def __init__(self, agentConfig):
try:
self.config_store = get_config_store(agentConfig=agentConfig)
except Exception as e:
log.error('Failed to instantiate the config store client. '
'Auto-config only will be used. %s' % str(e))
agentConfig['sd_config_backend'] = None
self.config_store = get_config_store(agentConfig=agentConfig)
self.dockerutil = DockerUtil(config_store=self.config_store)
self.docker_client = self.dockerutil.client
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.kubeutil = None
log.error(
"Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s"
% str(ex))
if Platform.is_nomad():
self.nomadutil = NomadUtil()
elif Platform.is_ecs_instance():
self.ecsutil = ECSUtil()
self.VAR_MAPPING = {
'host': self._get_host_address,
'port': self._get_port,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
def test_detect_agent(self, mock_init, mock_inspect, mock_gw, mock_get):
mock_get.return_value = MockResponse({}, 404)
mock_init.return_value = None
mock_gw.return_value = "10.0.2.2"
mock_inspect.return_value = {
'NetworkSettings': {
'IPAddress': '10.0.0.42',
'Ports': {
'1234/tcp': '1234/tcp'
}
}
}
probe_calls = [
mock.call('http://10.0.0.42:1234/', timeout=1),
mock.call('http://10.0.2.2:51678/', timeout=1),
mock.call('http://localhost:51678/', timeout=1)
]
util = ECSUtil()
mock_get.reset_mock()
util._detect_agent()
mock_get.assert_has_calls(probe_calls)
def test_host_tags(self, mock_init, mock_inspect, mock_get):
mock_inspect.return_value = {}
mock_get.return_value = MockResponse({"Cluster": "default-xvello",
"Version": "Amazon ECS Agent - v1.14.1 (467c3d7)"}, 200)
mock_init.return_value = None
util = ECSUtil()
util.agent_url = 'http://dummy'
mock_get.reset_mock()
tags = util.get_host_tags()
self.assertEqual(['ecs_version:1.14.1'], tags)
mock_get.assert_called_once_with('http://dummy/v1/metadata', timeout=1)
def test_extract_tags(self, mock_init, mock_get):
mock_get.return_value = MockResponse({}, 404)
mock_init.return_value = None
util = ECSUtil()
util.agent_url = 'http://dummy'
mock_get.reset_mock()
mock_get.return_value = MockResponse({"Tasks": [{"Family": "dd-agent-latest", "Version": "12",
"Containers": [{"DockerId": CO_ID}]}]}, 200)
tags = util._get_cacheable_tags(CO_ID)
self.assertEqual(['task_name:dd-agent-latest', 'task_version:12'], tags)
mock_get.assert_called_once_with('http://dummy/v1/tasks', timeout=1)
def __init__(self, agentConfig):
try:
self.config_store = get_config_store(agentConfig=agentConfig)
except Exception as e:
log.error('Failed to instantiate the config store client. '
'Auto-config only will be used. %s' % str(e))
agentConfig['sd_config_backend'] = None
self.config_store = get_config_store(agentConfig=agentConfig)
self.dockerutil = DockerUtil(config_store=self.config_store)
self.docker_client = self.dockerutil.client
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.kubeutil = None
log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
if Platform.is_nomad():
self.nomadutil = NomadUtil()
elif Platform.is_ecs_instance():
self.ecsutil = ECSUtil()
self.VAR_MAPPING = {
'host': self._get_host_address,
'port': self._get_port,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
def test_host_tags(self, mock_init, mock_inspect, mock_get):
mock_inspect.return_value = {}
mock_get.return_value = MockResponse(
{
"Cluster": "default-xvello",
"Version": "Amazon ECS Agent - v1.14.1 (467c3d7)"
}, 200)
mock_init.return_value = None
util = ECSUtil()
util.agent_url = 'http://dummy'
mock_get.reset_mock()
tags = util.get_host_tags()
self.assertEqual(['ecs_version:1.14.1'], tags)
mock_get.assert_called_once_with('http://dummy/v1/metadata', timeout=1)
def test_detect_agent(self, mock_init, mock_inspect, mock_gw, mock_get):
mock_get.return_value = MockResponse({}, 404)
mock_init.return_value = None
