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.kubeutil = None
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
log.error(
"Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s"
% str(ex))
self.metadata_collector = MetadataCollector()
self.VAR_MAPPING = {
'host': self._get_host_address,
'pid': self._get_container_pid,
'port': self._get_port,
'container-name': self._get_container_name,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
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))
self.metadata_collector = MetadataCollector()
self.VAR_MAPPING = {
'host': self._get_host_address,
'pid': self._get_container_pid,
'port': self._get_port,
'container-name': self._get_container_name,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
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.kubeutil = None
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
self.metadata_collector = MetadataCollector()
self.VAR_MAPPING = {
'host': self._get_host_address,
'pid': self._get_container_pid,
'port': self._get_port,
'container-name': self._get_container_name,
'tags': self._get_additional_tags,
}
# docker labels we'll add as tags to all instances SD configures
self.docker_labels_as_tags = agentConfig.get('docker_labels_as_tags', '')
if self.docker_labels_as_tags:
self.docker_labels_as_tags = [label.strip() for label in self.docker_labels_as_tags.split(',')]
else:
self.docker_labels_as_tags = []
AbstractSDBackend.__init__(self, agentConfig)
def init(self):
try:
instance = self.instances[0]
# Getting custom tags for service checks when docker is down
self.custom_tags = instance.get("tags", [])
self.docker_util = DockerUtil()
if not self.docker_util.client:
raise Exception("Failed to initialize Docker client.")
self.docker_gateway = DockerUtil.get_gateway()
self.metadata_collector = MetadataCollector()
self.kubeutil = None
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
# We configure the check with the right cgroup settings for this host
# Just needs to be done once
self._mountpoints = self.docker_util.get_mountpoints(CGROUP_METRICS)
self._latest_size_query = 0
self._filtered_containers = set()
self._disable_net_metrics = False
# Set tagging options
# The collect_labels_as_tags is legacy, only tagging docker metrics.
# It is replaced by docker_labels_as_tags in config.cfg.
# We keep this line for backward compatibility.
if "collect_labels_as_tags" in instance:
self.collect_labels_as_tags = instance.get("collect_labels_as_tags")
self.kube_pod_tags = {}
self.use_histogram = _is_affirmative(instance.get('use_histogram', False))
performance_tags = instance.get("performance_tags", DEFAULT_PERFORMANCE_TAGS)
self.tag_names = {
CONTAINER: instance.get("container_tags", DEFAULT_CONTAINER_TAGS),
PERFORMANCE: performance_tags,
IMAGE: instance.get('image_tags', DEFAULT_IMAGE_TAGS)
}
# Set filtering settings
if self.docker_util.filtering_enabled:
self.tag_names[FILTERED] = self.docker_util.filtered_tag_names
# Container network mapping cache
self.network_mappings = {}
# get the health check whitelist
self.whitelist_patterns = None
health_scs_whitelist = instance.get('health_service_check_whitelist', [])
if health_scs_whitelist:
patterns, whitelist_tags = compile_filter_rules(health_scs_whitelist)
self.whitelist_patterns = set(patterns)
self.tag_names[HEALTHCHECK] = set(whitelist_tags)
# Other options
self.collect_image_stats = _is_affirmative(instance.get('collect_images_stats', False))
self.collect_container_size = _is_affirmative(instance.get('collect_container_size', False))
self.collect_container_count = _is_affirmative(instance.get('collect_container_count', False))
self.collect_volume_count = _is_affirmative(instance.get('collect_volume_count', False))
self.collect_events = _is_affirmative(instance.get('collect_events', True))
self.event_attributes_as_tags = instance.get('event_attributes_as_tags', [])
self.collect_image_size = _is_affirmative(instance.get('collect_image_size', False))
self.collect_disk_stats = _is_affirmative(instance.get('collect_disk_stats', False))
self.collect_exit_codes = _is_affirmative(instance.get('collect_exit_codes', False))
self.collect_ecs_tags = _is_affirmative(instance.get('ecs_tags', True)) and Platform.is_ecs_instance()
self.filtered_event_types = tuple(instance.get("filtered_event_types", DEFAULT_FILTERED_EVENT_TYPES))
self.capped_metrics = instance.get('capped_metrics')
except Exception as e:
self.log.critical(e)
self.warning("Initialization failed. Will retry at next iteration")
else:
self.init_success = True
class DockerDaemon(AgentCheck):
"""Collect metrics and events from Docker API and cgroups."""
def __init__(self, name, init_config, agentConfig, instances=None):
if instances is not None and len(instances) > 1:
raise Exception("Docker check only supports one configured instance.")
AgentCheck.__init__(self, name, init_config,
agentConfig, instances=instances)
self.init_success = False
self._service_discovery = agentConfig.get('service_discovery') and \
agentConfig.get('service_discovery_backend') == 'docker'
global_labels_as_tags = agentConfig.get('docker_labels_as_tags')
if global_labels_as_tags:
self.collect_labels_as_tags = [label.strip() for label in global_labels_as_tags.split(',')]
else:
self.collect_labels_as_tags = DEFAULT_LABELS_AS_TAGS
self.init()
def init(self):
try:
instance = self.instances[0]
# Getting custom tags for service checks when docker is down
self.custom_tags = instance.get("tags", [])
self.docker_util = DockerUtil()
if not self.docker_util.client:
raise Exception("Failed to initialize Docker client.")
self.docker_gateway = DockerUtil.get_gateway()
self.metadata_collector = MetadataCollector()
self.kubeutil = None
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
# We configure the check with the right cgroup settings for this host
# Just needs to be done once
self._mountpoints = self.docker_util.get_mountpoints(CGROUP_METRICS)
self._latest_size_query = 0
self._filtered_containers = set()
self._disable_net_metrics = False
# Set tagging options
# The collect_labels_as_tags is legacy, only tagging docker metrics.
# It is replaced by docker_labels_as_tags in config.cfg.
# We keep this line for backward compatibility.
if "collect_labels_as_tags" in instance:
self.collect_labels_as_tags = instance.get("collect_labels_as_tags")
self.kube_pod_tags = {}
self.use_histogram = _is_affirmative(instance.get('use_histogram', False))
performance_tags = instance.get("performance_tags", DEFAULT_PERFORMANCE_TAGS)
self.tag_names = {
CONTAINER: instance.get("container_tags", DEFAULT_CONTAINER_TAGS),
PERFORMANCE: performance_tags,
IMAGE: instance.get('image_tags', DEFAULT_IMAGE_TAGS)
}
# Set filtering settings
if self.docker_util.filtering_enabled:
self.tag_names[FILTERED] = self.docker_util.filtered_tag_names
# Container network mapping cache
self.network_mappings = {}
# get the health check whitelist
self.whitelist_patterns = None
health_scs_whitelist = instance.get('health_service_check_whitelist', [])
if health_scs_whitelist:
patterns, whitelist_tags = compile_filter_rules(health_scs_whitelist)
self.whitelist_patterns = set(patterns)
self.tag_names[HEALTHCHECK] = set(whitelist_tags)
# Other options
self.collect_image_stats = _is_affirmative(instance.get('collect_images_stats', False))
self.collect_container_size = _is_affirmative(instance.get('collect_container_size', False))
self.collect_container_count = _is_affirmative(instance.get('collect_container_count', False))
self.collect_volume_count = _is_affirmative(instance.get('collect_volume_count', False))
self.collect_events = _is_affirmative(instance.get('collect_events', True))
self.event_attributes_as_tags = instance.get('event_attributes_as_tags', [])
self.collect_image_size = _is_affirmative(instance.get('collect_image_size', False))
self.collect_disk_stats = _is_affirmative(instance.get('collect_disk_stats', False))
self.collect_exit_codes = _is_affirmative(instance.get('collect_exit_codes', False))
self.collect_ecs_tags = _is_affirmative(instance.get('ecs_tags', True)) and Platform.is_ecs_instance()
self.filtered_event_types = tuple(instance.get("filtered_event_types", DEFAULT_FILTERED_EVENT_TYPES))
self.capped_metrics = instance.get('capped_metrics')
except Exception as e:
self.log.critical(e)
self.warning("Initialization failed. Will retry at next iteration")
else:
self.init_success = True
def check(self, instance):
"""Run the Docker check for one instance."""
