Gateway Stability and npm Update Failure

October 6, 2026

Topic: OpenClaw gateway operations, npm update path on chonk.

Session Activity

Gateway remained stable throughout the day after an earlier restart. The session ran as agent:main with active Ollama model serving (qwen3.6:latest). No external API calls were made — all inference routed through local ollama on chonk.

mcs-manager Infrastructure Check

Minecraft server probe across lab subnets returned negative results:

DNS minecraft.lab.bitsmasher.net — no resolution
Port 25565 on bitsmasher.net — connection refused
MariaDB blowfish: minecraft database still exists (orphaned, possibly retainable for restore)

Conclusion: the Minecraft server is offline/decommissioned. The DB schema remains if a future restore is needed.

npm Auto-Update Failure

Gateway restart triggered an global-install-permission-denied during auto-update. The gateway itself recovered fine — it's running on pid 2527 post-restart. The update mechanism for the global /usr/lib/node_modules/openclaw/ install is broken, likely because OpenClaw was installed via npx -y openclaw@latest which lacks root-write capability to that path.

Resolution: requires sudo npm install -g openclaw@latest on chonk. Not urgent — gateway function is unaffected.

Token Usage & Load Metrics

Ollama does not expose per-session token counts via its API. Available load data from ollama ps:

Model	Size	Processor	Context
qwen3.6:latest	22 GB	40%/60% CPU/GPU	32768

The primary model (qwen3.6) was loaded at ~60% GPU utilization with full 32k context window during this session. No secondary models were active.

Note: Token usage tracking requires either gateway-level telemetry or manual counting from session logs. The workspace-steward skill defines static token ceilings (SOUL.md <300 tok, AGENTS.md <500 tok, MEMORY.md <1000 tok) but does not track per-session inference tokens.

Model Inventory on chonk

Model	Size (MB)
qwen3.6:latest	21,573
devstral-small-2:latest	14,474
nimble:latest	9,086
qwen2.5-coder:14b	8,571
qwen2.5-coder:latest	4,466
gemma3:latest	3,184
tev1:latest	4,397
nomic-embed-text:latest	261

Total local model footprint: ~67 GB across chonk's disk. This is the complete inference fleet with no external model provider dependency active.