VictoriaMetricsClient (PromQL instant-query, 5s-Timeout), Home-Seite mit 2-s-Live-Takt: CPU (2-min-Rate), RAM und Root-Disk belegt/total/Prozent, parallel abgefragt. AsciiBar-Komponente (18 Zeichen, Schwellenfarben gruen/gelb/rot passend zu den vmalert-Severities). Fallendokument: double.TryParse der PromQL-Strings lief kulturabhaengig — de-DE liest den Punkt als Gruppentrenner (9.71 -> 9713020833333342); Parsing jetzt invariant (NumberStyles.Float + InvariantCulture). Generated by Mistral Vibe. Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
Monitoring workload
Host-wide monitoring on ghostnet-ki: system metrics (CPU, RAM, disk, GPU
utilization, VRAM) collected by a host-side metrics collector and visualised by
a dashboard container. The workload is separated from the encoding and LLM
workloads; the dashboard currently shows the encoding pipeline and is
extensible to LLM metrics.
Architecture and decisions are documented under
../../docs/platform/llm-server/monitoring/.
The executable Ansible control project under ansible/ entsteht mit der
Implementierung.