if not self.init_success:
# Initialization can fail if cgroups are not ready or docker daemon is down. So we retry if needed
self.init()
try:
if self.docker_util.client is None:
message = "Unable to connect to Docker daemon"
self.service_check(SERVICE_CHECK_NAME, AgentCheck.CRITICAL,
message=message, tags=self.custom_tags)
return
except Exception as ex:
self.service_check(SERVICE_CHECK_NAME, AgentCheck.CRITICAL,
message=str(ex), tags=self.custom_tags)
return
if not self.init_success:
# Initialization failed, will try later
return
try:
# Report image metrics
if self.collect_image_stats:
self._count_and_weigh_images()
if Platform.is_k8s():
self.kube_pod_tags = {}
if self.kubeutil:
try:
self.kube_pod_tags = self.kubeutil.get_kube_pod_tags()
except Exception as e:
self.log.warning('Could not retrieve kubernetes labels: %s' % str(e))
# containers running with custom cgroups?
custom_cgroups = _is_affirmative(instance.get('custom_cgroups', False))
# Get the list of containers and the index of their names
health_service_checks = True if self.whitelist_patterns else False
containers_by_id = self._get_and_count_containers(custom_cgroups, health_service_checks)
containers_by_id = self._crawl_container_pids(containers_by_id, custom_cgroups)
# Send events from Docker API
if self.collect_events or self._service_discovery or not self._disable_net_metrics or self.collect_exit_codes:
self._process_events(containers_by_id)
# Report performance container metrics (cpu, mem, net, io)
self._report_performance_metrics(containers_by_id)
if self.collect_container_size:
self._report_container_size(containers_by_id)
if self.collect_container_count:
self._report_container_count(containers_by_id)
if self.collect_volume_count:
self._report_volume_count()
# Collect disk stats from Docker info command
if self.collect_disk_stats:
self._report_disk_stats()
if health_service_checks:
self._send_container_healthcheck_sc(containers_by_id)
except:
self.log.exception("Docker_daemon check failed")
self.warning("Check failed. Will retry at next iteration")
if self.capped_metrics:
self.filter_capped_metrics()
def _count_and_weigh_images(self):
try:
tags = self._get_tags()
active_images = self.docker_util.client.images(all=False)
active_images_len = len(active_images)
all_images_len = len(self.docker_util.client.images(quiet=True, all=True))
self.gauge("docker.images.available", active_images_len, tags=tags)
self.gauge("docker.images.intermediate", (all_images_len - active_images_len), tags=tags)
if self.collect_image_size:
self._report_image_size(active_images)
except Exception as e:
# It's not an important metric, keep going if it fails
self.warning("Failed to count Docker images. Exception: {0}".format(e))
def _get_and_count_containers(self, custom_cgroups=False, healthchecks=False):
"""List all the containers from the API, filter and count them."""
# Querying the size of containers is slow, we don't do it at each run
must_query_size = self.collect_container_size and self._latest_size_query == 0
self._latest_size_query = (self._latest_size_query + 1) % SIZE_REFRESH_RATE
running_containers_count = Counter()
all_containers_count = Counter()
try:
containers = self.docker_util.client.containers(all=True, size=must_query_size)
except Exception as e:
message = "Unable to list Docker containers: {0}".format(e)
self.service_check(SERVICE_CHECK_NAME, AgentCheck.CRITICAL,
message=message, tags=self.custom_tags)
raise Exception(message)
else:
self.service_check(SERVICE_CHECK_NAME, AgentCheck.OK, tags=self.custom_tags)
# Create a set of filtered containers based on the exclude/include rules
# and cache these rules in docker_util
self._filter_containers(containers)
containers_by_id = {}
for container in containers:
container_name = DockerUtil.container_name_extractor(container)[0]
container_status_tags = self._get_tags(container, CONTAINER)
all_containers_count[tuple(sorted(container_status_tags))] += 1
if self._is_container_running(container):
running_containers_count[tuple(sorted(container_status_tags))] += 1
# Check if the container is included/excluded via its tags
if self._is_container_excluded(container):
self.log.debug("Container {0} is excluded".format(container_name))
continue
containers_by_id[container['Id']] = container
# grab pid via API if custom cgroups - otherwise we won't find process when
# crawling for pids.
if custom_cgroups or healthchecks:
try:
inspect_dict = self.docker_util.client.inspect_container(container_name)
container['_pid'] = inspect_dict['State']['Pid']
container['health'] = inspect_dict['State'].get('Health', {})
except Exception as e:
self.log.debug("Unable to inspect Docker container: %s", e)
total_count = 0
# TODO: deprecate these 2, they should be replaced by _report_container_count
for tags, count in running_containers_count.iteritems():
total_count += count
self.gauge("docker.containers.running", count, tags=list(tags))
self.gauge("docker.containers.running.total", total_count, tags=self.custom_tags)
total_count = 0
for tags, count in all_containers_count.iteritems():
stopped_count = count - running_containers_count[tags]
total_count += stopped_count
self.gauge("docker.containers.stopped", stopped_count, tags=list(tags))
self.gauge("docker.containers.stopped.total", total_count, tags=self.custom_tags)
return containers_by_id
def _is_container_running(self, container):
"""Tell if a container is running, according to its status.
There is no "nice" API field to figure it out. We just look at the "Status" field, knowing how it is generated.
See: https://github.com/docker/docker/blob/v1.6.2/daemon/state.go#L35
"""
return container["Status"].startswith("Up") or container["Status"].startswith("Restarting")
def _get_tags(self, entity=None, tag_type=None):
"""Generate the tags for a given entity (container or image) according to a list of tag names."""
# Start with custom tags
tags = list(self.custom_tags)
# Collect pod names as tags on kubernetes
if Platform.is_k8s() and KubeUtil.POD_NAME_LABEL not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(KubeUtil.POD_NAME_LABEL)
self.collect_labels_as_tags.append(KubeUtil.CONTAINER_NAME_LABEL)
# Collect container names as tags on rancher
if Platform.is_rancher():
if RANCHER_CONTAINER_NAME not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(RANCHER_CONTAINER_NAME)
if RANCHER_SVC_NAME not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(RANCHER_SVC_NAME)
if RANCHER_STACK_NAME not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(RANCHER_STACK_NAME)
if entity is not None:
pod_name = None
namespace = None
# Get labels as tags
labels = entity.get("Labels")
if labels is not None:
for k in self.collect_labels_as_tags:
if k in labels:
v = labels[k]
if k == KubeUtil.POD_NAME_LABEL and Platform.is_k8s():
pod_name = v
k = "pod_name"
if "-" in pod_name:
replication_controller = "-".join(pod_name.split("-")[:-1])
if "/" in replication_controller: # k8s <= 1.1
namespace, replication_controller = replication_controller.split("/", 1)
elif KubeUtil.NAMESPACE_LABEL in labels: # k8s >= 1.2
namespace = labels[KubeUtil.NAMESPACE_LABEL]
tags.append("kube_namespace:%s" % namespace)
tags.append("kube_replication_controller:%s" % replication_controller)
tags.append("pod_name:%s" % pod_name)
elif k == KubeUtil.CONTAINER_NAME_LABEL and Platform.is_k8s():
if v:
tags.append("kube_container_name:%s" % v)
elif k == SWARM_SVC_LABEL and Platform.is_swarm():
if v:
tags.append("swarm_service:%s" % v)
elif k == RANCHER_CONTAINER_NAME and Platform.is_rancher():
if v:
tags.append('rancher_container:%s' % v)
elif k == RANCHER_SVC_NAME and Platform.is_rancher():
if v:
tags.append('rancher_service:%s' % v)
elif k == RANCHER_STACK_NAME and Platform.is_rancher():
if v:
tags.append('rancher_stack:%s' % v)
elif not v:
tags.append(k)
else:
tags.append("%s:%s" % (k, v))
if k == KubeUtil.POD_NAME_LABEL and Platform.is_k8s() and k not in labels:
tags.append("pod_name:no_pod")
# Get entity specific tags
if tag_type is not None:
tag_names = self.tag_names[tag_type]
for tag_name in tag_names:
tag_value = self._extract_tag_value(entity, tag_name)
if tag_value is not None:
for t in tag_value:
tags.append('%s:%s' % (tag_name, str(t).strip()))
# Add kube labels and creator/service tags
if Platform.is_k8s() and namespace and pod_name:
kube_tags = self.kube_pod_tags.get("{0}/{1}".format(namespace, pod_name))
if kube_tags:
tags.extend(list(kube_tags))
if self.metadata_collector.has_detected():
orch_tags = self.metadata_collector.get_container_tags(co=entity)
tags.extend(orch_tags)
return tags
def _extract_tag_value(self, entity, tag_name):
"""Extra tag information from the API result (containers or images).
Cache extracted tags inside the entity object.
"""
if tag_name not in TAG_EXTRACTORS:
self.warning("{0} isn't a supported tag".format(tag_name))
return
# Check for already extracted tags
if "_tag_values" not in entity:
entity["_tag_values"] = {}
if tag_name not in entity["_tag_values"]:
entity["_tag_values"][tag_name] = TAG_EXTRACTORS[tag_name](entity)
return entity["_tag_values"][tag_name]
def _filter_containers(self, containers):
if not self.docker_util.filtering_enabled:
return
self._filtered_containers = set()
for container in containers:
container_tags = self._get_tags(container, FILTERED)
# exclude/include patterns are stored in docker_util to share them with other container-related checks
if self.docker_util.are_tags_filtered(container_tags):
container_name = DockerUtil.container_name_extractor(container)[0]
self._filtered_containers.add(container_name)
self.log.debug("Container {0} is filtered".format(container_name))
def _is_container_excluded(self, container):
"""Check if a container is excluded according to the filter rules.
Requires _filter_containers to run first.
