Host node agent & capacity (infrastructure)
Updated 2026-06-12
Host node agent
Game servers run on host nodes (shared-capacity machines). Each node can run a lightweight HTTP agent that reports live metrics (CPU, RAM, network, load averages, health score) to the platform.
CEO Admin → Nodes & capacity
- Updated — timestamp of the latest metric sample stored for that node.
- Instances — active game instances placed on the node.
- Health — a platform-computed score from agent and placement signals.
Agent not responding
The node watchdog is only for game-capacity nodes. It skips readonly, protected, webserver, external inventory and Kubernetes worker rows, because those use separate infrastructure monitoring. For eligible game host nodes, the watchdog checks only the stored metrics-agent port and token, so a closed legacy/fallback port cannot create false incidents. If the agent is unreachable, it dispatches the node-agent deployment workflow with the node id, hostname, token and port so the remote systemd unit is reinstalled without drifting from the database identity. The production AI control agent on agent.gamehostcloud.com is a different stack: it runs behind the production HTTPS route and must not be treated as a game-node metrics agent. Its recovery uses the dedicated AI-agent repair workflow: the platform syncs the current bridge token, refreshes /opt/ghc-agent, restarts only the AI agent and verifies /ready + protected /status without asking the user to run commands.
Kubernetes (rollout)
When workers run on Kubernetes, node tags (for example tag:k8s-worker) tie scheduling into the same capacity model. See your team’s internal deployment docs for current status.