Commit 2da17657 authored by Lisa Casino's avatar Lisa Casino

slapos/collect: Enable disk usage by default

parent 83407df8
...@@ -66,6 +66,12 @@ download-binary-dir-url = http://shadir.nxdcdn.com ...@@ -66,6 +66,12 @@ download-binary-dir-url = http://shadir.nxdcdn.com
#shadir-cert-file = /etc/opt/slapos/shacache/shacache.cert #shadir-cert-file = /etc/opt/slapos/shacache/shacache.cert
#shadir-key-file = /etc/opt/slapos/shacache/shacache.key #shadir-key-file = /etc/opt/slapos/shacache/shacache.key
[collect]
enable = True
time_cycle = 86400
pid_folder = /srv/slapgrid/var/run
use_quota = False
# List of signatures of uploaders we trust: # List of signatures of uploaders we trust:
# Sebastien Robin # Sebastien Robin
# Kazuhiko Shiozaki # Kazuhiko Shiozaki
......
...@@ -31,6 +31,8 @@ from slapos.collect import do_collect ...@@ -31,6 +31,8 @@ from slapos.collect import do_collect
from slapos.cli.command import must_be_root from slapos.cli.command import must_be_root
from slapos.cli.config import ConfigCommand from slapos.cli.config import ConfigCommand
import logging
class CollectCommand(ConfigCommand): class CollectCommand(ConfigCommand):
""" """
Collect system consumption and data and store. Collect system consumption and data and store.
...@@ -44,4 +46,5 @@ class CollectCommand(ConfigCommand): ...@@ -44,4 +46,5 @@ class CollectCommand(ConfigCommand):
@must_be_root @must_be_root
def take_action(self, args): def take_action(self, args):
configp = self.fetch_config(args) configp = self.fetch_config(args)
do_collect(configp) logger = self.app.log
do_collect(logger, configp)
...@@ -71,7 +71,7 @@ def current_state(user_dict): ...@@ -71,7 +71,7 @@ def current_state(user_dict):
for i, process in enumerate(process_list): for i, process in enumerate(process_list):
yield build_snapshot(process) yield build_snapshot(process)
def do_collect(conf): def do_collect(logger, conf):
""" """
Main function Main function
The idea here is to poll system every so many seconds The idea here is to poll system every so many seconds
...@@ -80,6 +80,7 @@ def do_collect(conf): ...@@ -80,6 +80,7 @@ def do_collect(conf):
Each user object is a dict, indexed on timestamp. We add every snapshot Each user object is a dict, indexed on timestamp. We add every snapshot
matching the user so that we get informations for each users matching the user so that we get informations for each users
""" """
logger.info('Collecting data...')
try: try:
collected_date, collected_time = _get_time() collected_date, collected_time = _get_time()
user_dict = get_user_list(conf) user_dict = get_user_list(conf)
...@@ -94,16 +95,20 @@ def do_collect(conf): ...@@ -94,16 +95,20 @@ def do_collect(conf):
if conf.has_option("slapos", "collect_cache"): if conf.has_option("slapos", "collect_cache"):
days_to_preserve = conf.getint("slapos", "collect_cache") days_to_preserve = conf.getint("slapos", "collect_cache")
log_directory = "%s/var/data-log" % conf.get("slapos", "instance_root") log_directory = "%s/var/data-log" % conf.get("slapos", "instance_root")
logger.debug("Log directory: %s", log_directory)
mkdir_p(log_directory, 0o755) mkdir_p(log_directory, 0o755)
consumption_report_directory = "%s/var/consumption-report" % \ consumption_report_directory = "%s/var/consumption-report" % \
conf.get("slapos", "instance_root") conf.get("slapos", "instance_root")
mkdir_p(consumption_report_directory, 0o755) mkdir_p(consumption_report_directory, 0o755)
logger.debug("Consumption report directory: %s", consumption_report_directory)
xml_report_directory = "%s/var/xml_report/%s" % \ xml_report_directory = "%s/var/xml_report/%s" % \
(conf.get("slapos", "instance_root"), (conf.get("slapos", "instance_root"),
