monitoring: llama-swap GPU/LLM stack (v250) — PrometheusRule, Grafana dashboard, scrape config, VRAM exporter

ansible/roles/llm-inference-multimodel/tasks/main.yml

@@ -60,3 +60,37 @@
 - include_tasks: preset.yml
   when: llm_router_preset_enabled | default(false)
   tags: [always]
+
+# Phase S — llama-swap mode hot-swap proxy (port 8001)
+# Gates on llm_swapmode_enabled (default false — complete no-op until enabled).
+# Replaces router mode entirely: single Go binary + YAML config, no INI presets.
+# Additive deployment (non-invasive); production router (port 8002) stays running during Phase 1 shadow.
+# 
+# When llm_swapmode_enabled: true, this phase:
+#   swapmode_binary  — download + install llama-swap binary
+#   swapmode_config  — render config.yaml.j2 template
+#   swapmode_systemd — deploy llama-swap.service unit
+#   swapmode_firewall — open port 8001 scoped to Hermes subnet
+#   swapmode_verify  — start service, run validation gates
+#
+# Added 2026-08-18 (t_c1e44190): llama-swap Phase 3 Ansible integration — Wong.
+- include_tasks: swapmode.yml
+  when: llm_swapmode_enabled | default(false)
+  tags: [always]
+
+# Phase M — GPU/LLM Monitoring (VRAM exporter + Prometheus + Grafana)
+# Gates on llm_monitoring_enabled (default true — but can be disabled per-host).
+# Deploys:
+#   - VRAM textfile exporter script (runs every minute via cron)
+#   - Prometheus scrape config template (for GitOps deployment)
+#   - Grafana dashboard JSON template (for GitOps deployment)
+#   - PrometheusRule alert rules template (for GitOps deployment)
+#
+# No cluster-facing changes here; templates are staged for manual review
+# and committed via Git. ArgoCD syncs them automatically afterward.
+#
+# Reference: roles/llm-inference-multimodel/references/monitoring-llm-homelab-ciro-luciotta-2026.md
+# Added 2026-08-18 (t_57a9f82f): GPU/LLM monitoring Phase 3 — Wong.
+- include_tasks: monitoring.yml
+  when: llm_monitoring_enabled | default(true)
+  tags: [always]