feat: add nvidia-smi pmon integration to detect NVENC/NVDEC processes and display GPU process activity with auto-refresh status

Add graphics_process_log_state global to track GPU process changes. Query nvidia-smi pmon with 1-count sample to detect video encode/decode processes missed by compute-apps query. Parse pmon output for SM/ENC/DEC utilization percentages and process types. Merge pmon results with compute-apps by PID, preferring pmon data when available. Log GPU process detection with PID/name

app.py

@@ -88,6 +88,7 @@ raid_refresh_lock = threading.Lock()
 datastore_cache = {"stored_at": 0.0, "snapshot": None, "raid": None, "raid_at": 0.0}
 raid_log_state = None
 graphics_log_state = None
+graphics_process_log_state = None
 hardware_log_once = set()
 
 LANG_NAMES = {"de":"Deutsch", "deu":"Deutsch", "ger":"Deutsch", "en":"Englisch", "eng":"Englisch", "ja":"Japanisch", "jpn":"Japanisch", "ko":"Koreanisch", "kor":"Koreanisch", "fr":"Französisch", "fra":"Französisch", "fre":"Französisch", "es":"Spanisch", "spa":"Spanisch", "it":"Italienisch", "ita":"Italienisch", "ru":"Russisch", "rus":"Russisch", "zh":"Chinesisch", "zho":"Chinesisch", "chi":"Chinesisch", "und":"Unbekannt", "":"Unbekannt"}
@@ -357,7 +358,7 @@ def io_pressure():
 
 def graphics_snapshot():
     """Read NVIDIA GPU details with short, non-blocking subprocess timeouts."""
-    global graphics_log_state
+    global graphics_log_state, graphics_process_log_state
     query = "index,name,uuid,serial,pci.bus_id,driver_version,pstate,temperature.gpu,utilization.gpu,utilization.memory,memory.total,memory.used,memory.free,power.draw,power.limit,fan.speed,clocks.current.graphics,clocks.current.memory,clocks.max.graphics,clocks.max.memory,compute_mode"
     try:
         result = subprocess.run([NVIDIA_SMI_BIN, f"--query-gpu={query}", "--format=csv,noheader,nounits"], capture_output=True, text=True, timeout=5, check=False)
@@ -388,15 +389,49 @@ def graphics_snapshot():
             try: gpu[key] = round(float(gpu[key]), 1)
             except (TypeError, ValueError): gpu[key] = None
         gpus.append(gpu)
-    processes = []
+    processes_by_pid = {}
     try:
         proc = subprocess.run([NVIDIA_SMI_BIN, "--query-compute-apps=pid,process_name,used_gpu_memory", "--format=csv,noheader,nounits"], capture_output=True, text=True, timeout=5, check=False)
-        for line in proc.stdout.splitlines():
-            values = [value.strip() for value in line.split(",")]
-            if len(values) >= 3:
-                processes.append({"pid": values[0], "name": values[1], "memory": values[2]})
+        if proc.returncode == 0:
+            for line in proc.stdout.splitlines():
+                values = [value.strip() for value in line.split(",")]
+                if len(values) >= 3 and values[0].isdigit():
+                    processes_by_pid[values[0]] = {"pid": values[0], "name": values[1], "memory": values[2], "type": "compute"}
     except (OSError, subprocess.SubprocessError):
         pass
+    pmon_error = ""
+    try:
+        # Compute-apps misses NVENC/NVDEC. pmon also reports video/graphics
+        # processes such as Jellyfin while a hardware transcode is running.
+        pmon = subprocess.run([NVIDIA_SMI_BIN, "pmon", "-c", "1", "-s", "um"], capture_output=True, text=True, timeout=5, check=False)
+        if pmon.returncode == 0:
+            for line in pmon.stdout.splitlines():
+                values = line.split()
+                if not values or values[0].startswith("#") or len(values) < 8 or not values[1].isdigit():
+                    continue
+                pid, process_type = values[1], values[2]
+                command = " ".join(values[7:]) or "GPU-Prozess"
+                processes_by_pid[pid] = {
+                    "pid": pid,
+                    "name": command,
+                    "memory": f"SM {values[3]}% · ENC {values[5]}% · DEC {values[6]}%",
+                    "type": process_type,
+                    "sm": values[3], "enc": values[5], "dec": values[6],
+                }
+        else:
+            pmon_error = " ".join((pmon.stderr or "pmon ohne Ausgabe").split())[:180]
+    except (OSError, subprocess.SubprocessError) as exc:
+        pmon_error = error_summary(exc)
+    processes = list(processes_by_pid.values())
+    process_signature = tuple((item["pid"], item["name"], item.get("type")) for item in processes)
+    if process_signature != graphics_process_log_state:
+        if processes:
+            log.info("GPU-Prozesse erkannt: %s", ", ".join(f"{item['name']} (PID {item['pid']})" for item in processes))
+        elif pmon_error:
+            log.warning("GPU-Prozessabfrage fehlgeschlagen: %s", pmon_error)
+        else:
+            log.info("Keine GPU-Prozesse erkannt")
+        graphics_process_log_state = process_signature
     signature = ("ok", tuple((gpu.get("index"), gpu.get("name"), gpu.get("temperature.gpu"), gpu.get("utilization.gpu")) for gpu in gpus))
     if signature != graphics_log_state:
         log.info("GPU-Status verfügbar: %s NVIDIA-GPU(s) via %s", len(gpus), NVIDIA_SMI_BIN)