"""
container_name = DockerUtil.container_name_extractor(container)[0]
return container_name in self._filtered_containers
def _report_container_size(self, containers_by_id):
for container in containers_by_id.itervalues():
if self._is_container_excluded(container):
continue
tags = self._get_tags(container, PERFORMANCE)
m_func = FUNC_MAP[GAUGE][self.use_histogram]
if "SizeRw" in container:
m_func(self, 'docker.container.size_rw', container['SizeRw'],
tags=tags)
if "SizeRootFs" in container:
m_func(
self, 'docker.container.size_rootfs', container['SizeRootFs'],
tags=tags)
def _send_container_healthcheck_sc(self, containers_by_id):
"""Send health service checks for containers."""
for container in containers_by_id.itervalues():
healthcheck_tags = self._get_tags(container, HEALTHCHECK)
match = False
for tag in healthcheck_tags:
for rule in self.whitelist_patterns:
if re.match(rule, tag):
match = True
self._submit_healthcheck_sc(container)
break
if match:
break
def _submit_healthcheck_sc(self, container):
health = container.get('health', {})
status = AgentCheck.UNKNOWN
if health:
_health = health.get('Status', '')
if _health == 'unhealthy':
status = AgentCheck.CRITICAL
elif _health == 'healthy':
status = AgentCheck.OK
tags = self._get_tags(container, CONTAINER)
self.service_check(HEALTHCHECK_SERVICE_CHECK_NAME, status, tags=tags)
def _report_container_count(self, containers_by_id):
"""Report container count per state"""
m_func = FUNC_MAP[GAUGE][self.use_histogram]
per_state_count = defaultdict(int)
filterlambda = lambda ctr: not self._is_container_excluded(ctr)
containers = list(filter(filterlambda, containers_by_id.values()))
for ctr in containers:
per_state_count[ctr.get('State', '')] += 1
for state in per_state_count:
if state:
m_func(self, 'docker.container.count', per_state_count[state], tags=['container_state:%s' % state.lower()])
def _report_volume_count(self):
"""Report volume count per state (dangling or not)"""
m_func = FUNC_MAP[GAUGE][self.use_histogram]
attached_volumes = self.docker_util.client.volumes(filters={'dangling': False})
dangling_volumes = self.docker_util.client.volumes(filters={'dangling': True})
attached_count = len(attached_volumes.get('Volumes', []) or [])
dangling_count = len(dangling_volumes.get('Volumes', []) or [])
m_func(self, 'docker.volume.count', attached_count, tags=['volume_state:attached'])
m_func(self, 'docker.volume.count', dangling_count, tags=['volume_state:dangling'])
def _report_image_size(self, images):
for image in images:
tags = self._get_tags(image, IMAGE)
if 'VirtualSize' in image:
self.gauge('docker.image.virtual_size', image['VirtualSize'], tags=tags)
if 'Size' in image:
self.gauge('docker.image.size', image['Size'], tags=tags)
# Performance metrics
def _report_performance_metrics(self, containers_by_id):
containers_without_proc_root = []
for container_id, container in containers_by_id.iteritems():
if self._is_container_excluded(container) or not self._is_container_running(container):
continue
tags = self._get_tags(container, PERFORMANCE)
try:
self._report_cgroup_metrics(container, tags)
if "_proc_root" not in container:
containers_without_proc_root.append(DockerUtil.container_name_extractor(container)[0])
continue
self._report_net_metrics(container, tags)
except BogusPIDException as e:
self.log.warning('Unable to report cgroup metrics for container %s: %s', container_id[:12], e)
if containers_without_proc_root:
message = "Couldn't find pid directory for containers: {0}. They'll be missing network metrics".format(
", ".join(containers_without_proc_root))
if not Platform.is_k8s():
self.warning(message)
else:
# On kubernetes, this is kind of expected. Network metrics will be collected by the kubernetes integration anyway
self.log.debug(message)
def _report_cgroup_metrics(self, container, tags):
cgroup_stat_file_failures = 0
if not container.get('_pid'):
raise BogusPIDException('Cannot report on bogus pid(0)')
for cgroup in CGROUP_METRICS:
try:
stat_file = self._get_cgroup_from_proc(cgroup["cgroup"], container['_pid'], cgroup['file'])
except MountException as e:
# We can't find a stat file
self.warning(str(e))
cgroup_stat_file_failures += 1
if cgroup_stat_file_failures >= len(CGROUP_METRICS):
self.warning("Couldn't find the cgroup files. Skipping the CGROUP_METRICS for now.")
except IOError as e:
self.log.debug("Cannot read cgroup file, container likely raced to finish : %s", e)
else:
stats = self._parse_cgroup_file(stat_file)
if stats:
for key, (dd_key, metric_func) in cgroup['metrics'].iteritems():
metric_func = FUNC_MAP[metric_func][self.use_histogram]
if key in stats:
metric_func(self, dd_key, int(stats[key]), tags=tags)
# Computed metrics
for mname, (key_list, fct, metric_func) in cgroup.get('to_compute', {}).iteritems():
values = [stats[key] for key in key_list if key in stats]
if len(values) != len(key_list):
self.log.debug("Couldn't compute {0}, some keys were missing.".format(mname))
continue
value = fct(*values)
metric_func = FUNC_MAP[metric_func][self.use_histogram]
if value is not None:
metric_func(self, mname, value, tags=tags)
def _report_net_metrics(self, container, tags):
"""Find container network metrics by looking at /proc/$PID/net/dev of the container process."""
if self._disable_net_metrics:
self.log.debug("Network metrics are disabled. Skipping")
return
proc_net_file = os.path.join(container['_proc_root'], 'net/dev')
try:
if container['Id'] in self.network_mappings:
networks = self.network_mappings[container['Id']]
else:
networks = self.docker_util.get_container_network_mapping(container)
if not networks:
networks = {'eth0': 'bridge'}
self.network_mappings[container['Id']] = networks
except Exception as e:
# Revert to previous behaviour if the method is missing or failing
# Debug message will only appear once per container, then the cache is used
self.log.debug("Failed to build docker network mapping, using failsafe. Exception: {0}".format(e))
networks = {'eth0': 'bridge'}
self.network_mappings[container['Id']] = networks
try:
with open(proc_net_file, 'r') as fp:
lines = fp.readlines()
"""Two first lines are headers:
Inter-| Receive | Transmit
face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed
"""
for l in lines[2:]:
cols = l.split(':', 1)
interface_name = str(cols[0]).strip()
if interface_name in networks:
net_tags = tags + ['docker_network:'+networks[interface_name]]
x = cols[1].split()
m_func = FUNC_MAP[RATE][self.use_histogram]
m_func(self, "docker.net.bytes_rcvd", long(x[0]), net_tags)
m_func(self, "docker.net.bytes_sent", long(x[8]), net_tags)
except Exception as e:
# It is possible that the container got stopped between the API call and now
self.warning("Failed to report IO metrics from file {0}. Exception: {1}".format(proc_net_file, e))
def _invalidate_network_mapping_cache(self, api_events):
for ev in api_events:
try:
if ev.get('Type') == 'network' and ev.get('Action').endswith('connect'):
container_id = ev.get('Actor').get('Attributes').get('container')
if container_id in self.network_mappings:
self.log.debug("Removing network mapping cache for container %s" % container_id)
del self.network_mappings[container_id]
except Exception:
self.log.warning('Malformed network event: %s' % str(ev))
def _process_events(self, containers_by_id):
api_events = self._get_events()
if self.collect_exit_codes:
self._report_exit_codes(api_events, containers_by_id)
if self.collect_events:
try:
aggregated_events = self._pre_aggregate_events(api_events, containers_by_id)
events = self._format_events(aggregated_events, containers_by_id)
except (socket.timeout, urllib2.URLError):
self.warning('Timeout when collecting events. Events will be missing.')
return
except Exception as e:
self.warning("Unexpected exception when collecting events: {0}. "
"Events will be missing".format(e))
return
for ev in events:
self.log.debug("Creating event: %s" % ev['msg_title'])
self.event(ev)
def _get_events(self):
"""Get the list of events."""
events, changed_container_ids = self.docker_util.get_events()
if not self._disable_net_metrics:
self._invalidate_network_mapping_cache(events)
if changed_container_ids and self._service_discovery:
get_sd_backend(self.agentConfig).update_checks(changed_container_ids)
if changed_container_ids:
self.metadata_collector.invalidate_cache(events)
return events
def _pre_aggregate_events(self, api_events, containers_by_id):
