def exclude_disk(self, part):
# skip cd-rom drives with no disk in it; they may raise
# ENOENT, pop-up a Windows GUI error for a non-ready
# partition or just hang;
# and all the other excluded disks
skip_win = Platform.is_win32() and ('cdrom' in part.opts or part.fstype == '')
return skip_win or self._exclude_disk(part.device, part.fstype, part.mountpoint)
def collect_metrics_psutil(self):
self._valid_disks = {}
for part in psutil.disk_partitions(all=True):
# we check all exclude conditions
if self.exclude_disk(part):
continue
# Get disk metrics here to be able to exclude on total usage
try:
disk_usage = timeout(5)(psutil.disk_usage)(part.mountpoint)
except TimeoutException:
self.log.warning(
u'Timeout while retrieving the disk usage of `%s` mountpoint. Skipping...', part.mountpoint
)
continue
except Exception as e:
self.log.warning('Unable to get disk metrics for %s: %s', part.mountpoint, e)
continue
# Exclude disks with total disk size 0
if disk_usage.total == 0:
continue
# For later, latency metrics
self._valid_disks[part.device] = (part.fstype, part.mountpoint)
self.log.debug('Passed: {}'.format(part.device))
device_name = part.mountpoint if self._use_mount else part.device
tags = [part.fstype, 'filesystem:{}'.format(part.fstype)] if self._tag_by_filesystem else []
tags.extend(self._custom_tags)
# apply device/mountpoint specific tags
for regex, device_tags in self._device_tag_re:
if regex.match(device_name):
tags.extend(device_tags)
if self.devices_label.get(device_name):
tags.append(self.devices_label.get(device_name))
# legacy check names c: vs psutil name C:\\
if Platform.is_win32():
device_name = device_name.strip('\\').lower()
tags.append('device:{}'.format(device_name))
for metric_name, metric_value in iteritems(self._collect_part_metrics(part, disk_usage)):
self.gauge(metric_name, metric_value, tags=tags)
# Add in a disk read write or read only check
if self._service_check_rw:
rwro = {'rw', 'ro'} & set(part.opts.split(','))
if len(rwro) == 1:
self.service_check(
'disk.read_write', AgentCheck.OK if rwro.pop() == 'rw' else AgentCheck.CRITICAL, tags=tags
)
else:
self.service_check('disk.read_write', AgentCheck.UNKNOWN, tags=tags)
self.collect_latency_metrics()
def psutil_wrapper(self, process, method, accessors, try_sudo, *args,
**kwargs):
"""
A psutil wrapper that is calling
* psutil.method(*args, **kwargs) and returns the result
OR
* psutil.method(*args, **kwargs).accessor[i] for each accessors
given in a list, the result being indexed in a dictionary
by the accessor name
"""
if accessors is None:
result = None
else:
result = {}
# Ban certain method that we know fail
if method == 'num_fds' and not Platform.is_unix():
return result
elif method == 'num_handles' and not Platform.is_win32():
return result
try:
res = getattr(process, method)(*args, **kwargs)
if accessors is None:
result = res
else:
for acc in accessors:
try:
result[acc] = getattr(res, acc)
except AttributeError:
self.log.debug(
"psutil.%s().%s attribute does not exist", method,
acc)
except (NotImplementedError, AttributeError):
self.log.debug("psutil method %s not implemented", method)
except psutil.AccessDenied:
self.log.debug("psutil was denied access for method %s", method)
if method == 'num_fds' and Platform.is_unix() and try_sudo:
try:
# It is up the agent's packager to grant
# corresponding sudo policy on unix platforms
ls_args = [
'sudo', 'ls', '/proc/{}/fd/'.format(process.pid)
]
process_ls = subprocess.check_output(ls_args)
result = len(process_ls.splitlines())
except subprocess.CalledProcessError as e:
self.log.exception(
"trying to retrieve %s with sudo failed with return code %s",
method, e.returncode)
except Exception:
self.log.exception(
"trying to retrieve %s with sudo also failed", method)
except psutil.NoSuchProcess:
self.warning("Process %s disappeared while scanning", process.pid)
return result
def psutil_wrapper(self, process, method, accessors=None, *args, **kwargs):
"""
A psutil wrapper that is calling
* psutil.method(*args, **kwargs) and returns the result
OR
* psutil.method(*args, **kwargs).accessor[i] for each accessors
given in a list, the result being indexed in a dictionary
by the accessor name
"""
if accessors is None:
result = None
else:
result = {}
# Ban certain method that we know fail
if method == 'num_fds' and not Platform.is_unix():
