Add one-shot GPU contention audit

servers/unraid/tail2-pose-server/compose.yaml

@@ -108,6 +108,41 @@ services:
       retries: 6
       start_period: 180s
 
+  gpu-audit:
+    image: tail2-pose-server:0.1.0
+    container_name: tail2-pose-gpu-audit
+    gpus: all
+    restart: "no"
+    volumes:
+      - /:/host:ro
+      - ./validation:/validation
+    entrypoint: ["/bin/bash", "-lc"]
+    command:
+      - |
+        set -euo pipefail
+        out=/validation/gpu-audit
+        mkdir -p "$$out"
+        nvidia-smi --query-gpu=index,uuid,pstate,utilization.gpu,utilization.memory,memory.used,memory.total,power.draw,clocks.current.sm,temperature.gpu --format=csv,noheader,nounits > "$$out/snapshot.csv"
+        nvidia-smi --query-compute-apps=gpu_uuid,pid,process_name,used_gpu_memory --format=csv,noheader,nounits > "$$out/processes.csv"
+        {
+          echo 'timestamp_utc,index,uuid,utilization_gpu_pct,utilization_memory_pct,memory_used_mib,pstate,power_draw_w,clocks_sm_mhz,temperature_c'
+          for i in $$(seq 1 60); do
+            ts=$$(date -u +%FT%T.%3NZ)
+            nvidia-smi --query-gpu=index,uuid,utilization.gpu,utilization.memory,memory.used,pstate,power.draw,clocks.current.sm,temperature.gpu --format=csv,noheader,nounits | sed "s/^/$$ts,/"
+            sleep 1
+          done
+        } > "$$out/utilization-60s.csv"
+        {
+          echo 'pid,comm,cgroup'
+          tail -n +1 "$$out/processes.csv" | while IFS=, read -r uuid pid name mem; do
+            pid=$$(echo "$$pid" | tr -d ' ')
+            comm=$$(cat "/host/proc/$$pid/comm" 2>/dev/null || echo unavailable)
+            cgroup=$$(tr '\n' ';' < "/host/proc/$$pid/cgroup" 2>/dev/null || true)
+            printf '%s,%s,%s\n' "$$pid" "$$comm" "$$cgroup"
+          done
+        } > "$$out/process-map.csv"
+        chmod -R a+rX,u+w "$$out"
+
   validate:
     profiles: ["validation"]
     image: tail2-pose-server:0.1.0