Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 1 | # Copyright 2016 The Chromium OS Authors. All rights reserved. |
| 2 | # Use of this source code is governed by a BSD-style license that can be |
| 3 | # found in the LICENSE file. |
| 4 | |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 5 | """Send system monitoring data to the timeseries monitoring API.""" |
Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 6 | |
Allen Li | 13bdf0c | 2017-03-02 15:18:16 -0800 | [diff] [blame] | 7 | from __future__ import absolute_import |
Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 8 | from __future__ import print_function |
Allen Li | 13bdf0c | 2017-03-02 15:18:16 -0800 | [diff] [blame] | 9 | from __future__ import unicode_literals |
Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 10 | |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 11 | import random |
| 12 | import time |
Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 13 | |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 14 | import psutil |
| 15 | |
| 16 | from chromite.lib import commandline |
Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 17 | from chromite.lib import cros_logging as logging |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 18 | from chromite.lib import metrics |
| 19 | from chromite.lib import ts_mon_config |
Allen Li | 325c076 | 2017-03-02 15:00:19 -0800 | [diff] [blame] | 20 | from infra_libs.ts_mon.common import interface |
| 21 | |
Allen Li | 24bf818 | 2017-03-02 16:41:20 -0800 | [diff] [blame^] | 22 | from chromite.scripts.sysmon import git_metrics |
Allen Li | baa422d | 2016-12-16 18:21:10 -0800 | [diff] [blame] | 23 | from chromite.scripts.sysmon import loop |
Allen Li | 325c076 | 2017-03-02 15:00:19 -0800 | [diff] [blame] | 24 | from chromite.scripts.sysmon import net_metrics |
Allen Li | baa422d | 2016-12-16 18:21:10 -0800 | [diff] [blame] | 25 | from chromite.scripts.sysmon import osinfo_metrics |
Allen Li | 79317bb | 2016-12-16 18:25:07 -0800 | [diff] [blame] | 26 | from chromite.scripts.sysmon import prod_metrics |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 27 | from chromite.scripts.sysmon import puppet_metrics |
| 28 | from chromite.scripts.sysmon import system_metrics |
Allen Li | ec5beb3 | 2016-09-08 15:31:41 -0700 | [diff] [blame] | 29 | |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 30 | logger = logging.getLogger(__name__) |
| 31 | |
| 32 | |
Allen Li | baa422d | 2016-12-16 18:21:10 -0800 | [diff] [blame] | 33 | class _MetricCollector(object): |
Allen Li | 38de641 | 2016-12-16 16:52:45 -0800 | [diff] [blame] | 34 | """Metric collector class.""" |
| 35 | |
Allen Li | 79317bb | 2016-12-16 18:25:07 -0800 | [diff] [blame] | 36 | def __init__(self, collect_prod_hosts=False): |
Allen Li | 45ae839 | 2017-03-02 14:19:35 -0800 | [diff] [blame] | 37 | self._collect_osinfo = _TimedCallback( |
| 38 | callback=osinfo_metrics.collect_os_info, |
Allen Li | baa422d | 2016-12-16 18:21:10 -0800 | [diff] [blame] | 39 | interval=60 * 60) |
Allen Li | 79317bb | 2016-12-16 18:25:07 -0800 | [diff] [blame] | 40 | if collect_prod_hosts: |
| 41 | logger.info('Enabling prod host metric collection.') |
Allen Li | 45ae839 | 2017-03-02 14:19:35 -0800 | [diff] [blame] | 42 | self._collect_prod_hosts = _TimedCallback( |
| 43 | callback=prod_metrics.collect_prod_hosts, |
Allen Li | 79317bb | 2016-12-16 18:25:07 -0800 | [diff] [blame] | 44 | interval=10 * 60) |
| 45 | else: |
Allen Li | 45ae839 | 2017-03-02 14:19:35 -0800 | [diff] [blame] | 46 | self._collect_prod_hosts = lambda: None |
Allen Li | 38de641 | 2016-12-16 16:52:45 -0800 | [diff] [blame] | 47 | |
| 48 | def __call__(self): |
| 49 | """Collect metrics.""" |
Allen Li | 45ae839 | 2017-03-02 14:19:35 -0800 | [diff] [blame] | 50 | system_metrics.collect_uptime() |
| 51 | system_metrics.collect_cpu_info() |