# Aggregate events, one per image. Put newer events first.
events = defaultdict(deque)
for event in api_events:
# Skip events related to filtered containers
container = containers_by_id.get(event.get('id'))
if container is not None and self._is_container_excluded(container):
self.log.debug("Excluded event: container {0} status changed to {1}".format(
event['id'], event['status']))
continue
# from may be missing (for network events for example)
if 'from' in event:
image_name = event['from']
if image_name.startswith('sha256:'):
image_name = self.docker_util.image_name_extractor({'Image': image_name})
events[image_name].appendleft(event)
return events
def _format_events(self, aggregated_events, containers_by_id):
events = []
for image_name, event_group in aggregated_events.iteritems():
container_tags = set()
filtered_events_count = 0
normal_prio_events = []
for event in event_group:
# Only keep events that are not configured to be filtered out
if event['status'].startswith(self.filtered_event_types):
filtered_events_count += 1
continue
container_name = event['id'][:11]
if event['id'] in containers_by_id:
cont = containers_by_id[event['id']]
container_name = DockerUtil.container_name_extractor(cont)[0]
container_tags.update(self._get_tags(cont, PERFORMANCE))
container_tags.add('container_name:%s' % container_name)
# Add additionnal docker event attributes as tag
for attr in self.event_attributes_as_tags:
if attr in event['Actor']['Attributes'] and attr not in EXCLUDED_ATTRIBUTES:
container_tags.add('%s:%s' % (attr, event['Actor']['Attributes'][attr]))
normal_prio_events.append((event, container_name))
if filtered_events_count:
self.log.debug('%d events were filtered out because of ignored event type' % filtered_events_count)
normal_event = self._create_dd_event(normal_prio_events, image_name, container_tags, priority='Normal')
if normal_event:
events.append(normal_event)
return events
def _report_exit_codes(self, events, containers_by_id):
for event in events:
container_tags = set()
container = containers_by_id.get(event.get('id'))
# Skip events related to filtered containers
if container is not None and self._is_container_excluded(container):
continue
# Report the exit code in case of a DIE event
if container is not None and event['status'] == 'die':
container_name = DockerUtil.container_name_extractor(container)[0]
container_tags.update(self._get_tags(container, CONTAINER))
container_tags.add('container_name:%s' % container_name)
try:
exit_code = int(event['Actor']['Attributes']['exitCode'])
message = 'Container %s exited with %s' % (container_name, exit_code)
status = AgentCheck.OK if exit_code == 0 else AgentCheck.CRITICAL
self.service_check(EXIT_SERVICE_CHECK_NAME, status, tags=list(container_tags), message=message)
except KeyError:
self.log.warning('Unable to collect the exit code for container %s' % container_name)
def _create_dd_event(self, events, image, c_tags, priority='Normal'):
"""Create the actual event to submit from a list of similar docker events"""
if not events:
return
max_timestamp = 0
status = defaultdict(int)
status_change = []
for ev, c_name in events:
max_timestamp = max(max_timestamp, int(ev['time']))
status[ev['status']] += 1
status_change.append([c_name, ev['status']])
status_text = ", ".join(["%d %s" % (count, st) for st, count in status.iteritems()])
msg_title = "%s %s on %s" % (image, status_text, self.hostname)
msg_body = (
"%%%\n"
"{image_name} {status} on {hostname}\n"
"```\n{status_changes}\n```\n"
"%%%"
).format(
image_name=image,
status=status_text,
hostname=self.hostname,
status_changes="\n".join(
["%s \t%s" % (change[1].upper(), change[0]) for change in status_change])
)
if any(error in status_text for error in ERROR_ALERT_TYPE):
alert_type = "error"
else:
alert_type = None
return {
'timestamp': max_timestamp,
'host': self.hostname,
'event_type': EVENT_TYPE,
'msg_title': msg_title,
'msg_text': msg_body,
'source_type_name': EVENT_TYPE,
'event_object': 'docker:%s' % image,
'tags': list(c_tags),
'alert_type': alert_type,
'priority': priority
}
def _report_disk_stats(self):
"""Report metrics about the volume space usage"""
stats = {
'docker.data.used': None,
'docker.data.total': None,
'docker.data.free': None,
'docker.metadata.used': None,
'docker.metadata.total': None,
'docker.metadata.free': None
# these two are calculated by _calc_percent_disk_stats
# 'docker.data.percent': None,
# 'docker.metadata.percent': None
}
info = self.docker_util.client.info()
driver_status = info.get('DriverStatus', [])
if not driver_status:
self.log.warning('Disk metrics collection is enabled but docker info did not'
' report any. Your storage driver might not support them, skipping.')
return
for metric in driver_status:
# only consider metrics about disk space
if len(metric) == 2 and 'Space' in metric[0]:
# identify Data and Metadata metrics
mtype = 'data'
if 'Metadata' in metric[0]:
mtype = 'metadata'
if 'Used' in metric[0]:
stats['docker.{0}.used'.format(mtype)] = metric[1]
elif 'Space Total' in metric[0]:
stats['docker.{0}.total'.format(mtype)] = metric[1]
elif 'Space Available' in metric[0]:
stats['docker.{0}.free'.format(mtype)] = metric[1]
stats = self._format_disk_metrics(stats)
stats.update(self._calc_percent_disk_stats(stats))
tags = self._get_tags()
for name, val in stats.iteritems():
if val is not None:
self.gauge(name, val, tags)
def _format_disk_metrics(self, metrics):
"""Cast the disk stats to float and convert them to bytes"""
for name, raw_val in metrics.iteritems():
if raw_val:
match = DISK_STATS_RE.search(raw_val)
if match is None or len(match.groups()) != 2:
self.log.warning('Can\'t parse value %s for disk metric %s. Dropping it.' % (raw_val, name))
metrics[name] = None
val, unit = match.groups()
# by default some are uppercased others lowercased. That's error prone.
unit = unit.lower()
try:
val = int(float(val) * UNIT_MAP[unit])
metrics[name] = val
except KeyError:
self.log.error('Unrecognized unit %s for disk metric %s. Dropping it.' % (unit, name))
metrics[name] = None
return metrics
def _calc_percent_disk_stats(self, stats):
"""Calculate a percentage of used disk space for data and metadata"""
mtypes = ['data', 'metadata']
percs = {}
for mtype in mtypes:
used = stats.get('docker.{0}.used'.format(mtype))
total = stats.get('docker.{0}.total'.format(mtype))
free = stats.get('docker.{0}.free'.format(mtype))
if used and total and free and ceil(total) < free + used:
self.log.debug('used, free, and total disk metrics may be wrong, '
'used: %s, free: %s, total: %s',
used, free, total)
total = used + free
try:
if isinstance(used, int):
percs['docker.{0}.percent'.format(mtype)] = round(100 * float(used) / float(total), 2)
elif isinstance(free, int):
percs['docker.{0}.percent'.format(mtype)] = round(100 * (1.0 - (float(free) / float(total))), 2)
except ZeroDivisionError:
self.log.error('docker.{0}.total is 0, calculating docker.{1}.percent'
' is not possible.'.format(mtype, mtype))
return percs
# Cgroups
def _get_cgroup_from_proc(self, cgroup, pid, filename):
"""Find a specific cgroup file, containing metrics to extract."""
params = {
"file": filename,
}
return DockerUtil.find_cgroup_from_proc(self._mountpoints, pid, cgroup, self.docker_util._docker_root) % (params)
def _parse_cgroup_file(self, stat_file):
"""Parse a cgroup pseudo file for key/values."""
self.log.debug("Opening cgroup file: %s" % stat_file)
try:
with open(stat_file, 'r') as fp:
if 'blkio' in stat_file:
return self._parse_blkio_metrics(fp.read().splitlines())
elif 'cpuacct.usage' in stat_file:
return dict({'usage': str(int(fp.read())/10000000)})
elif 'memory.soft_limit_in_bytes' in stat_file:
value = int(fp.read())
# do not report kernel max default value (uint64 * 4096)
# see https://github.com/torvalds/linux/blob/5b36577109be007a6ecf4b65b54cbc9118463c2b/mm/memcontrol.c#L2844-L2845
# 2 ** 60 is kept for consistency of other cgroups metrics
if value < 2 ** 60:
return dict({'softlimit': value})
else:
return dict(map(lambda x: x.split(' ', 1), fp.read().splitlines()))
except IOError:
# It is possible that the container got stopped between the API call and now.
# Some files can also be missing (like cpu.stat) and that's fine.
self.log.debug("Can't open %s. Its metrics will be missing." % stat_file)
def _parse_blkio_metrics(self, stats):
"""Parse the blkio metrics."""
metrics = {
'io_read': 0,
'io_write': 0,
}
for line in stats:
if 'Read' in line:
metrics['io_read'] += int(line.split()[2])
if 'Write' in line:
metrics['io_write'] += int(line.split()[2])
return metrics
def _is_container_cgroup(self, line, selinux_policy):
if line[1] not in ('cpu,cpuacct', 'cpuacct,cpu', 'cpuacct') or line[2] == '/docker-daemon':
return False
if 'docker' in line[2]: # general case
return True
if 'docker' in selinux_policy: # selinux
return True
if line[2].startswith('/') and re.match(CONTAINER_ID_RE, line[2][1:]): # kubernetes
return True
if line[2].startswith('/') and re.match(CONTAINER_ID_RE, line[2].split('/')[-1]): # kube 1.6+ qos hierarchy
return True
return False
# proc files
def _crawl_container_pids(self, container_dict, custom_cgroups=False):
"""Crawl `/proc` to find container PIDs and add them to `containers_by_id`."""
proc_path = os.path.join(self.docker_util._docker_root, 'proc')
pid_dirs = [_dir for _dir in os.listdir(proc_path) if _dir.isdigit()]
if len(pid_dirs) == 0:
self.warning("Unable to find any pid directory in {0}. "
"If you are running the agent in a container, make sure to "
'share the volume properly: "/proc:/host/proc:ro". '
"See https://github.com/serverdensity/docker-sd-agent/blob/master/README.md for more information. "
"Network metrics will be missing".format(proc_path))
self._disable_net_metrics = True
return container_dict
self._disable_net_metrics = False
for folder in pid_dirs:
try:
path = os.path.join(proc_path, folder, 'cgroup')
with open(path, 'r') as f:
content = [line.strip().split(':') for line in f.readlines()]
selinux_policy = ''
path = os.path.join(proc_path, folder, 'attr', 'current')
if os.path.exists(path):
with open(path, 'r') as f:
selinux_policy = f.readlines()[0]
except IOError, e:
# Issue #2074
self.log.debug("Cannot read %s, process likely raced to finish : %s", path, e)
except Exception as e:
self.warning("Cannot read %s : %s" % (path, str(e)))
continue
try:
for line in content:
if self._is_container_cgroup(line, selinux_policy):
cpuacct = line[2]
break
else:
continue
matches = re.findall(CONTAINER_ID_RE, cpuacct)
if matches:
container_id = matches[-1]
if container_id not in container_dict:
self.log.debug(
"Container %s not in container_dict, it's likely excluded", container_id
)
continue
container_dict[container_id]['_pid'] = folder
container_dict[container_id]['_proc_root'] = os.path.join(proc_path, folder)
elif custom_cgroups: # if we match by pid that should be enough (?) - O(n) ugh!