mock_gw.return_value = "10.0.2.2"
mock_inspect.return_value = {'NetworkSettings': {'IPAddress': '10.0.0.42',
'Ports': {'1234/tcp': '1234/tcp'}}}
probe_calls = [mock.call('http://10.0.0.42:1234/', timeout=1),
mock.call('http://10.0.2.2:51678/', timeout=1),
mock.call('http://localhost:51678/', timeout=1)]
util = ECSUtil()
mock_get.reset_mock()
util._detect_agent()
mock_get.assert_has_calls(probe_calls)
def test_extract_tags(self, mock_init, mock_get):
mock_get.return_value = MockResponse({}, 404)
mock_init.return_value = None
util = ECSUtil()
util.agent_url = 'http://dummy'
mock_get.reset_mock()
mock_get.return_value = MockResponse(
{
"Tasks": [{
"Family": "dd-agent-latest",
"Version": "12",
"Containers": [{
"DockerId": CO_ID
}]
}]
}, 200)
tags = util._get_cacheable_tags(CO_ID)
self.assertEqual(['task_name:dd-agent-latest', 'task_version:12'],
tags)
mock_get.assert_called_once_with('http://dummy/v1/tasks', timeout=1)
class SDDockerBackend(AbstractSDBackend):
"""Docker-based service discovery"""
def __init__(self, agentConfig):
try:
self.config_store = get_config_store(agentConfig=agentConfig)
except Exception as e:
log.error('Failed to instantiate the config store client. '
'Auto-config only will be used. %s' % str(e))
agentConfig['sd_config_backend'] = None
self.config_store = get_config_store(agentConfig=agentConfig)
self.dockerutil = DockerUtil(config_store=self.config_store)
self.docker_client = self.dockerutil.client
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.kubeutil = None
log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
if Platform.is_nomad():
self.nomadutil = NomadUtil()
elif Platform.is_ecs_instance():
self.ecsutil = ECSUtil()
self.VAR_MAPPING = {
'host': self._get_host_address,
'port': self._get_port,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
def _make_fetch_state(self):
pod_list = []
if Platform.is_k8s():
if not self.kubeutil:
log.error("kubelet client not created, cannot retrieve pod list.")
else:
try:
pod_list = self.kubeutil.retrieve_pods_list().get('items', [])
except Exception as ex:
log.warning("Failed to retrieve pod list: %s" % str(ex))
return _SDDockerBackendConfigFetchState(self.docker_client.inspect_container, pod_list)
def update_checks(self, changed_containers):
state = self._make_fetch_state()
if Platform.is_k8s():
self.kubeutil.check_services_cache_freshness()
conf_reload_set = set()
for c_id in changed_containers:
checks = self._get_checks_to_refresh(state, c_id)
if checks:
conf_reload_set.update(set(checks))
if conf_reload_set:
self.reload_check_configs = conf_reload_set
def _get_checks_to_refresh(self, state, c_id):
"""Get the list of checks applied to a container from the identifier_to_checks cache in the config store.
Use the DATADOG_ID label or the image."""
inspect = state.inspect_container(c_id)
# If the container was removed we can't tell which check is concerned
# so we have to reload everything.
# Same thing if it's stopped and we're on Kubernetes in auto_conf mode
# because the pod was deleted and its template could have been in the annotations.
if not inspect or \
(not inspect.get('State', {}).get('Running')
and Platform.is_k8s() and not self.agentConfig.get('sd_config_backend')):
self.reload_check_configs = True
return
identifier = inspect.get('Config', {}).get('Labels', {}).get(DATADOG_ID) or \
self.dockerutil.image_name_extractor(inspect)
platform_kwargs = {}
if Platform.is_k8s():
kube_metadata = state.get_kube_config(c_id, 'metadata') or {}
platform_kwargs = {
'kube_annotations': kube_metadata.get('annotations'),
'kube_container_name': state.get_kube_container_name(c_id),
}
return self.config_store.get_checks_to_refresh(identifier, **platform_kwargs)
def _get_host_address(self, state, c_id, tpl_var):
"""Extract the container IP from a docker inspect object, or the kubelet API."""