return result
elif method == 'num_handles' and not Platform.is_win32():
return result
# Try running `num_fds` with sudo if possible
if method == 'num_fds' and self.try_sudo:
self.log.debug("Running num_fds using sudo")
try:
ls_args = ['sudo', 'ls', '/proc/{}/fd/'.format(process.pid)]
process_ls = subprocess.check_output(ls_args)
result = len(process_ls.splitlines())
except Exception as e:
self.log.exception(
"Trying to retrieve %s with sudo failed with error: %s",
method, e)
else:
try:
res = getattr(process, method)(*args, **kwargs)
if accessors is None:
result = res
else:
for acc in accessors:
try:
result[acc] = getattr(res, acc)
except AttributeError:
self.log.debug(
"psutil.%s().%s attribute does not exist",
method, acc)
except (NotImplementedError, AttributeError):
self.log.debug("psutil method %s not implemented", method)
except psutil.AccessDenied:
self.log.debug("psutil was denied access for method %s",
method)
except psutil.NoSuchProcess:
self.log.debug("Process %s disappeared while scanning",
process.pid)
return result
from datadog_checks.base import AgentCheck
from datadog_checks.base.utils.platform import Platform
PY3 = sys.version_info[0] == 3
if PY3:
# use higher precision clock available in Python3
time_func = time.perf_counter
else:
time_func = time.time
# These imports are necessary because otherwise dynamic type
# resolution will fail on windows without it.
# See more here: https://github.com/rthalley/dnspython/issues/39.
if Platform.is_win32():
from dns.rdtypes.ANY import * # noqa
from dns.rdtypes.IN import * # noqa
# for tiny time deltas, time.time on Windows reports the same value
# of the clock more than once, causing the computation of response_time
# to be often 0; let's use time.clock that is more precise.
if not PY3:
time_func = time.clock
class BadConfException(Exception):
pass
class DNSCheck(AgentCheck):
def check(self, instance):
"""Get disk space/inode stats"""
if self._tag_by_label and Platform.is_linux():
self.devices_label = self._get_devices_label()
self._valid_disks = {}
for part in psutil.disk_partitions(all=self._include_all_devices):
# we check all exclude conditions
if self.exclude_disk(part):
continue
# Get disk metrics here to be able to exclude on total usage
try:
disk_usage = timeout(self._timeout)(psutil.disk_usage)(
part.mountpoint)
except TimeoutException:
self.log.warning(
u'Timeout after %d seconds while retrieving the disk usage of `%s` mountpoint. '
u'You might want to change the timeout length in the settings.',
self._timeout,
part.mountpoint,
)
continue
except Exception as e:
self.log.warning(
u'Unable to get disk metrics for %s: %s. '
u'You can exclude this mountpoint in the settings if it is invalid.',
part.mountpoint,
e,
)
continue
# Exclude disks with size less than min_disk_size
if disk_usage.total <= self._min_disk_size:
if disk_usage.total > 0:
self.log.info(
'Excluding device %s with total disk size %s',
part.device, disk_usage.total)
continue
# For later, latency metrics
self._valid_disks[part.device] = (part.fstype, part.mountpoint)
self.log.debug('Passed: %s', part.device)
device_name = part.mountpoint if self._use_mount else part.device
tags = [part.fstype, 'filesystem:{}'.format(part.fstype)
] if self._tag_by_filesystem else []
tags.extend(self._custom_tags)
# apply device/mountpoint specific tags
for regex, device_tags in self._device_tag_re:
if regex.match(device_name):
tags.extend(device_tags)
if self.devices_label.get(device_name):
tags.extend(self.devices_label.get(device_name))
# legacy check names c: vs psutil name C:\\
if Platform.is_win32():
device_name = device_name.strip('\\').lower()
tags.append('device:{}'.format(device_name))
tags.append('device_name:{}'.format(_base_device_name(
part.device)))
for metric_name, metric_value in iteritems(
self._collect_part_metrics(part, disk_usage)):
self.gauge(metric_name, metric_value, tags=tags)
# Add in a disk read write or read only check
if self._service_check_rw:
rwro = {'rw', 'ro'} & set(part.opts.split(','))
if len(rwro) == 1:
self.service_check('disk.read_write',
AgentCheck.OK if rwro.pop() == 'rw' else
AgentCheck.CRITICAL,
tags=tags)
else:
self.service_check('disk.read_write',
AgentCheck.UNKNOWN,
tags=tags)
self.collect_latency_metrics()