for _, container in container_dict.iteritems():
if container.get('_pid') == int(folder):
container['_proc_root'] = os.path.join(proc_path, folder)
break
except Exception, e:
self.warning("Cannot parse %s content: %s" % (path, str(e)))
continue
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.kubeutil = None
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
log.error(
"Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s"
% str(ex))
self.metadata_collector = MetadataCollector()
self.VAR_MAPPING = {
'host': self._get_host_address,
'pid': self._get_container_pid,
'port': self._get_port,
'container-name': self._get_container_name,
'tags': self._get_additional_tags,
}
AbstractSDBackend.__init__(self, agentConfig)
def _make_fetch_state(self):
pod_list = []
if Platform.is_k8s():
if not self.kubeutil or not self.kubeutil.init_success:
log.error(
"kubelet client not initialized, 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.dockerutil.client.inspect_container, pod_list)
def update_checks(self, changed_containers):
"""
Takes a list of container IDs that changed recently
and marks their corresponding checks as
"""
if not self.dockerutil.client:
log.warning(
"Docker client is not initialized, pausing auto discovery.")
return
state = self._make_fetch_state()
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 STACKSTATE_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
labels = inspect.get('Config', {}).get('Labels', {})
identifier = labels.get(STACKSTATE_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),
}
if labels:
platform_kwargs['docker_labels'] = labels
return self.config_store.get_checks_to_refresh(identifier,
**platform_kwargs)
def _get_container_pid(self, state, cid, tpl_var):
"""Extract the host-namespace pid of the container pid 0"""
pid = state.inspect_container(cid).get('State', {}).get('Pid')
if not pid:
return None
return str(pid)
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_inspect.get('Id', '')
c_img = self.dockerutil.image_name_extractor(c_inspect)
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():
if not self.kubeutil.init_success:
log.warning(
"kubelet client not initialized, kubernetes tags will be missing."
)
return tags
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 will be missing." % c_id[:12])
return tags
# 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)
if not self.kubeutil:
log.warning("The agent can't connect to kubelet, creator and "
"service tags will be missing for container %s." %
c_id[:12])
else:
# add creator tags
creator_tags = self.kubeutil.get_pod_creator_tags(pod_metadata)
tags.extend(creator_tags)
# add services tags
if self.kubeutil.collect_service_tag:
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)
if self.metadata_collector.has_detected():
orch_tags = self.metadata_collector.get_container_tags(
co=c_inspect)
tags.extend(orch_tags)
return tags
def _get_container_name(self, state, c_id, tpl_var):
container_inspect = state.inspect_container(c_id)
return container_inspect.get('Name', '').lstrip('/')
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'))
c_inspect = state.inspect_container(c_id)
c_name = c_inspect.get('Config', {}).get('Labels', {}).get(
KubeUtil.CONTAINER_NAME_LABEL)
if c_name:
tags.append('kube_container_name:%s' % c_name)
return tags
def get_configs(self):
"""Get the config for all docker containers running on the host."""
configs = {}
if not self.dockerutil.client:
log.warning(
"Docker client is not initialized, pausing auto discovery.")
return configs
state = self._make_fetch_state()
containers = [(self.dockerutil.image_name_extractor(container),
container.get('Id'), container.get('Labels'))
for container in self.dockerutil.client.containers()]
for image, cid, labels in containers:
try:
# value of the STACKSTATE_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, labels) 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:
if isinstance(instance, list):
configs[check_name] = (source, (init_config,
instance))
else:
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))
if isinstance(instance, list):
for inst in instance:
configs[check_name][1][1].append(inst)
else:
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 STACKSTATE_ID label, return its value or the image name if missing"""
return labels.get(STACKSTATE_ID) or image
def _get_check_configs(self, state, c_id, identifier, labels=None):
"""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'),
}
if labels:
platform_kwargs['docker_labels'] = labels
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
# covering mono-instance and multi-instances cases
tmpl_array = instance_tpl
if not isinstance(instance_tpl, list):
tmpl_array = [instance_tpl]
# insert tags in instance_tpl and process values for template variables
result_instances = []
result_init_config = None
for inst_tmpl in tmpl_array:
instance_tpl, var_values = self._fill_tpl(
state, c_id, inst_tmpl, 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
result_instances.append(instance)
if not result_init_config:
result_init_config = init_config
elif result_init_config != init_config:
self.log.warning(
"Different versions of `init_config` found for "
"check {}. Keeping the first one found.".format(
'check_name'))
check_configs.append(
(source, (check_name, result_init_config, result_instances)))
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) and not isinstance(
instance_tpl, list):
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, c_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 only default c_tags to the instance to avoid duplicate tags from conf
if c_tags:
tags = c_tags[:] # shallow copy of the c_tags array
else:
tags = []
if tags:
tpl_tags = instance_tpl.get('tags', [])
if isinstance(tpl_tags, dict):
for key, val in tpl_tags.iteritems():
tags.append("{}:{}".format(key, val))
else:
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
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(
'/'))
if self.agentConfig['collect_orchestrator_tags']:
host_container_metadata = MetadataCollector(
).get_host_metadata()
if host_container_metadata:
payload['container-meta'] = host_container_metadata
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 self.agentConfig['collect_orchestrator_tags']:
host_docker_tags = MetadataCollector().get_host_tags()
if host_docker_tags:
host_tags.extend(host_docker_tags)
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
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.kubeutil = None
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
self.metadata_collector = MetadataCollector()
self.VAR_MAPPING = {
'host': self._get_host_address,
'pid': self._get_container_pid,
'port': self._get_port,
'container-name': self._get_container_name,
'tags': self._get_additional_tags,
}
# docker labels we'll add as tags to all instances SD configures
self.docker_labels_as_tags = agentConfig.get('docker_labels_as_tags', '')
if self.docker_labels_as_tags:
self.docker_labels_as_tags = [label.strip() for label in self.docker_labels_as_tags.split(',')]
else:
self.docker_labels_as_tags = []
AbstractSDBackend.__init__(self, agentConfig)
def _make_fetch_state(self):
pod_list = []
if Platform.is_k8s():
if not self.kubeutil or not self.kubeutil.init_success:
log.error("kubelet client not initialized, 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.dockerutil.client.inspect_container, pod_list)
def update_checks(self, changed_containers):
"""
Takes a list of container IDs that changed recently
and marks their corresponding checks as
"""
if not self.dockerutil.client:
log.warning("Docker client is not initialized, pausing auto discovery.")
return
state = self._make_fetch_state()
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 SD_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
labels = inspect.get('Config', {}).get('Labels', {})
identifier = labels.get(SD_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),
}
if labels:
platform_kwargs['docker_labels'] = labels
return self.config_store.get_checks_to_refresh(identifier, **platform_kwargs)
def _get_container_pid(self, state, cid, tpl_var):
"""Extract the host-namespace pid of the container pid 0"""
pid = state.inspect_container(cid).get('State', {}).get('Pid')
if not pid:
return None
return str(pid)
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_inspect.get('Id', '')
c_img = self.dockerutil.image_name_extractor(c_inspect)
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
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)
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)
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']
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)
ports = []
try:
ports = [x.split('/')[0] for x in container_inspect['NetworkSettings']['Ports'].keys()]
if len(ports) == 0:
raise IndexError
except (IndexError, KeyError, AttributeError):
if Platform.is_k8s():
spec = state.get_kube_container_spec(c_id)
if spec:
ports = [str(x.get('containerPort')) for x in spec.get('ports', [])]
else:
ports = [p.split('/')[0] for p in container_inspect['Config'].get('ExposedPorts', {}).keys()]
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, self.docker_labels_as_tags)
if Platform.is_k8s():
if not self.kubeutil.init_success:
log.warning("kubelet client not initialized, kubernetes tags will be missing.")
return tags
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 will be missing." % c_id[:12])
return tags
# 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)
# get kubernetes container name
kube_container_name = state.get_kube_container_name(c_id)
if kube_container_name:
tags.append('kube_container_name:%s' % kube_container_name)
if not self.kubeutil:
log.warning("The agent can't connect to kubelet, creator and "
"service tags will be missing for container %s." % c_id[:12])
else:
# add creator tags
creator_tags = self.kubeutil.get_pod_creator_tags(pod_metadata)
tags.extend(creator_tags)
# add services tags
if self.kubeutil.collect_service_tag:
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)
if self.metadata_collector.has_detected():
orch_tags = self.metadata_collector.get_container_tags(co=c_inspect)
tags.extend(orch_tags)
return tags
def _get_container_name(self, state, c_id, tpl_var):
container_inspect = state.inspect_container(c_id)
return container_inspect.get('Name', '').lstrip('/')
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'))
c_inspect = state.inspect_container(c_id)
c_name = c_inspect.get('Config', {}).get('Labels', {}).get(KubeUtil.CONTAINER_NAME_LABEL)
if c_name:
tags.append('kube_container_name:%s' % c_name)
return tags
def get_configs(self):
"""Get the config for all docker containers running on the host."""
configs = {}
if not self.dockerutil.client:
log.warning("Docker client is not initialized, pausing auto discovery.")