c_inspect = state.inspect_container(c_id)
c_id, c_img = c_inspect.get('Id', ''), c_inspect.get('Config', {}).get('Image', '')
networks = c_inspect.get('NetworkSettings', {}).get('Networks') or {}
ip_dict = {}
for net_name, net_desc in networks.iteritems():
ip = net_desc.get('IPAddress')
if ip:
ip_dict[net_name] = ip
ip_addr = self._extract_ip_from_networks(ip_dict, tpl_var)
if ip_addr:
return ip_addr
# try to get the bridge (default) IP address
log.debug("No IP address was found in container %s (%s) "
"networks, trying with the IPAddress field" % (c_id[:12], c_img))
ip_addr = c_inspect.get('NetworkSettings', {}).get('IPAddress')
if ip_addr:
return ip_addr
if Platform.is_k8s():
# kubernetes case
log.debug("Couldn't find the IP address for container %s (%s), "
"using the kubernetes way." % (c_id[:12], c_img))
pod_ip = state.get_kube_config(c_id, 'status').get('podIP')
if pod_ip:
return pod_ip
if Platform.is_rancher():
# try to get the rancher IP address
log.debug("No IP address was found in container %s (%s) "
"trying with the Rancher label" % (c_id[:12], c_img))
ip_addr = c_inspect.get('Config', {}).get('Labels', {}).get(RANCHER_CONTAINER_IP)
if ip_addr:
return ip_addr.split('/')[0]
log.error("No IP address was found for container %s (%s)" % (c_id[:12], c_img))
return None
def _extract_ip_from_networks(self, ip_dict, tpl_var):
"""Extract a single IP from a dictionary made of network names and IPs."""
if not ip_dict:
return None
tpl_parts = tpl_var.split('_', 1)
# no specifier
if len(tpl_parts) < 2:
log.debug("No key was passed for template variable %s." % tpl_var)
return self._get_fallback_ip(ip_dict)
else:
res = ip_dict.get(tpl_parts[-1])
if res is None:
log.warning("The key passed for template variable %s was not found." % tpl_var)
return self._get_fallback_ip(ip_dict)
else:
return res
def _get_fallback_ip(self, ip_dict):
"""try to pick the bridge key, falls back to the value of the last key"""
if 'bridge' in ip_dict:
log.debug("Using the bridge network.")
return ip_dict['bridge']
else:
last_key = sorted(ip_dict.iterkeys())[-1]
log.debug("Trying with the last (sorted) network: '%s'." % last_key)
return ip_dict[last_key]
def _get_port(self, state, c_id, tpl_var):
"""Extract a port from a container_inspect or the k8s API given a template variable."""
container_inspect = state.inspect_container(c_id)
try:
ports = map(lambda x: x.split('/')[0], container_inspect['NetworkSettings']['Ports'].keys())
if len(ports) == 0: # There might be a key Port in NetworkSettings but no ports so we raise IndexError to check in ExposedPorts
raise IndexError
except (IndexError, KeyError, AttributeError):
# try to get ports from the docker API. Works if the image has an EXPOSE instruction
ports = map(lambda x: x.split('/')[0], container_inspect['Config'].get('ExposedPorts', {}).keys())
# if it failed, try with the kubernetes API
if not ports and Platform.is_k8s():
log.debug("Didn't find the port for container %s (%s), trying the kubernetes way." %
(c_id[:12], container_inspect.get('Config', {}).get('Image', '')))
spec = state.get_kube_container_spec(c_id)
if spec:
ports = [str(x.get('containerPort')) for x in spec.get('ports', [])]
ports = sorted(ports, key=int)
return self._extract_port_from_list(ports, tpl_var)
def _extract_port_from_list(self, ports, tpl_var):
if not ports:
return None
tpl_parts = tpl_var.split('_', 1)
if len(tpl_parts) == 1:
log.debug("No index was passed for template variable %s. "
"Trying with the last element." % tpl_var)
return ports[-1]
try:
idx = tpl_parts[-1]
return ports[int(idx)]
except ValueError:
log.error("Port index is not an integer. Using the last element instead.")
except IndexError:
log.error("Port index is out of range. Using the last element instead.")
return ports[-1]
def get_tags(self, state, c_id):
"""Extract useful tags from docker or platform APIs. These are collected by default."""
tags = []
ctr = state.inspect_container(c_id)
# TODO: extend with labels, container ID, etc.
tags.append('docker_image:%s' % self.dockerutil.image_name_extractor(ctr))
tags.append('image_name:%s' % self.dockerutil.image_tag_extractor(ctr, 0)[0])
tags.append('image_tag:%s' % self.dockerutil.image_tag_extractor(ctr, 1)[0])
if Platform.is_k8s():
pod_metadata = state.get_kube_config(c_id, 'metadata')
if pod_metadata is None:
log.warning("Failed to fetch pod metadata for container %s."
