fix(node-agent): detect GPUs via docker run --gpus all ubuntu:22.04

nsenter approach failed (requires SYS_ADMIN in container).
nvidia-smi bind-mount failed (Alpine vs Ubuntu glibc incompatibility).

Working solution: spawn 'docker run --rm --gpus all ubuntu:22.04 nvidia-smi'
via the Docker socket. The NVIDIA Container Runtime injects nvidia-smi and
driver libs into any container with --gpus all, regardless of the base image.
ubuntu:22.04 is already cached on GPU nodes.

Result: GPU reported with name, memory_mb, driver_version — shows as BOUND
in the cluster UI.

docker-compose.worker.yml

@@ -41,7 +41,6 @@ services:
     build: ./services/node-agent
     restart: unless-stopped
     network_mode: host
-    pid: host    # share host PID namespace so nsenter can run host binaries
     environment:
       MAM_API_URL:   ${MAM_API_URL}
       NODE_TOKEN:    ${NODE_TOKEN:-}

services/node-agent/index.js

@@ -183,41 +183,37 @@ function sampleCpu() {
 // ── Hardware detection ────────────────────────────────────────────────────
 // GPU_COUNT / BMD_COUNT env vars override filesystem detection when /dev isn't mapped
 function detectGpusViaSmi() {
-  // Run nvidia-smi in the host's mount namespace via nsenter so it uses the
+  // Spawn a temporary container with --gpus all so the NVIDIA Container Runtime
-  // host's glibc and NVIDIA driver libs (not the container's Alpine libc).
+  // injects nvidia-smi + driver libs into ubuntu:22.04 (already cached on GPU hosts).
-  // Requires pid: host in docker-compose so we can access /proc/1/ns/mnt.
+  // This sidesteps the Alpine/glibc incompatibility in the node-agent image itself.
-  // Falls back to direct nvidia-smi call (works if container has glibc).
+  // Returns null if nvidia runtime isn't available — caller falls back to /dev scan.
-  // Returns null if nvidia-smi isn't available — caller falls back to /dev scan.
   const { execFileSync } = require('child_process');
   const QUERY = '--query-gpu=index,name,memory.total,driver_version';
   const FMT   = '--format=csv,noheader,nounits';
 
-  const attempts = [
+  try {
-    // nsenter into host mount ns (preferred — uses host glibc)
+    const out = execFileSync('docker', [
-    () => execFileSync('nsenter', ['--mount=/proc/1/ns/mnt', '--', 'nvidia-smi', QUERY, FMT],
+      'run', '--rm', '--gpus', 'all',
-                       { timeout: 8000, encoding: 'utf8' }),
+      'ubuntu:22.04',
-    // Direct call (works if container has compatible glibc)
+      'nvidia-smi', QUERY, FMT,
-    () => execFileSync('nvidia-smi', [QUERY, FMT],
+    ], { timeout: 15000, encoding: 'utf8' });
-                       { timeout: 5000, encoding: 'utf8' }),
-  ];
 
-  for (const attempt of attempts) {
+    const lines = out.trim().split('\n').filter(l => /^\d+,/.test(l.trim()));
-    try {
+    if (!lines.length) return null;
-      const out = attempt();
+    return lines.map(line => {
-      return out.trim().split('\n').filter(Boolean).map(line => {
+      const [idx, name, memMiB, driver] = line.split(',').map(s => s.trim());
-        const [idx, name, memMiB, driver] = line.split(',').map(s => s.trim());
+      return {
-        return {
+        index:     parseInt(idx, 10),
-          index:     parseInt(idx, 10),
+        name:      name,
-          name:      name,
+        memory_mb: parseInt(memMiB, 10),
-          memory_mb: parseInt(memMiB, 10),
+        driver:    driver,
-          driver:    driver,
+        device:    `/dev/nvidia${idx}`,
-          device:    `/dev/nvidia${idx}`,
+        type:      'nvidia',
-          type:      'nvidia',
+      };
-        };
+    });
-      });
+  } catch (_) {
-    } catch (_) {}
+    return null; // nvidia runtime not available — caller uses /dev scan
   }
-  return null; // all attempts failed — caller falls back to /dev scan
 }
 
 function detectHardware() {