def _get_hostname_metadata(self):
"""
Returns a dictionnary that contains hostname metadata.
"""
metadata = EC2.get_metadata(self.agentConfig)
if metadata.get('hostname'):
metadata['ec2-hostname'] = metadata.get('hostname')
del metadata['hostname']
if self.agentConfig.get('hostname'):
metadata['agent-hostname'] = self.agentConfig.get('hostname')
else:
try:
metadata["socket-hostname"] = socket.gethostname()
except Exception:
pass
try:
metadata["socket-fqdn"] = socket.getfqdn()
except Exception:
pass
metadata["hostname"] = self.hostname
metadata["timezones"] = self._decode_tzname(time.tzname)
# Add cloud provider aliases
host_aliases = GCE.get_host_aliases(self.agentConfig)
if host_aliases:
metadata['host_aliases'] = host_aliases
return metadata
def _set_agent_config_hostname(self, agentConfig):
# Try to fetch instance Id from EC2 if not hostname has been set
# in the config file.
# DEPRECATED
if agentConfig.get('hostname') is None and agentConfig.get('use_ec2_instance_id'):
instanceId = EC2.get_instance_id(agentConfig)
if instanceId is not None:
log.info("Running on EC2, instanceId: %s" % instanceId)
agentConfig['hostname'] = instanceId
else:
log.info('Not running on EC2, using hostname to identify this server')
return agentConfig
def test_metadata(self):
# Reset metadata just to be sure
EC2.metadata = {}
# Test gathering metadata from ec2
start = time.time()
d = EC2.get_metadata({'collect_instance_metadata': True})
end = time.time()
self.assertTrue(isinstance(d, types.DictType))
# Either we're on ec2 or we're not (at least 7 attributes expected)
assert len(d) == 0 or len(d) >= 7, d
if "instance-id" in d:
assert d["instance-id"].startswith("i-"), d
assert end - start <= 1.15, "It took %s seconds to get ec2 metadata" % (end-start)
def _populate_payload_metadata(self, payload, check_statuses, start_event=True):
"""
Periodically populate the payload with metadata related to the system, host, and/or checks.
"""
now = time.time()
# Include system stats on first postback
if start_event and self._is_first_run():
payload['systemStats'] = self.agentConfig.get('system_stats', {})
# Also post an event in the newsfeed
payload['events']['System'] = [{
'api_key': self.agentConfig['api_key'],
'host': self.hostname,
'timestamp': now,
'event_type':'Agent Startup',
'msg_text': 'Version %s' % get_version()
}]
# Periodically send the host metadata.
if self._should_send_additional_data('host_metadata'):
# gather metadata with gohai
gohai_metadata = self._run_gohai_metadata()
if gohai_metadata:
payload['gohai'] = gohai_metadata
payload['systemStats'] = get_system_stats(
proc_path=self.agentConfig.get('procfs_path', '/proc').rstrip('/')
)
payload['meta'] = self._get_hostname_metadata()
self.hostname_metadata_cache = payload['meta']
# Add static tags from the configuration file
host_tags = []
if self.agentConfig['tags'] is not None:
host_tags.extend([unicode(tag.strip())
for tag in self.agentConfig['tags'].split(",")])
if self.agentConfig['collect_ec2_tags']:
host_tags.extend(EC2.get_tags(self.agentConfig))
if host_tags:
payload['host-tags']['system'] = host_tags
# If required by the user, let's create the dd_check:xxx host tags
if self.agentConfig['create_dd_check_tags']:
app_tags_list = [DD_CHECK_TAG.format(c.name) for c in self.initialized_checks_d]
app_tags_list.extend([DD_CHECK_TAG.format(cname) for cname
in JMXFiles.get_jmx_appnames()])
if 'system' not in payload['host-tags']:
payload['host-tags']['system'] = []
payload['host-tags']['system'].extend(app_tags_list)
GCE_tags = GCE.get_tags(self.agentConfig)
if GCE_tags is not None:
payload['host-tags'][GCE.SOURCE_TYPE_NAME] = GCE_tags
# Log the metadata on the first run
if self._is_first_run():
log.info("Hostnames: %s, tags: %s" %
(repr(self.hostname_metadata_cache), payload['host-tags']))
# Periodically send extra hosts metadata (vsphere)
# Metadata of hosts that are not the host where the agent runs, not all the checks use
# that
external_host_tags = []
if self._should_send_additional_data('external_host_tags'):
for check in self.initialized_checks_d:
try:
getter = getattr(check, 'get_external_host_tags')
check_tags = getter()
external_host_tags.extend(check_tags)
except AttributeError:
pass
if external_host_tags:
payload['external_host_tags'] = external_host_tags
# Periodically send agent_checks metadata
if self._should_send_additional_data('agent_checks'):
# Add agent checks statuses and error/warning messages
agent_checks = []
for check in check_statuses:
if check.instance_statuses is not None:
for i, instance_status in enumerate(check.instance_statuses):
agent_checks.append(
(
check.name, check.source_type_name,
instance_status.instance_id,
instance_status.status,
# put error message or list of warning messages in the same field
# it will be handled by the UI
instance_status.error or instance_status.warnings or "",
check.service_metadata[i]
)
)
else:
agent_checks.append(
(
check.name, check.source_type_name,
"initialization",
check.status, repr(check.init_failed_error)
)
)
payload['agent_checks'] = agent_checks
payload['meta'] = self.hostname_metadata_cache # add hostname metadata
def get_hostname(config=None):
"""
Get the canonical host name this agent should identify as. This is
the authoritative source of the host name for the agent.
Tries, in order:
* agent config (datadog.conf, "hostname:")
* 'hostname -f' (on unix)
* socket.gethostname()
"""
hostname = None
# first, try the config
if config is None:
from config import get_config
config = get_config(parse_args=True)
config_hostname = config.get('hostname')
if config_hostname and is_valid_hostname(config_hostname):
return config_hostname
# Try to get GCE instance name
gce_hostname = GCE.get_hostname(config)
if gce_hostname is not None:
if is_valid_hostname(gce_hostname):
return gce_hostname
# Try to get the docker hostname
if Platform.is_containerized():
# First we try from the Docker API
docker_util = DockerUtil()
docker_hostname = docker_util.get_hostname(use_default_gw=False)
if docker_hostname is not None and is_valid_hostname(docker_hostname):
hostname = docker_hostname
elif Platform.is_k8s(): # Let's try from the kubelet
kube_util = KubeUtil()
_, kube_hostname = kube_util.get_node_info()
if kube_hostname is not None and is_valid_hostname(kube_hostname):
hostname = kube_hostname
# then move on to os-specific detection
if hostname is None:
if Platform.is_unix() or Platform.is_solaris():
unix_hostname = _get_hostname_unix()
if unix_hostname and is_valid_hostname(unix_hostname):
hostname = unix_hostname
# if we have an ec2 default hostname, see if there's an instance-id available
if (Platform.is_ecs_instance()) or (hostname is not None and EC2.is_default(hostname)):
instanceid = EC2.get_instance_id(config)
if instanceid:
hostname = instanceid
# fall back on socket.gethostname(), socket.getfqdn() is too unreliable
if hostname is None:
try:
socket_hostname = socket.gethostname()
except socket.error:
socket_hostname = None
if socket_hostname and is_valid_hostname(socket_hostname):
hostname = socket_hostname
if hostname is None:
log.critical('Unable to reliably determine host name. You can define one in datadog.conf or in your hosts file')
raise Exception('Unable to reliably determine host name. You can define one in datadog.conf or in your hosts file')
return hostname
def get_hostname(config=None):
"""
Get the canonical host name this agent should identify as. This is
the authoritative source of the host name for the agent.
Tries, in order:
* agent config (datadog.conf, "hostname:")
* 'hostname -f' (on unix)
* socket.gethostname()
"""
hostname = None
# first, try the config
if config is None:
from config import get_config
config = get_config(parse_args=True)
config_hostname = config.get('hostname')
if config_hostname and is_valid_hostname(config_hostname):
return config_hostname
# Try to get GCE instance name
gce_hostname = GCE.get_hostname(config)
if gce_hostname is not None:
if is_valid_hostname(gce_hostname):
return gce_hostname
# Try to get the docker hostname
if Platform.is_containerized():
# First we try from the Docker API
docker_util = DockerUtil()
docker_hostname = docker_util.get_hostname(use_default_gw=False)
if docker_hostname is not None and is_valid_hostname(docker_hostname):
hostname = docker_hostname
elif Platform.is_k8s(): # Let's try from the kubelet
try:
kube_util = KubeUtil()
except Exception as ex:
log.error("Couldn't instantiate the kubernetes client, "
"getting the k8s hostname won't work. Error: %s" %
str(ex))
else:
_, kube_hostname = kube_util.get_node_info()
if kube_hostname is not None and is_valid_hostname(
kube_hostname):
hostname = kube_hostname
# then move on to os-specific detection
if hostname is None:
if Platform.is_unix() or Platform.is_solaris():
unix_hostname = _get_hostname_unix()
if unix_hostname and is_valid_hostname(unix_hostname):
hostname = unix_hostname
# if we don't have a hostname, or we have an ec2 default hostname,
# see if there's an instance-id available
if not Platform.is_windows() and (hostname is None
or Platform.is_ecs_instance()
or EC2.is_default(hostname)):
instanceid = EC2.get_instance_id(config)
if instanceid:
hostname = instanceid
# fall back on socket.gethostname(), socket.getfqdn() is too unreliable
if hostname is None:
try:
socket_hostname = socket.gethostname()
except socket.error:
socket_hostname = None
if socket_hostname and is_valid_hostname(socket_hostname):
hostname = socket_hostname
if hostname is None:
log.critical(
'Unable to reliably determine host name. You can define one in datadog.conf or in your hosts file'
)
raise Exception(
'Unable to reliably determine host name. You can define one in datadog.conf or in your hosts file'
)
return hostname
def test_is_default_hostname(self):
for hostname in ['ip-172-31-16-235', 'domU-12-31-38-00-A4-A2', 'domU-12-31-39-02-14-35']:
self.assertTrue(EC2.is_default(hostname))
for hostname in ['i-672d49da', 'localhost', 'robert.redf.org']:
self.assertFalse(EC2.is_default(hostname))