return configs
state = self._make_fetch_state()
containers = [(
self.dockerutil.image_name_extractor(container),
container.get('Id'), container.get('Labels')
) for container in self.dockerutil.client.containers()]
for image, cid, labels in containers:
try:
# value of the SD_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, labels) 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:
if isinstance(instance, list):
configs[check_name] = (source, (init_config, instance))
else:
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))
if isinstance(instance, list):
for inst in instance:
configs[check_name][1][1].append(inst)
else:
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 SD_ID label, return its value or the image name if missing"""
return labels.get(SD_ID) or image
def _get_check_configs(self, state, c_id, identifier, labels=None):
"""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'),
}
if labels:
platform_kwargs['docker_labels'] = labels
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
# covering mono-instance and multi-instances cases
tmpl_array = instance_tpl
if not isinstance(instance_tpl, list):
tmpl_array = [instance_tpl]
# insert tags in instance_tpl and process values for template variables
result_instances = []
result_init_config = None
for inst_tmpl in tmpl_array:
instance_tpl, var_values = self._fill_tpl(state, c_id, inst_tmpl, 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
result_instances.append(instance)
if not result_init_config:
result_init_config = init_config
elif result_init_config != init_config:
log.warning("Different versions of `init_config` found for "
"check {}. Keeping the first one found.".format('check_name'))
check_configs.append((source, (check_name, result_init_config, result_instances)))
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 = []
auto_conf = not bool(config_backend)
# 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 = [var.strip('%') for var in variables]
if not isinstance(init_config_tpl, dict):
init_config_tpl = json.loads(init_config_tpl or '{}')
if not isinstance(instance_tpl, dict) and not isinstance(instance_tpl, list):
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, c_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 only default c_tags to the instance to avoid duplicate tags from conf
if c_tags:
tags = c_tags[:] # shallow copy of the c_tags array
else:
tags = []
if tags:
tpl_tags = instance_tpl.get('tags', [])
if isinstance(tpl_tags, dict):
for key, val in tpl_tags.iteritems():
tags.append("{}:{}".format(key, val))
else:
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
def init(self):
try:
instance = self.instances[0]
self.docker_util = DockerUtil()
self.docker_client = self.docker_util.client
self.docker_gateway = DockerUtil.get_gateway()
self.metadata_collector = MetadataCollector()
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.kubeutil = None
self.log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
# We configure the check with the right cgroup settings for this host
# Just needs to be done once
self._mountpoints = self.docker_util.get_mountpoints(CGROUP_METRICS)
self._latest_size_query = 0
self._filtered_containers = set()
self._disable_net_metrics = False
# Set tagging options
self.custom_tags = instance.get("tags", [])
self.collect_labels_as_tags = instance.get("collect_labels_as_tags", DEFAULT_LABELS_AS_TAGS)
self.kube_pod_tags = {}
self.use_histogram = _is_affirmative(instance.get('use_histogram', False))
performance_tags = instance.get("performance_tags", DEFAULT_PERFORMANCE_TAGS)
self.tag_names = {
CONTAINER: instance.get("container_tags", DEFAULT_CONTAINER_TAGS),
PERFORMANCE: performance_tags,
IMAGE: instance.get('image_tags', DEFAULT_IMAGE_TAGS)
}
# Set filtering settings
if self.docker_util.filtering_enabled:
self.tag_names[FILTERED] = self.docker_util.filtered_tag_names
# Container network mapping cache
self.network_mappings = {}
# get the health check whitelist
self.whitelist_patterns = None
health_scs_whitelist = instance.get('health_service_check_whitelist', [])
if health_scs_whitelist:
patterns, whitelist_tags = compile_filter_rules(health_scs_whitelist)
self.whitelist_patterns = set(patterns)
self.tag_names[HEALTHCHECK] = set(whitelist_tags)
# Other options
self.collect_image_stats = _is_affirmative(instance.get('collect_images_stats', False))
self.collect_container_size = _is_affirmative(instance.get('collect_container_size', False))
self.collect_container_count = _is_affirmative(instance.get('collect_container_count', False))
self.collect_volume_count = _is_affirmative(instance.get('collect_volume_count', False))
self.collect_events = _is_affirmative(instance.get('collect_events', True))
self.event_attributes_as_tags = instance.get('event_attributes_as_tags', [])
self.collect_image_size = _is_affirmative(instance.get('collect_image_size', False))
self.collect_disk_stats = _is_affirmative(instance.get('collect_disk_stats', False))
self.collect_exit_codes = _is_affirmative(instance.get('collect_exit_codes', False))
self.collect_ecs_tags = _is_affirmative(instance.get('ecs_tags', True)) and Platform.is_ecs_instance()
self.capped_metrics = instance.get('capped_metrics')
except Exception as e:
self.log.critical(e)
self.warning("Initialization failed. Will retry at next iteration")
else:
self.init_success = True
class DockerDaemon(AgentCheck):
"""Collect metrics and events from Docker API and cgroups."""
def __init__(self, name, init_config, agentConfig, instances=None):
if instances is not None and len(instances) > 1:
raise Exception("Docker check only supports one configured instance.")
AgentCheck.__init__(self, name, init_config,
agentConfig, instances=instances)
self.init_success = False
self.docker_client = None
self._service_discovery = agentConfig.get('service_discovery') and \
agentConfig.get('service_discovery_backend') == 'docker'
self.init()
def init(self):
try:
instance = self.instances[0]
self.docker_util = DockerUtil()
self.docker_client = self.docker_util.client
self.docker_gateway = DockerUtil.get_gateway()
self.metadata_collector = MetadataCollector()
if Platform.is_k8s():
try:
self.kubeutil = KubeUtil()
except Exception as ex:
self.kubeutil = None
self.log.error("Couldn't instantiate the kubernetes client, "
"subsequent kubernetes calls will fail as well. Error: %s" % str(ex))
# We configure the check with the right cgroup settings for this host
# Just needs to be done once
self._mountpoints = self.docker_util.get_mountpoints(CGROUP_METRICS)
self._latest_size_query = 0
self._filtered_containers = set()
self._disable_net_metrics = False
# Set tagging options
self.custom_tags = instance.get("tags", [])
self.collect_labels_as_tags = instance.get("collect_labels_as_tags", DEFAULT_LABELS_AS_TAGS)
self.kube_pod_tags = {}
self.use_histogram = _is_affirmative(instance.get('use_histogram', False))
performance_tags = instance.get("performance_tags", DEFAULT_PERFORMANCE_TAGS)
self.tag_names = {
CONTAINER: instance.get("container_tags", DEFAULT_CONTAINER_TAGS),
PERFORMANCE: performance_tags,
IMAGE: instance.get('image_tags', DEFAULT_IMAGE_TAGS)
}
# Set filtering settings
if self.docker_util.filtering_enabled:
self.tag_names[FILTERED] = self.docker_util.filtered_tag_names
# Container network mapping cache
self.network_mappings = {}
# get the health check whitelist
self.whitelist_patterns = None
health_scs_whitelist = instance.get('health_service_check_whitelist', [])
if health_scs_whitelist:
patterns, whitelist_tags = compile_filter_rules(health_scs_whitelist)
self.whitelist_patterns = set(patterns)
self.tag_names[HEALTHCHECK] = set(whitelist_tags)
# Other options
self.collect_image_stats = _is_affirmative(instance.get('collect_images_stats', False))
self.collect_container_size = _is_affirmative(instance.get('collect_container_size', False))
self.collect_container_count = _is_affirmative(instance.get('collect_container_count', False))
self.collect_volume_count = _is_affirmative(instance.get('collect_volume_count', False))
self.collect_events = _is_affirmative(instance.get('collect_events', True))
self.event_attributes_as_tags = instance.get('event_attributes_as_tags', [])
self.collect_image_size = _is_affirmative(instance.get('collect_image_size', False))
self.collect_disk_stats = _is_affirmative(instance.get('collect_disk_stats', False))
self.collect_exit_codes = _is_affirmative(instance.get('collect_exit_codes', False))
self.collect_ecs_tags = _is_affirmative(instance.get('ecs_tags', True)) and Platform.is_ecs_instance()
self.capped_metrics = instance.get('capped_metrics')
except Exception as e:
self.log.critical(e)
self.warning("Initialization failed. Will retry at next iteration")
else:
self.init_success = True
def check(self, instance):
"""Run the Docker check for one instance."""
if not self.init_success:
# Initialization can fail if cgroups are not ready or docker daemon is down. So we retry if needed
# https://github.com/DataDog/dd-agent/issues/1896
self.init()
if self.docker_client is None:
message = "Unable to connect to Docker daemon"
self.service_check(SERVICE_CHECK_NAME, AgentCheck.CRITICAL,
message=message)
return
if not self.init_success:
# Initialization failed, will try later
return
try:
# Report image metrics
if self.collect_image_stats:
self._count_and_weigh_images()
if Platform.is_k8s():
self.kube_pod_tags = {}
if self.kubeutil:
try:
self.kube_pod_tags = self.kubeutil.get_kube_pod_tags()
except Exception as e:
self.log.warning('Could not retrieve kubernetes labels: %s' % str(e))
# containers running with custom cgroups?
custom_cgroups = _is_affirmative(instance.get('custom_cgroups', False))
# Get the list of containers and the index of their names
health_service_checks = True if self.whitelist_patterns else False
containers_by_id = self._get_and_count_containers(custom_cgroups, health_service_checks)
containers_by_id = self._crawl_container_pids(containers_by_id, custom_cgroups)
# Send events from Docker API
if self.collect_events or self._service_discovery or not self._disable_net_metrics or self.collect_exit_codes:
self._process_events(containers_by_id)
# Report performance container metrics (cpu, mem, net, io)
self._report_performance_metrics(containers_by_id)
if self.collect_container_size:
self._report_container_size(containers_by_id)
if self.collect_container_count:
self._report_container_count(containers_by_id)
if self.collect_volume_count:
self._report_volume_count()
# Collect disk stats from Docker info command
if self.collect_disk_stats:
self._report_disk_stats()
if health_service_checks:
self._send_container_healthcheck_sc(containers_by_id)
except:
self.log.exception("Docker_daemon check failed")
self.warning("Check failed. Will retry at next iteration")
if self.capped_metrics:
self.filter_capped_metrics()
def _count_and_weigh_images(self):
try:
tags = self._get_tags()
active_images = self.docker_client.images(all=False)
active_images_len = len(active_images)
all_images_len = len(self.docker_client.images(quiet=True, all=True))
self.gauge("docker.images.available", active_images_len, tags=tags)
self.gauge("docker.images.intermediate", (all_images_len - active_images_len), tags=tags)
if self.collect_image_size:
self._report_image_size(active_images)
except Exception as e:
# It's not an important metric, keep going if it fails
self.warning("Failed to count Docker images. Exception: {0}".format(e))
def _get_and_count_containers(self, custom_cgroups=False, healthchecks=False):
"""List all the containers from the API, filter and count them."""
