From a7f3d49fea6c867c8c73e078669ed51e27181d45 Mon Sep 17 00:00:00 2001
From: Brad Stein <Brad.Stein@gmail.com>
Date: Tue, 27 Jan 2026 16:34:31 -0300
Subject: [PATCH] monitoring: read tegrastats per scrape

---
 .../jetson-tegrastats-exporter.yaml           |  2 +-
 .../scripts/jetson_tegrastats_exporter.py     | 32 ++++++++-----------
 2 files changed, 14 insertions(+), 20 deletions(-)

diff --git a/services/monitoring/jetson-tegrastats-exporter.yaml b/services/monitoring/jetson-tegrastats-exporter.yaml
index d80d83e..3679938 100644
--- a/services/monitoring/jetson-tegrastats-exporter.yaml
+++ b/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: "4"
+        monitoring.bstein.dev/restart-rev: "5"
     spec:
       serviceAccountName: default
       hostPID: true
diff --git a/services/monitoring/scripts/jetson_tegrastats_exporter.py b/services/monitoring/scripts/jetson_tegrastats_exporter.py
index 204e439..8314ad7 100644
--- a/services/monitoring/scripts/jetson_tegrastats_exporter.py
+++ b/services/monitoring/scripts/jetson_tegrastats_exporter.py
@@ -7,7 +7,6 @@ from time import time
 
 PORT = int(os.environ.get("JETSON_EXPORTER_PORT", "9100"))
 NODE_NAME = os.environ.get("NODE_NAME") or os.uname().nodename
-LOGFILE = "/tmp/tegrastats.log"
 BASE_METRICS = {
     "gr3d_freq_percent": 0.0,
     "gpu_temp_c": 0.0,
@@ -39,25 +38,21 @@ def parse_line(line: str) -> dict:
         updates["power_5v_in_mw"] = float(m.group(1))
     return updates
 
-def start_tegrastats():
-    subprocess.Popen(
-        ["/host/usr/bin/tegrastats", "--interval", "1000", "--logfile", LOGFILE],
-        stdout=subprocess.DEVNULL,
-        stderr=subprocess.STDOUT,
-        text=True,
-    )
-
-
 def read_latest_line() -> str:
-    if not os.path.exists(LOGFILE):
-        return ""
     try:
-        with open(LOGFILE, "rb") as handle:
-            handle.seek(0, os.SEEK_END)
-            size = handle.tell()
-            handle.seek(max(size - 4096, 0), os.SEEK_SET)
-            tail = handle.read().decode("utf-8", errors="ignore").splitlines()
-            return tail[-1] if tail else ""
+        proc = subprocess.Popen(
+            ["/host/usr/bin/tegrastats", "--interval", "1000"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            text=True,
+        )
+        line = proc.stdout.readline()
+        proc.terminate()
+        try:
+            proc.wait(timeout=1)
+        except subprocess.TimeoutExpired:
+            proc.kill()
+        return line
     except OSError:
         return ""
 
@@ -88,6 +83,5 @@ class Handler(http.server.BaseHTTPRequestHandler):
         return
 
 if __name__ == "__main__":
-    start_tegrastats()
     with socketserver.TCPServer(("", PORT), Handler) as httpd:
         httpd.serve_forever()