conf.get("slapos", "computer_id")) conf.get("slapos", "computer_id"))
mkdir_p(xml_report_directory, 0o755) mkdir_p(xml_report_directory, 0o755)
logger.debug("XML report directory: %s", xml_report_directory)
if stat.S_IMODE(os.stat(log_directory).st_mode) != 0o755: if stat.S_IMODE(os.stat(log_directory).st_mode) != 0o755:
os.chmod(log_directory, 0o755) os.chmod(log_directory, 0o755)
...@@ -113,11 +118,12 @@ def do_collect(conf): ...@@ -113,11 +118,12 @@ def do_collect(conf):
if conf.has_option("slapformat", "computer_model_id"): if conf.has_option("slapformat", "computer_model_id"):
computer_model_id = conf.get("slapformat", computer_model_id = conf.get("slapformat",
"computer_model_id") "computer_model_id")
else: else:
computer_model_id = "no_model" computer_model_id = "no_model"
logger.debug("Computer model id: %s", computer_model_id)
uptime = _get_uptime() uptime = _get_uptime()
if conf.has_option("slapformat", "heating_sensor_id"): if conf.has_option("slapformat", "heating_sensor_id"):
heating_sensor_id = conf.get("slapformat", heating_sensor_id = conf.get("slapformat",
"heating_sensor_id") "heating_sensor_id")
...@@ -130,18 +136,30 @@ def do_collect(conf): ...@@ -130,18 +136,30 @@ def do_collect(conf):
else: else:
heating_sensor_id = "no_sensor" heating_sensor_id = "no_sensor"
test_heating = False test_heating = False
logger.debug("Heating sensor id: %s", heating_sensor_id)
logger.info("Inserting computer information into database...")
computer = Computer(ComputerSnapshot(model_id=computer_model_id, computer = Computer(ComputerSnapshot(model_id=computer_model_id,
sensor_id = heating_sensor_id, sensor_id = heating_sensor_id,
test_heating=test_heating)) test_heating=test_heating))
# Insert computer's data
computer.save(database, collected_date, collected_time) computer.save(database, collected_date, collected_time)
logger.info("Done.")
logger.info("Inserting user information into database...")
# Insert TABLE user + TABLE folder
for user in user_dict.values(): for user in user_dict.values():
user.save(database, collected_date, collected_time) user.save(database, collected_date, collected_time)
logger.info("Done.")
logger.info("Writing csv, XML and JSON files...")
# Write a csv with dumped data in the log_directory
SystemCSVReporterDumper(database).dump(log_directory) SystemCSVReporterDumper(database).dump(log_directory)
RawCSVDumper(database).dump(log_directory) RawCSVDumper(database).dump(log_directory)
# Write xml files
consumption_report = ConsumptionReport( consumption_report = ConsumptionReport(
computer_id=conf.get("slapos", "computer_id"), computer_id=conf.get("slapos", "computer_id"),
user_list=user_dict, user_list=user_dict,
...@@ -156,16 +174,23 @@ def do_collect(conf): ...@@ -156,16 +174,23 @@ def do_collect(conf):
if report_file is not None: if report_file is not None:
shutil.copy(report_file, xml_report_directory) shutil.copy(report_file, xml_report_directory)
# write json
partition_report = PartitionReport( partition_report = PartitionReport(
database=database, database=database,
user_list=user_dict) user_list=user_dict)
partition_report.buildJSONMonitorReport() partition_report.buildJSONMonitorReport()
# Put dumped csv in a current_date.tar.gz
compressLogFolder(log_directory) compressLogFolder(log_directory)
logger.info("Done.")
# Drop older entries already reported # Drop older entries already reported
database.garbageCollect(days_to_preserve) database.garbageCollect(days_to_preserve)
logger.info("Finished collecting.")
logger.info('=' * 80)
except AccessDenied: except AccessDenied:
print("You HAVE TO execute this script with root permission.") logger.error("You HAVE TO execute this script with root permission.")
