monitoring: fix jetson gpu metrics

scripts/dashboards_render_atlas.py

@@ -221,6 +221,13 @@ def jetson_gpu_util_by_node():
     return 'max by (node) (jetson_gr3d_freq_percent{node!=""})'
 
 
+def jetson_gpu_util_by_hostname():
+    return (
+        'label_replace(max by (node) (jetson_gr3d_freq_percent{node!=""}), '
+        '"Hostname", "$1", "node", "(.*)")'
+    )
+
+
 def jetson_gpu_requests(scope_var):
     return (
         "sum by (namespace,node) ("
@@ -2688,7 +2695,7 @@ def build_gpu_dashboard():
         timeseries_panel(
             3,
             "GPU Util by Node",
-            'sum by (Hostname) (DCGM_FI_DEV_GPU_UTIL{pod!=""})',
+            f'(sum by (Hostname) (DCGM_FI_DEV_GPU_UTIL{{pod!=""}})) or ({jetson_gpu_util_by_hostname()})',
             {"h": 8, "w": 12, "x": 0, "y": 8},
             unit="percent",
             legend="{{Hostname}}",

services/monitoring/dashboards/atlas-gpu.json

@@ -126,7 +126,7 @@
       },
       "targets": [
         {
-          "expr": "sum by (Hostname) (DCGM_FI_DEV_GPU_UTIL{pod!=\"\"})",
+          "expr": "(sum by (Hostname) (DCGM_FI_DEV_GPU_UTIL{pod!=\"\"})) or (label_replace(max by (node) (jetson_gr3d_freq_percent{node!=\"\"}), \"Hostname\", \"$1\", \"node\", \"(.*)\"))",
           "refId": "A",
           "legendFormat": "{{Hostname}}"
         }

services/monitoring/grafana-dashboard-gpu.yaml

@@ -135,7 +135,7 @@ data:
           },
           "targets": [
             {
-              "expr": "sum by (Hostname) (DCGM_FI_DEV_GPU_UTIL{pod!=\"\"})",
+              "expr": "(sum by (Hostname) (DCGM_FI_DEV_GPU_UTIL{pod!=\"\"})) or (label_replace(max by (node) (jetson_gr3d_freq_percent{node!=\"\"}), \"Hostname\", \"$1\", \"node\", \"(.*)\"))",
               "refId": "A",
               "legendFormat": "{{Hostname}}"
             }

services/monitoring/jetson-tegrastats-exporter.yaml

@@ -17,7 +17,7 @@ spec:
       annotations:
         prometheus.io/scrape: "true"
         prometheus.io/port: "9100"
-        monitoring.bstein.dev/restart-rev: "1"
+        monitoring.bstein.dev/restart-rev: "2"
     spec:
       serviceAccountName: default
       hostPID: true

services/monitoring/scripts/jetson_tegrastats_exporter.py

@@ -4,7 +4,7 @@ import re
 import socketserver
 import subprocess
 import threading
-from time import time
+from time import sleep, time
 
 PORT = int(os.environ.get("JETSON_EXPORTER_PORT", "9100"))
 NODE_NAME = os.environ.get("NODE_NAME") or os.uname().nodename
@@ -20,6 +20,7 @@ METRICS = {
 LOCK = threading.Lock()
 
 def parse_line(line: str):
+    line = line.strip()
     updates = {}
     m = re.search(r"GR3D_FREQ\\s+(\\d+)%", line)
     if m:
@@ -34,7 +35,7 @@ def parse_line(line: str):
     if m:
         updates["ram_used_mb"] = float(m.group(1))
         updates["ram_total_mb"] = float(m.group(2))
-    m = re.search(r"POM_5V_IN\\s+(\\d+)/(\\d+)", line)
+    m = re.search(r"(?:POM_5V_IN|VDD_IN)\\s+(\\d+)/(\\d+)", line)
     if m:
         updates["power_5v_in_mw"] = float(m.group(1))
     with LOCK:
@@ -42,15 +43,23 @@ def parse_line(line: str):
         METRICS["last_scrape_ts"] = time()
 
 def run_tegrastats():
-    proc = subprocess.Popen(
-        ["/host/usr/bin/tegrastats", "--interval", "1000"],
-        stdout=subprocess.PIPE,
+    logfile = "/tmp/tegrastats.log"
+    subprocess.Popen(
+        ["/host/usr/bin/tegrastats", "--interval", "1000", "--logfile", logfile],
+        stdout=subprocess.DEVNULL,
         stderr=subprocess.STDOUT,
         text=True,
-        bufsize=1,
     )
-    for line in proc.stdout:
-        parse_line(line)
+    while not os.path.exists(logfile):
+        sleep(0.1)
+    with open(logfile, "r", encoding="utf-8", errors="ignore") as handle:
+        handle.seek(0, os.SEEK_END)
+        while True:
+            line = handle.readline()
+            if not line:
+                sleep(0.2)
+                continue
+            parse_line(line)
 
 class Handler(http.server.BaseHTTPRequestHandler):
     def do_GET(self):