brave Web Search? [y/N]:" → y
(Brave key prompt did NOT appear in this run — BRAVE_API_KEY env
was already set, so the wizard accepted it silently)
[5/8] Messaging channels → Enter (skip)
... policies / preset stage continues ...
[8/8] Building sandbox image (image export, upload)
5. Observe: "Still uploading image into OpenShell gateway... (525s elapsed)"
6. After ~9 min, "Image openshell/sandbox-from: is available in the gateway"
7. Wizard then: "Starting NIM container..." (well, OpenClaw sandbox container)
then: "Waiting for sandbox to become ready..."
8. After 180s, wizard prints failure and removes the orphan.
Expected Result
After a successful sandbox image upload to the gateway, the readiness

wait must give the sandbox enough time to finish initialization
(Ollama proxy bring-up + OpenClaw agent boot + brave preset egress
rules + policy version load) on a normal workstation. Either:
(a) extend the per-readiness-poll deadline (180s → e.g. 600s) so
large preset stacks have time to come up, OR
(b) actively surface what subsystem is still not-yet-ready so the
user knows whether to wait or abort, OR
(c) decouple readiness from a fixed deadline and let the user
`nemoclaw status` afterwards to inspect.

Either way, on a non-pathological host (40+ GB free VRAM, fast disk,
healthy gateway, valid API key), `nemoclaw onboard --fresh` with
the brave preset enabled should produce a Ready sandbox in
`nemoclaw list` on the first run.

Actual Result
Wizard tail:

[progress] Exported 609 MiB
[progress] Uploaded to gateway
Image openshell/sandbox-from:1778494022 is available in the gateway.
Still uploading image into OpenShell gateway... (525s elapsed)
Create stream exited with code 1 after sandbox was created.
Checking whether the sandbox reaches Ready state...
Waiting for sandbox to become ready...
✓ Deleted sandbox brave-test
Sandbox 'brave-test' was created but did not become ready within 180s.
The orphaned sandbox has been removed — you can safely retry.
Retry: nemoclaw onboard

$ nemoclaw list
Sandboxes:
my-assistant * (from a separate earlier silent-install run)
$ # no 'brave-test' present

Logs
Full per-step capture: /home/lab/day0-automation/20260511/report-T560826.txt

Driver: /home/lab/day0-automation/20260511/drive_T560826.py
Onboard state-machine transitions:
[10:06:33] inference (selected Ollama at item 7)
[10:06:40] model menu (Enter, default qwen2.5:7b)
[10:06:47] yn (Apply config → Enter)
[10:06:53] brave_yn (Enable Brave Web Search → y)
[10:07:00] messaging (Enter, skip)
[10:07:06] wait (image build + upload + sandbox-create)
[10:18:52] bash (wizard ended; "did not become ready within 180s")

Note: another sandbox 'my-assistant' DOES carry the brave preset
(applied by an earlier silent-install run T560899), so the brave
preset itself can be applied successfully. The defect is specifically
the 180s readiness wait timing out in the fresh-onboard path.

## Bug Details

| Field | Value |
|-------|-------|
| Priority | Unprioritized |
| Action | Dev - Open - To fix |
| Disposition | Open issue |
| Module | Machine Learning - NemoClaw |
| Keyword | NemoClaw, NEMOCLAW_GH_SYNC_APPROVAL, NemoClaw_Onboard, NemoClaw_Sandbox |

---
[NVB#6164319]

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.