...@@ -28,9 +28,12 @@ ...@@ -28,9 +28,12 @@
############################################################################## ##############################################################################
import os import os
import logging
from datetime import datetime, timedelta from datetime import datetime, timedelta
from slapos.collect.snapshot import FolderSizeSnapshot from slapos.collect.snapshot import FolderSizeSnapshot
logger = logging.getLogger(__name__)
def get_user_list(config): def get_user_list(config):
nb_user = int(config.get("slapformat", "partition_amount")) nb_user = int(config.get("slapformat", "partition_amount"))
name_prefix = config.get("slapformat", "user_base_name") name_prefix = config.get("slapformat", "user_base_name")
...@@ -38,14 +41,19 @@ def get_user_list(config): ...@@ -38,14 +41,19 @@ def get_user_list(config):
instance_root = config.get("slapos", "instance_root") instance_root = config.get("slapos", "instance_root")
# By default, enable disk snapshot, # By default, enable disk snapshot,
# and set time_cycle to 24hours after the first disk snapshot run # and set time_cycle to 24hours after the first disk snapshot run
disk_snapshot_params = {'enable': False, 'time_cycle': 86400} pid_folder_tmp = instance_root + "/var/run"
disk_snapshot_params = {'enable': True,
'time_cycle': 86400,
'pid_folder': pid_folder_tmp,
'use_quota': False}
if config.has_section('collect'): if config.has_section('collect'):
collect_section = dict(config.items("collect")) collect_section = dict(config.items("collect"))
disk_snapshot_params = dict( disk_snapshot_params = dict(
enable=collect_section.get("report_disk_usage", "False").lower() in ('true', 'on', '1'), enable = collect_section.get("report_disk_usage", "True").lower() in ('true', 'on', '1'),
pid_folder=collect_section.get("disk_snapshot_process_pid_foder", None), pid_folder = collect_section.get("disk_snapshot_process_pid_foder", pid_folder_tmp),
time_cycle=int(collect_section.get("disk_snapshot_time_cycle", 86400)), time_cycle = int(collect_section.get("disk_snapshot_time_cycle", 86400)),
use_quota=collect_section.get("disk_snapshot_use_quota", "True").lower() in ('true', 'on', '1'), use_quota = collect_section.get("disk_snapshot_use_quota", "False").lower() in ('true', 'on', '1'),
) )
user_dict = {name: User(name, path, disk_snapshot_params) user_dict = {name: User(name, path, disk_snapshot_params)
for name, path in [ for name, path in [
...@@ -72,6 +80,7 @@ class User(object): ...@@ -72,6 +80,7 @@ class User(object):
def _insertDiskSnapShot(self, database, collected_date, collected_time): def _insertDiskSnapShot(self, database, collected_date, collected_time):
if self.disk_snapshot_params['enable']: if self.disk_snapshot_params['enable']:
time_cycle = self.disk_snapshot_params.get('time_cycle', 0) time_cycle = self.disk_snapshot_params.get('time_cycle', 0)
database.connect() database.connect()
if time_cycle: if time_cycle:
for date_time in database.select(table="folder", columns="date, time", for date_time in database.select(table="folder", columns="date, time",
...@@ -79,18 +88,24 @@ class User(object): ...@@ -79,18 +88,24 @@ class User(object):
where="partition='%s'" % self.name): where="partition='%s'" % self.name):
latest_date = datetime.strptime('%s %s' % date_time, latest_date = datetime.strptime('%s %s' % date_time,
"%Y-%m-%d %H:%M:%S") "%Y-%m-%d %H:%M:%S")
if (datetime.utcnow() - latest_date).seconds < time_cycle: time_spent = (datetime.utcnow() - latest_date).total_seconds()
if time_spent < time_cycle:
# wait the time cycle # wait the time cycle
logger.info("Time cycle is not over (%s seconds remaining). No computation of "
"disk usage on the partition %s.", time_cycle - time_spent, self.name)
return return
break break
pid_file = self.disk_snapshot_params.get('pid_folder', None) pid_file = self.disk_snapshot_params.get('pid_folder', None)
if pid_file is not None: if pid_file is not None:
pid_file = os.path.join(pid_file, '%s_disk_size.pid' % self.name) pid_file = os.path.join(pid_file, '%s_disk_size.pid' % self.name)
disk_snapshot = FolderSizeSnapshot(self.path, pid_file) disk_snapshot = FolderSizeSnapshot(self.path, pid_file)
disk_snapshot.update_folder_size() disk_snapshot.update_folder_size()
# Skeep insert empty partition: size <= 1Mb # Skeep insert empty partition: size <= 1Mb
if disk_snapshot.disk_usage <= 1024.0 and \ if disk_snapshot.disk_usage <= 1024.0 and \
not self.disk_snapshot_params.get('testing', False): not self.disk_snapshot_params.get('testing', False):
logger.debug("Disk usage of the partition %s: %s. "
"Ignoring insertion in the dataset.", self.name, disk_snapshot.disk_usage)
return return
database.inserFolderSnapshot(self.name, database.inserFolderSnapshot(self.name,
disk_usage=disk_snapshot.get("disk_usage"), disk_usage=disk_snapshot.get("disk_usage"),
...@@ -102,7 +117,6 @@ class User(object): ...@@ -102,7 +117,6 @@ class User(object):
def save(self, database, collected_date, collected_time): def save(self, database, collected_date, collected_time):
""" Insert collected data on user collector """ """ Insert collected data on user collector """
database.connect() database.connect()
snapshot_counter = len(self.snapshot_list)
for snapshot_item in self.snapshot_list: for snapshot_item in self.snapshot_list:
snapshot_item.update_cpu_percent() snapshot_item.update_cpu_percent()
database.insertUserSnapshot(self.name, database.insertUserSnapshot(self.name,
......