| 52 | system_metrics.collect_disk_info() |
| 53 | system_metrics.collect_mem_info() |
Allen Li | 325c076 | 2017-03-02 15:00:19 -0800 | [diff] [blame] | 54 | net_metrics.collect_net_info() |
Allen Li | 45ae839 | 2017-03-02 14:19:35 -0800 | [diff] [blame] | 55 | system_metrics.collect_proc_info() |
| 56 | system_metrics.collect_load_avg() |
| 57 | puppet_metrics.collect_puppet_summary() |
Allen Li | 24bf818 | 2017-03-02 16:41:20 -0800 | [diff] [blame^] | 58 | git_metrics.collect_git_metrics() |
Allen Li | 45ae839 | 2017-03-02 14:19:35 -0800 | [diff] [blame] | 59 | self._collect_prod_hosts() |
| 60 | self._collect_osinfo() |
| 61 | system_metrics.collect_unix_time() # must be just before flush |
Allen Li | 38de641 | 2016-12-16 16:52:45 -0800 | [diff] [blame] | 62 | metrics.Flush() |
| 63 | |
| 64 | @property |
| 65 | def _next_osinfo_collection(self): |
| 66 | return self._last_osinfo_collection + (60 * 60) |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 67 | |
| 68 | |
Allen Li | baa422d | 2016-12-16 18:21:10 -0800 | [diff] [blame] | 69 | class _TimedCallback(object): |
| 70 | """Limits callback to one call in a given interval.""" |
| 71 | |
| 72 | def __init__(self, callback, interval): |
| 73 | """Initialize instance. |
| 74 | |
| 75 | Args: |
| 76 | callback: function to call |
| 77 | interval: Number of seconds between allowed calls |
| 78 | """ |
| 79 | self._callback = callback |
| 80 | self._interval = interval |
| 81 | self._last_called = time.time() - interval |
| 82 | |
| 83 | def __call__(self): |
| 84 | if time.time() > self._next_call: |
| 85 | self._callback() |
| 86 | self._last_called = time.time() |
| 87 | |
| 88 | @property |
| 89 | def _next_call(self): |
| 90 | return self._last_called + self._interval |
| 91 | |
| 92 | |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 93 | def main(): |
| 94 | parser = commandline.ArgumentParser( |
| 95 | description=__doc__, |
Allen Li | 337729d | 2016-12-16 16:45:49 -0800 | [diff] [blame] | 96 | default_log_level='DEBUG') |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 97 | parser.add_argument( |
| 98 | '--interval', |
Allen Li | 337729d | 2016-12-16 16:45:49 -0800 | [diff] [blame] | 99 | default=60, |
| 100 | type=int, |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 101 | help='time (in seconds) between sampling system metrics') |
Allen Li | 79317bb | 2016-12-16 18:25:07 -0800 | [diff] [blame] | 102 | parser.add_argument( |
| 103 | '--collect-prod-hosts', |
| 104 | action='store_true', |
| 105 | help='Enable collection of prod host metrics, like roles') |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 106 | opts = parser.parse_args() |
| 107 | opts.Freeze() |
| 108 | |
| 109 | # This returns a 0 value the first time it's called. Call it now and |
| 110 | # discard the return value. |
| 111 | psutil.cpu_times_percent() |
| 112 | |
| 113 | # Wait a random amount of time before starting the loop in case sysmon |
| 114 | # is started at exactly the same time on all machines. |
| 115 | time.sleep(random.uniform(0, opts.interval)) |
| 116 | |
| 117 | # This call returns a context manager that doesn't do anything, so we |
| 118 | # ignore the return value. |
| 119 | ts_mon_config.SetupTsMonGlobalState('sysmon', auto_flush=False) |
| 120 | # The default prefix is '/chrome/infra/'. |
| 121 | interface.state.metric_name_prefix = (interface.state.metric_name_prefix |
| 122 | + 'chromeos/sysmon/') |
| 123 | |
Allen Li | 79317bb | 2016-12-16 18:25:07 -0800 | [diff] [blame] | 124 | collector = _MetricCollector(collect_prod_hosts=opts.collect_prod_hosts) |
| 125 | loop.SleepLoop(callback=collector, |
Allen Li | 26d1008 | 2016-12-16 16:31:02 -0800 | [diff] [blame] | 126 | interval=opts.interval).loop_forever() |
Allen Li | 788d167 | 2016-12-16 15:58:23 -0800 | [diff] [blame] | 127 | |
| 128 | |
| 129 | if __name__ == '__main__': |
| 130 | main() |