" Kubernetes tags may be missing." % c_id[:12])
return []
# get pod labels
kube_labels = pod_metadata.get('labels', {})
for label, value in kube_labels.iteritems():
tags.append('%s:%s' % (label, value))
# get kubernetes namespace
namespace = pod_metadata.get('namespace')
tags.append('kube_namespace:%s' % namespace)
# add creator tags
creator_tags = self.kubeutil.get_pod_creator_tags(pod_metadata)
tags.extend(creator_tags)
# add services tags
services = self.kubeutil.match_services_for_pod(pod_metadata)
for s in services:
if s is not None:
tags.append('kube_service:%s' % s)
elif Platform.is_swarm():
c_labels = state.inspect_container(c_id).get('Config', {}).get('Labels', {})
swarm_svc = c_labels.get(SWARM_SVC_LABEL)
if swarm_svc:
tags.append('swarm_service:%s' % swarm_svc)
elif Platform.is_rancher():
c_inspect = state.inspect_container(c_id)
service_name = c_inspect.get('Config', {}).get('Labels', {}).get(RANCHER_SVC_NAME)
stack_name = c_inspect.get('Config', {}).get('Labels', {}).get(RANCHER_STACK_NAME)
container_name = c_inspect.get('Config', {}).get('Labels', {}).get(RANCHER_CONTAINER_NAME)
if service_name:
tags.append('rancher_service:%s' % service_name)
if stack_name:
tags.append('rancher_stack:%s' % stack_name)
if container_name:
tags.append('rancher_container:%s' % container_name)
elif Platform.is_nomad():
nomad_tags = self.nomadutil.extract_container_tags(state.inspect_container(c_id))
if nomad_tags:
tags.extend(nomad_tags)
elif Platform.is_ecs_instance():
ecs_tags = self.ecsutil.extract_container_tags(state.inspect_container(c_id))
tags.extend(ecs_tags)
return tags
def _get_additional_tags(self, state, c_id, *args):
tags = []
if Platform.is_k8s():
pod_metadata = state.get_kube_config(c_id, 'metadata')
pod_spec = state.get_kube_config(c_id, 'spec')
if pod_metadata is None or pod_spec is None:
log.warning("Failed to fetch pod metadata or pod spec for container %s."
" Additional Kubernetes tags may be missing." % c_id[:12])
return []
tags.append('node_name:%s' % pod_spec.get('nodeName'))
tags.append('pod_name:%s' % pod_metadata.get('name'))
return tags
def get_configs(self):
"""Get the config for all docker containers running on the host."""
configs = {}
state = self._make_fetch_state()
containers = [(
self.dockerutil.image_name_extractor(container),
container.get('Id'), container.get('Labels')
) for container in self.docker_client.containers()]
if Platform.is_k8s():
self.kubeutil.check_services_cache_freshness()
for image, cid, labels in containers:
try:
# value of the DATADOG_ID tag or the image name if the label is missing
identifier = self.get_config_id(image, labels)
check_configs = self._get_check_configs(state, cid, identifier) or []
for conf in check_configs:
source, (check_name, init_config, instance) = conf
# build instances list if needed
if configs.get(check_name) is None:
configs[check_name] = (source, (init_config, [instance]))
else:
conflict_init_msg = 'Different versions of `init_config` found for check {}. ' \
'Keeping the first one found.'
if configs[check_name][1][0] != init_config:
log.warning(conflict_init_msg.format(check_name))
configs[check_name][1][1].append(instance)
except Exception:
log.exception('Building config for container %s based on image %s using service '
'discovery failed, leaving it alone.' % (cid[:12], image))
return configs
def get_config_id(self, image, labels):
"""Look for a DATADOG_ID label, return its value or the image name if missing"""
return labels.get(DATADOG_ID) or image
def _get_check_configs(self, state, c_id, identifier):
"""Retrieve configuration templates and fill them with data pulled from docker and tags."""
platform_kwargs = {}
if Platform.is_k8s():
kube_metadata = state.get_kube_config(c_id, 'metadata') or {}
platform_kwargs = {
'kube_container_name': state.get_kube_container_name(c_id),
'kube_annotations': kube_metadata.get('annotations'),
}
config_templates = self._get_config_templates(identifier, **platform_kwargs)
if not config_templates:
return None
check_configs = []
tags = self.get_tags(state, c_id)
for config_tpl in config_templates:
source, config_tpl = config_tpl
check_name, init_config_tpl, instance_tpl, variables = config_tpl
# insert tags in instance_tpl and process values for template variables
instance_tpl, var_values = self._fill_tpl(state, c_id, instance_tpl, variables, tags)
tpl = self._render_template(init_config_tpl or {}, instance_tpl or {}, var_values)
if tpl and len(tpl) == 2:
init_config, instance = tpl
check_configs.append((source, (check_name, init_config, instance)))
return check_configs
def _get_config_templates(self, identifier, **platform_kwargs):
"""Extract config templates for an identifier from a K/V store and returns it as a dict object."""
config_backend = self.agentConfig.get('sd_config_backend')
templates = []
if config_backend is None:
auto_conf = True
else:
auto_conf = False
# format [(source, ('ident', {init_tpl}, {instance_tpl}))]
raw_tpls = self.config_store.get_check_tpls(identifier, auto_conf=auto_conf, **platform_kwargs)
for tpl in raw_tpls:
# each template can come from either auto configuration or user-supplied templates
try:
source, (check_name, init_config_tpl, instance_tpl) = tpl
except (TypeError, IndexError, ValueError):
log.debug('No template was found for identifier %s, leaving it alone: %s' % (identifier, tpl))
return None
try:
# build a list of all variables to replace in the template
variables = self.PLACEHOLDER_REGEX.findall(str(init_config_tpl)) + \
self.PLACEHOLDER_REGEX.findall(str(instance_tpl))
variables = map(lambda x: x.strip('%'), variables)
if not isinstance(init_config_tpl, dict):
init_config_tpl = json.loads(init_config_tpl or '{}')
if not isinstance(instance_tpl, dict):
instance_tpl = json.loads(instance_tpl or '{}')
except json.JSONDecodeError:
log.exception('Failed to decode the JSON template fetched for check {0}. Its configuration'
' by service discovery failed for ident {1}.'.format(check_name, identifier))
return None
templates.append((source,
(check_name, init_config_tpl, instance_tpl, variables)))
return templates
def _fill_tpl(self, state, c_id, instance_tpl, variables, tags=None):
"""Add container tags to instance templates and build a
dict from template variable names and their values."""
var_values = {}
c_image = state.inspect_container(c_id).get('Config', {}).get('Image', '')
# add default tags to the instance
if tags:
tpl_tags = instance_tpl.get('tags', [])
tags += tpl_tags if isinstance(tpl_tags, list) else [tpl_tags]
instance_tpl['tags'] = list(set(tags))
for var in variables:
# variables can be suffixed with an index in case several values are found
if var.split('_')[0] in self.VAR_MAPPING:
try:
res = self.VAR_MAPPING[var.split('_')[0]](state, c_id, var)
if res is None:
raise ValueError("Invalid value for variable %s." % var)
var_values[var] = res
except Exception as ex:
log.error("Could not find a value for the template variable %s for container %s "
"(%s): %s" % (var, c_id[:12], c_image, str(ex)))
else:
log.error("No method was found to interpolate template variable %s for container %s "
"(%s)." % (var, c_id[:12], c_image))
return instance_tpl, var_values
class SDDockerBackend(AbstractSDBackend):
"""Docker-based service discovery"""
def __init__(self, agentConfig):
try:
self.config_store = get_config_store(agentConfig=agentConfig)
except Exception as e:
log.error('Failed to instantiate the config store client. '
'Auto-config only will be used. %s' % str(e))
agentConfig['sd_config_backend'] = None
self.config_store = get_config_store(agentConfig=agentConfig)
self.dockerutil = DockerUtil(config_store=self.config_store)
self.docker_client = self.dockerutil.client
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.kubeutil = None
log.error(
"Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s"
% str(ex))
if Platform.is_nomad():
self.nomadutil = NomadUtil()
elif Platform.is_ecs_instance():
self.ecsutil = ECSUtil()
self.VAR_MAPPING = {
'host': self._get_host_address,
'port': self._get_port,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
def _make_fetch_state(self):
pod_list = []
if Platform.is_k8s():
if not self.kubeutil:
log.error(
"kubelet client not created, cannot retrieve pod list.")
else:
try:
pod_list = self.kubeutil.retrieve_pods_list().get(
'items', [])
except Exception as ex:
log.warning("Failed to retrieve pod list: %s" % str(ex))
return _SDDockerBackendConfigFetchState(
self.docker_client.inspect_container, pod_list)
def update_checks(self, changed_containers):
state = self._make_fetch_state()
if Platform.is_k8s():
self.kubeutil.check_services_cache_freshness()
conf_reload_set = set()
for c_id in changed_containers:
checks = self._get_checks_to_refresh(state, c_id)
if checks:
conf_reload_set.update(set(checks))
if conf_reload_set:
self.reload_check_configs = conf_reload_set
def _get_checks_to_refresh(self, state, c_id):
"""Get the list of checks applied to a container from the identifier_to_checks cache in the config store.
Use the DATADOG_ID label or the image."""
inspect = state.inspect_container(c_id)
# If the container was removed we can't tell which check is concerned
# so we have to reload everything.
# Same thing if it's stopped and we're on Kubernetes in auto_conf mode
# because the pod was deleted and its template could have been in the annotations.
if not inspect or \
(not inspect.get('State', {}).get('Running')
and Platform.is_k8s() and not self.agentConfig.get('sd_config_backend')):
self.reload_check_configs = True
return
identifier = inspect.get('Config', {}).get('Labels', {}).get(DATADOG_ID) or \
self.dockerutil.image_name_extractor(inspect)
platform_kwargs = {}
if Platform.is_k8s():
kube_metadata = state.get_kube_config(c_id, 'metadata') or {}
platform_kwargs = {
'kube_annotations': kube_metadata.get('annotations'),
'kube_container_name': state.get_kube_container_name(c_id),
}
return self.config_store.get_checks_to_refresh(identifier,
**platform_kwargs)
def _get_host_address(self, state, c_id, tpl_var):
"""Extract the container IP from a docker inspect object, or the kubelet API."""
c_inspect = state.inspect_container(c_id)
c_id, c_img = c_inspect.get('Id',
''), c_inspect.get('Config',
{}).get('Image', '')
networks = c_inspect.get('NetworkSettings', {}).get('Networks') or {}
ip_dict = {}
for net_name, net_desc in networks.iteritems():
ip = net_desc.get('IPAddress')
if ip:
ip_dict[net_name] = ip
ip_addr = self._extract_ip_from_networks(ip_dict, tpl_var)
if ip_addr:
return ip_addr
# try to get the bridge (default) IP address
log.debug("No IP address was found in container %s (%s) "
"networks, trying with the IPAddress field" %
(c_id[:12], c_img))
ip_addr = c_inspect.get('NetworkSettings', {}).get('IPAddress')
if ip_addr:
return ip_addr
if Platform.is_k8s():
# kubernetes case
log.debug("Couldn't find the IP address for container %s (%s), "
"using the kubernetes way." % (c_id[:12], c_img))
pod_ip = state.get_kube_config(c_id, 'status').get('podIP')
if pod_ip:
return pod_ip
if Platform.is_rancher():
# try to get the rancher IP address
log.debug("No IP address was found in container %s (%s) "
"trying with the Rancher label" % (c_id[:12], c_img))
ip_addr = c_inspect.get('Config',
{}).get('Labels',
{}).get(RANCHER_CONTAINER_IP)
if ip_addr:
return ip_addr.split('/')[0]
log.error("No IP address was found for container %s (%s)" %
(c_id[:12], c_img))
return None
def _extract_ip_from_networks(self, ip_dict, tpl_var):
"""Extract a single IP from a dictionary made of network names and IPs."""
if not ip_dict:
return None
tpl_parts = tpl_var.split('_', 1)
# no specifier
if len(tpl_parts) < 2:
log.debug("No key was passed for template variable %s." % tpl_var)
return self._get_fallback_ip(ip_dict)
else:
res = ip_dict.get(tpl_parts[-1])
if res is None:
log.warning(
"The key passed for template variable %s was not found." %
tpl_var)
return self._get_fallback_ip(ip_dict)
else:
return res
def _get_fallback_ip(self, ip_dict):
"""try to pick the bridge key, falls back to the value of the last key"""
if 'bridge' in ip_dict:
log.debug("Using the bridge network.")
return ip_dict['bridge']
else:
last_key = sorted(ip_dict.iterkeys())[-1]
log.debug("Trying with the last (sorted) network: '%s'." %
last_key)
return ip_dict[last_key]
def _get_port(self, state, c_id, tpl_var):
"""Extract a port from a container_inspect or the k8s API given a template variable."""
container_inspect = state.inspect_container(c_id)
try:
ports = map(lambda x: x.split('/')[0],
container_inspect['NetworkSettings']['Ports'].keys())
if len(
ports
) == 0: # There might be a key Port in NetworkSettings but no ports so we raise IndexError to check in ExposedPorts
raise IndexError
except (IndexError, KeyError, AttributeError):
# try to get ports from the docker API. Works if the image has an EXPOSE instruction
ports = map(
lambda x: x.split('/')[0],
container_inspect['Config'].get('ExposedPorts', {}).keys())
# if it failed, try with the kubernetes API
if not ports and Platform.is_k8s():
log.debug(
"Didn't find the port for container %s (%s), trying the kubernetes way."
% (c_id[:12], container_inspect.get('Config', {}).get(
'Image', '')))
spec = state.get_kube_container_spec(c_id)
if spec:
ports = [
str(x.get('containerPort'))
for x in spec.get('ports', [])
]
ports = sorted(ports, key=int)
return self._extract_port_from_list(ports, tpl_var)
def _extract_port_from_list(self, ports, tpl_var):
if not ports:
return None
tpl_parts = tpl_var.split('_', 1)
if len(tpl_parts) == 1:
log.debug("No index was passed for template variable %s. "
"Trying with the last element." % tpl_var)
return ports[-1]
try:
idx = tpl_parts[-1]
return ports[int(idx)]
except ValueError:
log.error(
"Port index is not an integer. Using the last element instead."
)
except IndexError:
log.error(
"Port index is out of range. Using the last element instead.")
return ports[-1]
def get_tags(self, state, c_id):
"""Extract useful tags from docker or platform APIs. These are collected by default."""
c_inspect = state.inspect_container(c_id)
tags = self.dockerutil.extract_container_tags(c_inspect)
if Platform.is_k8s():
pod_metadata = state.get_kube_config(c_id, 'metadata')
if pod_metadata is None:
log.warning("Failed to fetch pod metadata for container %s."
" Kubernetes tags may be missing." % c_id[:12])
return []
# get pod labels
kube_labels = pod_metadata.get('labels', {})
for label, value in kube_labels.iteritems():
tags.append('%s:%s' % (label, value))
# get kubernetes namespace
namespace = pod_metadata.get('namespace')
tags.append('kube_namespace:%s' % namespace)
# add creator tags
creator_tags = self.kubeutil.get_pod_creator_tags(pod_metadata)
tags.extend(creator_tags)
# add services tags
services = self.kubeutil.match_services_for_pod(pod_metadata)
for s in services:
if s is not None:
tags.append('kube_service:%s' % s)
elif Platform.is_swarm():
c_labels = c_inspect.get('Config', {}).get('Labels', {})
swarm_svc = c_labels.get(SWARM_SVC_LABEL)
if swarm_svc:
tags.append('swarm_service:%s' % swarm_svc)
elif Platform.is_rancher():
service_name = c_inspect.get('Config',
{}).get('Labels',
{}).get(RANCHER_SVC_NAME)
stack_name = c_inspect.get('Config',
{}).get('Labels',
{}).get(RANCHER_STACK_NAME)
container_name = c_inspect.get('Config', {}).get(
'Labels', {}).get(RANCHER_CONTAINER_NAME)
if service_name:
tags.append('rancher_service:%s' % service_name)
if stack_name:
tags.append('rancher_stack:%s' % stack_name)
if container_name:
tags.append('rancher_container:%s' % container_name)
elif Platform.is_nomad():
nomad_tags = self.nomadutil.extract_container_tags(c_inspect)
if nomad_tags:
tags.extend(nomad_tags)
elif Platform.is_ecs_instance():
ecs_tags = self.ecsutil.extract_container_tags(c_inspect)
tags.extend(ecs_tags)
return tags
def _get_additional_tags(self, state, c_id, *args):
tags = []
if Platform.is_k8s():
pod_metadata = state.get_kube_config(c_id, 'metadata')
pod_spec = state.get_kube_config(c_id, 'spec')
if pod_metadata is None or pod_spec is None:
log.warning(
"Failed to fetch pod metadata or pod spec for container %s."
" Additional Kubernetes tags may be missing." % c_id[:12])
return []
tags.append('node_name:%s' % pod_spec.get('nodeName'))
tags.append('pod_name:%s' % pod_metadata.get('name'))
return tags
def get_configs(self):
"""Get the config for all docker containers running on the host."""
configs = {}
state = self._make_fetch_state()
containers = [(self.dockerutil.image_name_extractor(container),
container.get('Id'), container.get('Labels'))
for container in self.docker_client.containers()]
if Platform.is_k8s():
self.kubeutil.check_services_cache_freshness()
for image, cid, labels in containers:
try:
# value of the DATADOG_ID tag or the image name if the label is missing
identifier = self.get_config_id(image, labels)
check_configs = self._get_check_configs(
state, cid, identifier) or []
for conf in check_configs:
source, (check_name, init_config, instance) = conf
# build instances list if needed
if configs.get(check_name) is None:
configs[check_name] = (source, (init_config,
[instance]))
else:
conflict_init_msg = 'Different versions of `init_config` found for check {}. ' \
'Keeping the first one found.'
if configs[check_name][1][0] != init_config:
log.warning(conflict_init_msg.format(check_name))
configs[check_name][1][1].append(instance)
except Exception:
log.exception(
'Building config for container %s based on image %s using service '
'discovery failed, leaving it alone.' % (cid[:12], image))
return configs
def get_config_id(self, image, labels):
"""Look for a DATADOG_ID label, return its value or the image name if missing"""
return labels.get(DATADOG_ID) or image
def _get_check_configs(self, state, c_id, identifier):
"""Retrieve configuration templates and fill them with data pulled from docker and tags."""
platform_kwargs = {}
if Platform.is_k8s():
kube_metadata = state.get_kube_config(c_id, 'metadata') or {}
platform_kwargs = {
'kube_container_name': state.get_kube_container_name(c_id),
'kube_annotations': kube_metadata.get('annotations'),
}
config_templates = self._get_config_templates(identifier,
**platform_kwargs)
if not config_templates:
return None
check_configs = []
tags = self.get_tags(state, c_id)
for config_tpl in config_templates:
source, config_tpl = config_tpl
check_name, init_config_tpl, instance_tpl, variables = config_tpl
# insert tags in instance_tpl and process values for template variables
instance_tpl, var_values = self._fill_tpl(state, c_id,
instance_tpl, variables,
tags)
tpl = self._render_template(init_config_tpl or {}, instance_tpl
or {}, var_values)
if tpl and len(tpl) == 2:
init_config, instance = tpl
check_configs.append(
(source, (check_name, init_config, instance)))
return check_configs
def _get_config_templates(self, identifier, **platform_kwargs):
"""Extract config templates for an identifier from a K/V store and returns it as a dict object."""
config_backend = self.agentConfig.get('sd_config_backend')
templates = []
if config_backend is None:
auto_conf = True
else:
auto_conf = False
# format [(source, ('ident', {init_tpl}, {instance_tpl}))]
raw_tpls = self.config_store.get_check_tpls(identifier,
auto_conf=auto_conf,
**platform_kwargs)
for tpl in raw_tpls:
# each template can come from either auto configuration or user-supplied templates
try:
source, (check_name, init_config_tpl, instance_tpl) = tpl
except (TypeError, IndexError, ValueError):
log.debug(
'No template was found for identifier %s, leaving it alone: %s'
% (identifier, tpl))
return None
try:
# build a list of all variables to replace in the template
variables = self.PLACEHOLDER_REGEX.findall(str(init_config_tpl)) + \
self.PLACEHOLDER_REGEX.findall(str(instance_tpl))
variables = map(lambda x: x.strip('%'), variables)
if not isinstance(init_config_tpl, dict):
init_config_tpl = json.loads(init_config_tpl or '{}')
if not isinstance(instance_tpl, dict):
instance_tpl = json.loads(instance_tpl or '{}')
except json.JSONDecodeError:
log.exception(
'Failed to decode the JSON template fetched for check {0}. Its configuration'
' by service discovery failed for ident {1}.'.format(
check_name, identifier))
return None
templates.append((source, (check_name, init_config_tpl,
instance_tpl, variables)))
return templates
def _fill_tpl(self, state, c_id, instance_tpl, variables, tags=None):
"""Add container tags to instance templates and build a
dict from template variable names and their values."""
var_values = {}
c_image = state.inspect_container(c_id).get('Config',
{}).get('Image', '')
# add default tags to the instance
if tags:
tpl_tags = instance_tpl.get('tags', [])
tags += tpl_tags if isinstance(tpl_tags, list) else [tpl_tags]
instance_tpl['tags'] = list(set(tags))
for var in variables:
# variables can be suffixed with an index in case several values are found
if var.split('_')[0] in self.VAR_MAPPING:
try:
res = self.VAR_MAPPING[var.split('_')[0]](state, c_id, var)
if res is None:
raise ValueError("Invalid value for variable %s." %
var)
var_values[var] = res
except Exception as ex:
log.error(
"Could not find a value for the template variable %s for container %s "
"(%s): %s" % (var, c_id[:12], c_image, str(ex)))
else:
log.error(
"No method was found to interpolate template variable %s for container %s "
"(%s)." % (var, c_id[:12], c_image))
return instance_tpl, var_values