...@@ -31,6 +31,7 @@ from __future__ import print_function ...@@ -31,6 +31,7 @@ from __future__ import print_function
import psutil import psutil
import os import os
import subprocess import subprocess
import logging
from .temperature import collectComputerTemperature, launchTemperatureTest from .temperature import collectComputerTemperature, launchTemperatureTest
from .temperature.heating import get_contribution_ratio from .temperature.heating import get_contribution_ratio
...@@ -39,6 +40,8 @@ import six ...@@ -39,6 +40,8 @@ import six
MEASURE_INTERVAL = 5 MEASURE_INTERVAL = 5
logger = logging.getLogger(__name__)
class _Snapshot(object): class _Snapshot(object):
def get(self, property, default=None): def get(self, property, default=None):
return getattr(self, property, default) return getattr(self, property, default)
...@@ -97,18 +100,21 @@ class FolderSizeSnapshot(_Snapshot): ...@@ -97,18 +100,21 @@ class FolderSizeSnapshot(_Snapshot):
except OSError: except OSError:
pass pass
else: else:
logger.warning("Process %s still in progress. Try later.", pid)
return return
self.disk_usage = self._getSize(self.folder_path) self.disk_usage = self._getSize(self.folder_path)
# If extra disk added to partition # If extra disk added to partition
data_dir = os.path.join(self.folder_path, 'DATA') data_dir = os.path.join(self.folder_path, 'DATA')
if os.path.exists(data_dir): if os.path.exists(data_dir):
logger.debug("Extra disk added to the partition")
for filename in os.listdir(data_dir): for filename in os.listdir(data_dir):
extra_path = os.path.join(data_dir, filename) extra_path = os.path.join(data_dir, filename)
if os.path.islink(extra_path) and os.path.isdir('%s/' % extra_path): if os.path.islink(extra_path) and os.path.isdir('%s/' % extra_path):
self.disk_usage += self._getSize('%s/' % extra_path) self.disk_usage += self._getSize('%s/' % extra_path)
def _getSize(self, file_path): def _getSize(self, file_path):
size = 0 size = 0
command = 'du -s %s' % file_path command = 'du -s %s' % file_path
process = subprocess.Popen(command, stdout=subprocess.PIPE, process = subprocess.Popen(command, stdout=subprocess.PIPE,
...@@ -119,6 +125,7 @@ class FolderSizeSnapshot(_Snapshot): ...@@ -119,6 +125,7 @@ class FolderSizeSnapshot(_Snapshot):
result = process.communicate()[0] result = process.communicate()[0]
if process.returncode == 0: if process.returncode == 0:
size, _ = result.strip().split() size, _ = result.strip().split()
return float(size) return float(size)
class SystemSnapshot(_Snapshot): class SystemSnapshot(_Snapshot):
...@@ -161,8 +168,7 @@ class HeatingContributionSnapshot(_Snapshot): ...@@ -161,8 +168,7 @@ class HeatingContributionSnapshot(_Snapshot):
result = launchTemperatureTest(sensor_id) result = launchTemperatureTest(sensor_id)
if result is None: if result is None:
print("Impossible to test sensor: %s " % sensor_id) logger.warning("Impossible to test sensor: %s", sensor_id)
initial_temperature, final_temperature, duration = result initial_temperature, final_temperature, duration = result
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment