Create daemon to monitor android device temperatures

scripts/slave/recipe_modules/chromium_android/resources/spawn_device_temp_monitor.py

+#!/usr/bin/env python
+# Copyright 2015 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+"""Launches a daemon to monitor android device temperaures.
+
+This script will repeatedly poll the given devices for their
+temperatures every 30 seconds via adb and uploads them to monarch

[Sergey Berezin, 2015/09/09 16:47:13]

Please don't mention internal tools in external CLs. Just ts_mon or timeseries
monitoring.

[Sergey Berezin, 2015/09/09 16:47:13]

nit: our retention policy starts with once a minute, so every 30 sec will
generate (soft) errors on the backend API. Once a minute would be perfect.

[bpastene, 2015/09/09 18:32:20]

Done.

[bpastene, 2015/09/09 18:32:20]

Done.

+through infra's ts_mon.
+"""
+
+import json
+import logging
+import os
+import optparse
+import re
+import signal
+import subprocess
+import sys
+import time
+
+# Common name of sensor found in nexus devices to measure cpu (core0) temp.
+_CPU_TEMP_SENSOR = 'tsens_tz_sensor0'
+
+# TODO(bpastene): change the following once infra.git becomes a checked
+# out repo on slaves instead of a cipd managed package.

[Sergey Berezin, 2015/09/09 16:47:14]

nit: don't rely on this happening; we might do away with deployment through
checkout altogether. But at the moment, the future is uncertain.

[bpastene, 2015/09/09 18:32:20]

Done.

+
+# Location of the infra-python package's run script.
+_RUN_PY = '/opt/infra-python/run.py'
+
+
+def get_device_args(adb_path, metric_prefix, device):
+  bat_temp = None
+  cpu_temp = None
+  # Search for the file that the _CPU_TEMP_SENSOR dumps to and cat it.
+  cmd = [adb_path, '-s', device, 'shell',
+         'grep -l "%s" /sys/class/thermal/thermal_zone*/type'
+           % (_CPU_TEMP_SENSOR)]
+  try:
+    cpu_temp_files = subprocess.check_output(cmd)
+    if (len(cpu_temp_files.splitlines()) == 1):
+        cpu_temp_file = re.sub('type$', 'temp', cpu_temp_files.strip())
+        cmd = [adb_path, '-s', device, 'shell',
+               'cat %s' % (cpu_temp_file)]
+        file_contents = subprocess.check_output(cmd)
+        cpu_temp = int(file_contents)
+  except (subprocess.CalledProcessError, TypeError, ValueError):
+    cpu_temp = None
+
+  # Dump system battery info and grab the temp.
+  cmd = [adb_path, '-s', device, 'shell', 'dumpsys battery']
+  try:
+    battery_info = subprocess.check_output(cmd)
+    for line in battery_info.splitlines():
+      m = re.match('^\s*temperature: ([0-9]+)\s*$', line)
+      if m:
+        bat_temp = int(m.group(1))
+  except (subprocess.CalledProcessError, TypeError, ValueError):
+    bat_temp = None
+
+  cpu_dict = {'name': "%s/%s/cpu_temp" % (metric_prefix, device),
+              'value': cpu_temp}
+  cpu_temp_args = ['--float', json.dumps(cpu_dict)] if cpu_temp else []
+  battery_dict = {'name': '%s/%s/battery_temp' % (metric_prefix,
+                  device), 'value': bat_temp}
+  bat_temp_args = ['--float', 
+                   json.dumps(battery_dict)] if bat_temp else []
+  return cpu_temp_args + bat_temp_args
+
+
+def main(adb_path,
+         devices_json,
+         metric_prefix="android_device"):
+  """Launches the device temperature monitor.
+
+  Polls the devices for their battery and cpu temperatures
+  every 30 seconds and uploads them to monarch through infra's
+  ts_mon. Fully qualified, the metric names would be
+  /chrome/infra/${metric_prefix}/${device_serial}/(battery|cpu)_temp

[Sergey Berezin, 2015/09/09 16:47:13]

This is not the best metric name - you'll have hundreds of metrics that you
won't be able to aggregate over. Instead, use a single metric name for
temperature across all devices, e.g.

  /chrome/infra/dev/android/temperature

Or, if it makes sense, you may split the device and battery temperature into two
metrics:

  /chrome/infra/dev/battery/temperature

Individual device IDs and possibly the slave name they are attached to, should
be specified as metric fields. For a single metric, the type of temperature
(device or battery) should also be a field.

Let's talk offline about what's the best format; it depends on how you'll be
processing it on the backend.

[bpastene, 2015/09/09 18:32:20]

Done. As per our discussion offline, metrics will be both
/chrome/infra/dev/cpu/temperature and
/chrome/infra/dev/battery/temperature with metric fields to specify devices and
slaves.

+
+  Args:
+    adb_path: Path to adb binary.
+    devices_json: Json list of device serials to poll.
+    metric_prefix: Prefix of the metric name.
+  """
+
+  devices = json.loads(devices_json)
+  while True:
+    upload_cmd_args = []
+    for device in devices:
+      upload_cmd_args += get_device_args(adb_path, metric_prefix, device)
+
+    cmd = [_RUN_PY, 'infra.tools.send_ts_mon_values'] + upload_cmd_args
+    try:
+      subprocess.Popen(cmd)
+    except OSError:
+      logging.exception('Unable to call %s', _RUN_PY)
+
+    time.sleep(30)
+
+
+if __name__ == '__main__':
+  sys.exit(main(*sys.argv[1:]))