# Querying the size of containers is slow, we don't do it at each run
must_query_size = self.collect_container_size and self._latest_size_query == 0
self._latest_size_query = (self._latest_size_query + 1) % SIZE_REFRESH_RATE
running_containers_count = Counter()
all_containers_count = Counter()
try:
containers = self.docker_client.containers(all=True, size=must_query_size)
except Exception as e:
message = "Unable to list Docker containers: {0}".format(e)
self.service_check(SERVICE_CHECK_NAME, AgentCheck.CRITICAL,
message=message)
raise Exception(message)
else:
self.service_check(SERVICE_CHECK_NAME, AgentCheck.OK)
# Create a set of filtered containers based on the exclude/include rules
# and cache these rules in docker_util
self._filter_containers(containers)
containers_by_id = {}
for container in containers:
container_name = DockerUtil.container_name_extractor(container)[0]
container_status_tags = self._get_tags(container, CONTAINER)
all_containers_count[tuple(sorted(container_status_tags))] += 1
if self._is_container_running(container):
running_containers_count[tuple(sorted(container_status_tags))] += 1
# Check if the container is included/excluded via its tags
if self._is_container_excluded(container):
self.log.debug("Container {0} is excluded".format(container_name))
continue
containers_by_id[container['Id']] = container
# grab pid via API if custom cgroups - otherwise we won't find process when
# crawling for pids.
if custom_cgroups or healthchecks:
try:
inspect_dict = self.docker_client.inspect_container(container_name)
container['_pid'] = inspect_dict['State']['Pid']
container['health'] = inspect_dict['State'].get('Health', {})
except Exception as e:
self.log.debug("Unable to inspect Docker container: %s", e)
# TODO: deprecate these 2, they should be replaced by _report_container_count
for tags, count in running_containers_count.iteritems():
self.gauge("docker.containers.running", count, tags=list(tags))
for tags, count in all_containers_count.iteritems():
stopped_count = count - running_containers_count[tags]
self.gauge("docker.containers.stopped", stopped_count, tags=list(tags))
return containers_by_id
def _is_container_running(self, container):
"""Tell if a container is running, according to its status.
There is no "nice" API field to figure it out. We just look at the "Status" field, knowing how it is generated.
See: https://github.com/docker/docker/blob/v1.6.2/daemon/state.go#L35
"""
return container["Status"].startswith("Up") or container["Status"].startswith("Restarting")
def _get_tags(self, entity=None, tag_type=None):
"""Generate the tags for a given entity (container or image) according to a list of tag names."""
# Start with custom tags
tags = list(self.custom_tags)
# Collect pod names as tags on kubernetes
if Platform.is_k8s() and KubeUtil.POD_NAME_LABEL not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(KubeUtil.POD_NAME_LABEL)
# Collect container names as tags on rancher
if Platform.is_rancher():
if RANCHER_CONTAINER_NAME not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(RANCHER_CONTAINER_NAME)
if RANCHER_SVC_NAME not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(RANCHER_SVC_NAME)
if RANCHER_STACK_NAME not in self.collect_labels_as_tags:
self.collect_labels_as_tags.append(RANCHER_STACK_NAME)
if entity is not None:
pod_name = None
# Get labels as tags
labels = entity.get("Labels")
if labels is not None:
for k in self.collect_labels_as_tags:
if k in labels:
v = labels[k]
if k == KubeUtil.POD_NAME_LABEL and Platform.is_k8s():
pod_name = v
k = "pod_name"
if "-" in pod_name:
replication_controller = "-".join(pod_name.split("-")[:-1])
if "/" in replication_controller: # k8s <= 1.1
namespace, replication_controller = replication_controller.split("/", 1)
elif KubeUtil.NAMESPACE_LABEL in labels: # k8s >= 1.2
namespace = labels[KubeUtil.NAMESPACE_LABEL]
pod_name = "{0}/{1}".format(namespace, pod_name)
tags.append("kube_namespace:%s" % namespace)
tags.append("kube_replication_controller:%s" % replication_controller)
tags.append("pod_name:%s" % pod_name)
elif k == SWARM_SVC_LABEL and Platform.is_swarm():
if v:
tags.append("swarm_service:%s" % v)
elif k == RANCHER_CONTAINER_NAME and Platform.is_rancher():
if v:
tags.append('rancher_container:%s' % v)
elif k == RANCHER_SVC_NAME and Platform.is_rancher():
if v:
tags.append('rancher_service:%s' % v)
elif k == RANCHER_STACK_NAME and Platform.is_rancher():
if v:
tags.append('rancher_stack:%s' % v)
elif not v:
tags.append(k)
else:
tags.append("%s:%s" % (k,v))
if k == KubeUtil.POD_NAME_LABEL and Platform.is_k8s() and k not in labels:
tags.append("pod_name:no_pod")
# Get entity specific tags
if tag_type is not None:
tag_names = self.tag_names[tag_type]
for tag_name in tag_names:
tag_value = self._extract_tag_value(entity, tag_name)
if tag_value is not None:
for t in tag_value:
tags.append('%s:%s' % (tag_name, str(t).strip()))
# Add kube labels and creator/service tags
if Platform.is_k8s():
kube_tags = self.kube_pod_tags.get(pod_name)
if kube_tags:
tags.extend(list(kube_tags))
if self.metadata_collector.has_detected():
orch_tags = self.metadata_collector.get_container_tags(co=entity)
tags.extend(orch_tags)
return tags
def _extract_tag_value(self, entity, tag_name):
"""Extra tag information from the API result (containers or images).
Cache extracted tags inside the entity object.
"""
if tag_name not in TAG_EXTRACTORS:
self.warning("{0} isn't a supported tag".format(tag_name))
return
# Check for already extracted tags
if "_tag_values" not in entity:
entity["_tag_values"] = {}
if tag_name not in entity["_tag_values"]:
entity["_tag_values"][tag_name] = TAG_EXTRACTORS[tag_name](entity)
return entity["_tag_values"][tag_name]
def _filter_containers(self, containers):
if not self.docker_util.filtering_enabled:
return
self._filtered_containers = set()
for container in containers:
container_tags = self._get_tags(container, FILTERED)
# exclude/include patterns are stored in docker_util to share them with other container-related checks
if self.docker_util.are_tags_filtered(container_tags):
container_name = DockerUtil.container_name_extractor(container)[0]
self._filtered_containers.add(container_name)
self.log.debug("Container {0} is filtered".format(container_name))
def _is_container_excluded(self, container):
"""Check if a container is excluded according to the filter rules.
Requires _filter_containers to run first.
"""
container_name = DockerUtil.container_name_extractor(container)[0]
return container_name in self._filtered_containers
def _report_container_size(self, containers_by_id):
for container in containers_by_id.itervalues():
if self._is_container_excluded(container):
continue
tags = self._get_tags(container, PERFORMANCE)
m_func = FUNC_MAP[GAUGE][self.use_histogram]
if "SizeRw" in container:
m_func(self, 'docker.container.size_rw', container['SizeRw'],
tags=tags)
if "SizeRootFs" in container:
m_func(
self, 'docker.container.size_rootfs', container['SizeRootFs'],
tags=tags)
def _send_container_healthcheck_sc(self, containers_by_id):
"""Send health service checks for containers."""
for container in containers_by_id.itervalues():
healthcheck_tags = self._get_tags(container, HEALTHCHECK)
match = False
for tag in healthcheck_tags:
for rule in self.whitelist_patterns:
if re.match(rule, tag):
match = True
self._submit_healthcheck_sc(container)
break
if match:
break
def _submit_healthcheck_sc(self, container):
health = container.get('health', {})
status = AgentCheck.UNKNOWN
if health:
_health = health.get('Status', '')
if _health == 'unhealthy':
status = AgentCheck.CRITICAL
elif _health == 'healthy':
status = AgentCheck.OK
tags = self._get_tags(container, CONTAINER)
self.service_check(HEALTHCHECK_SERVICE_CHECK_NAME, status, tags=tags)
def _report_container_count(self, containers_by_id):
"""Report container count per state"""
m_func = FUNC_MAP[GAUGE][self.use_histogram]
per_state_count = defaultdict(int)
filterlambda = lambda ctr: not self._is_container_excluded(ctr)
containers = list(filter(filterlambda, containers_by_id.values()))
for ctr in containers:
per_state_count[ctr.get('State', '')] += 1
for state in per_state_count:
if state:
m_func(self, 'docker.container.count', per_state_count[state], tags=['container_state:%s' % state.lower()])
def _report_volume_count(self):
"""Report volume count per state (dangling or not)"""
m_func = FUNC_MAP[GAUGE][self.use_histogram]
attached_volumes = self.docker_client.volumes(filters={'dangling': False})
dangling_volumes = self.docker_client.volumes(filters={'dangling': True})
attached_count = len(attached_volumes['Volumes'])
dangling_count = len(dangling_volumes['Volumes'])
m_func(self, 'docker.volume.count', attached_count, tags=['volume_state:attached'])
m_func(self, 'docker.volume.count', dangling_count, tags=['volume_state:dangling'])
def _report_image_size(self, images):
for image in images:
tags = self._get_tags(image, IMAGE)
if 'VirtualSize' in image:
self.gauge('docker.image.virtual_size', image['VirtualSize'], tags=tags)
if 'Size' in image:
self.gauge('docker.image.size', image['Size'], tags=tags)
# Performance metrics
def _report_performance_metrics(self, containers_by_id):
containers_without_proc_root = []
for container in containers_by_id.itervalues():
if self._is_container_excluded(container) or not self._is_container_running(container):
continue
tags = self._get_tags(container, PERFORMANCE)
try:
self._report_cgroup_metrics(container, tags)
if "_proc_root" not in container:
containers_without_proc_root.append(DockerUtil.container_name_extractor(container)[0])
continue
self._report_net_metrics(container, tags)
except BogusPIDException as e:
self.log.warning('Unable to report cgroup metrics: %s', e)
if containers_without_proc_root:
message = "Couldn't find pid directory for containers: {0}. They'll be missing network metrics".format(
", ".join(containers_without_proc_root))
if not Platform.is_k8s():
self.warning(message)
else:
# On kubernetes, this is kind of expected. Network metrics will be collected by the kubernetes integration anyway
self.log.debug(message)
def _report_cgroup_metrics(self, container, tags):
cgroup_stat_file_failures = 0
if not container.get('_pid'):
raise BogusPIDException('Cannot report on bogus pid(0)')
for cgroup in CGROUP_METRICS:
try:
stat_file = self._get_cgroup_from_proc(cgroup["cgroup"], container['_pid'], cgroup['file'])
except MountException as e:
# We can't find a stat file
self.warning(str(e))
cgroup_stat_file_failures += 1
if cgroup_stat_file_failures >= len(CGROUP_METRICS):
self.warning("Couldn't find the cgroup files. Skipping the CGROUP_METRICS for now.")
except IOError as e:
self.log.debug("Cannot read cgroup file, container likely raced to finish : %s", e)
else:
stats = self._parse_cgroup_file(stat_file)
if stats:
for key, (dd_key, metric_func) in cgroup['metrics'].iteritems():
metric_func = FUNC_MAP[metric_func][self.use_histogram]
if key in stats:
metric_func(self, dd_key, int(stats[key]), tags=tags)
# Computed metrics
for mname, (key_list, fct, metric_func) in cgroup.get('to_compute', {}).iteritems():
values = [stats[key] for key in key_list if key in stats]
if len(values) != len(key_list):
self.log.debug("Couldn't compute {0}, some keys were missing.".format(mname))
continue
value = fct(*values)
metric_func = FUNC_MAP[metric_func][self.use_histogram]
if value is not None:
metric_func(self, mname, value, tags=tags)
def _report_net_metrics(self, container, tags):
"""Find container network metrics by looking at /proc/$PID/net/dev of the container process."""
if self._disable_net_metrics:
self.log.debug("Network metrics are disabled. Skipping")
return
proc_net_file = os.path.join(container['_proc_root'], 'net/dev')
try:
if container['Id'] in self.network_mappings:
networks = self.network_mappings[container['Id']]
else:
networks = self.docker_util.get_container_network_mapping(container)
if not networks:
networks = {'eth0': 'bridge'}
self.network_mappings[container['Id']] = networks
except Exception as e:
# Revert to previous behaviour if the method is missing or failing
# Debug message will only appear once per container, then the cache is used
self.log.debug("Failed to build docker network mapping, using failsafe. Exception: {0}".format(e))
networks = {'eth0': 'bridge'}
self.network_mappings[container['Id']] = networks
try:
with open(proc_net_file, 'r') as fp:
lines = fp.readlines()
"""Two first lines are headers:
Inter-| Receive | Transmit
face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed
"""
for l in lines[2:]:
cols = l.split(':', 1)
interface_name = str(cols[0]).strip()
if interface_name in networks:
net_tags = tags + ['docker_network:'+networks[interface_name]]
x = cols[1].split()
m_func = FUNC_MAP[RATE][self.use_histogram]
m_func(self, "docker.net.bytes_rcvd", long(x[0]), net_tags)
m_func(self, "docker.net.bytes_sent", long(x[8]), net_tags)
except Exception as e:
# It is possible that the container got stopped between the API call and now
self.warning("Failed to report IO metrics from file {0}. Exception: {1}".format(proc_net_file, e))
def _invalidate_network_mapping_cache(self, api_events):
for ev in api_events:
try:
if ev.get('Type') == 'network' and ev.get('Action').endswith('connect'):
container_id = ev.get('Actor').get('Attributes').get('container')
if container_id in self.network_mappings:
self.log.debug("Removing network mapping cache for container %s" % container_id)
del self.network_mappings[container_id]
except Exception:
self.log.warning('Malformed network event: %s' % str(ev))
def _process_events(self, containers_by_id):
api_events = self._get_events()
if self.collect_exit_codes:
self._report_exit_codes(api_events, containers_by_id)
if self.collect_events:
try:
aggregated_events = self._pre_aggregate_events(api_events, containers_by_id)
events = self._format_events(aggregated_events, containers_by_id)
except (socket.timeout, urllib2.URLError):
self.warning('Timeout when collecting events. Events will be missing.')
return
except Exception as e:
self.warning("Unexpected exception when collecting events: {0}. "
"Events will be missing".format(e))
return
for ev in events:
self.log.debug("Creating event: %s" % ev['msg_title'])
self.event(ev)
def _get_events(self):
"""Get the list of events."""
events, changed_container_ids = self.docker_util.get_events()
if not self._disable_net_metrics:
self._invalidate_network_mapping_cache(events)
if changed_container_ids and self._service_discovery:
get_sd_backend(self.agentConfig).update_checks(changed_container_ids)
if changed_container_ids:
self.metadata_collector.invalidate_cache(events)
if Platform.is_nomad():
self.nomadutil.invalidate_cache(events)
elif Platform.is_ecs_instance():
self.ecsutil.invalidate_cache(events)
return events
def _pre_aggregate_events(self, api_events, containers_by_id):
# Aggregate events, one per image. Put newer events first.
events = defaultdict(deque)
for event in api_events:
# Skip events related to filtered containers
container = containers_by_id.get(event.get('id'))
if container is not None and self._is_container_excluded(container):
self.log.debug("Excluded event: container {0} status changed to {1}".format(
event['id'], event['status']))
continue
# from may be missing (for network events for example)
if 'from' in event:
image_name = event['from']
if image_name.startswith('sha256:'):
image_name = self.docker_util.image_name_extractor({'Image': image_name})
events[image_name].appendleft(event)
return events
def _format_events(self, aggregated_events, containers_by_id):
events = []
for image_name, event_group in aggregated_events.iteritems():
container_tags = set()
low_prio_events = []
normal_prio_events = []
for event in event_group:
container_name = event['id'][:11]
if event['id'] in containers_by_id:
cont = containers_by_id[event['id']]
container_name = DockerUtil.container_name_extractor(cont)[0]
container_tags.update(self._get_tags(cont, PERFORMANCE))
container_tags.add('container_name:%s' % container_name)
# Add additionnal docker event attributes as tag
for attr in self.event_attributes_as_tags:
if attr in event['Actor']['Attributes'] and attr not in EXCLUDED_ATTRIBUTES:
container_tags.add('%s:%s' % (attr, event['Actor']['Attributes'][attr]))
# health checks generate tons of these so we treat them separately and lower their priority
if event['status'].startswith('exec_create:') or event['status'].startswith('exec_start:'):
low_prio_events.append((event, container_name))
else:
normal_prio_events.append((event, container_name))
exec_event = self._create_dd_event(low_prio_events, image_name, container_tags, priority='Low')
if exec_event:
events.append(exec_event)
normal_event = self._create_dd_event(normal_prio_events, image_name, container_tags, priority='Normal')
if normal_event:
events.append(normal_event)
return events
def _report_exit_codes(self, events, containers_by_id):
for event in events:
container_tags = set()
container = containers_by_id.get(event.get('id'))
# Skip events related to filtered containers
if container is not None and self._is_container_excluded(container):
continue
# Report the exit code in case of a DIE event
if container is not None and event['status'] == 'die':
container_name = DockerUtil.container_name_extractor(container)[0]
container_tags.update(self._get_tags(container, CONTAINER))
container_tags.add('container_name:%s' % container_name)
try:
exit_code = int(event['Actor']['Attributes']['exitCode'])
message = 'Container %s exited with %s' % (container_name, exit_code)
status = AgentCheck.OK if exit_code == 0 else AgentCheck.CRITICAL
self.service_check(EXIT_SERVICE_CHECK_NAME, status, tags=list(container_tags), message=message)
except KeyError:
self.log.warning('Unable to collect the exit code for container %s' % container_name)
def _create_dd_event(self, events, image, c_tags, priority='Normal'):
"""Create the actual event to submit from a list of similar docker events"""
if not events:
return
max_timestamp = 0
status = defaultdict(int)
status_change = []
for ev, c_name in events:
max_timestamp = max(max_timestamp, int(ev['time']))
status[ev['status']] += 1
status_change.append([c_name, ev['status']])
status_text = ", ".join(["%d %s" % (count, st) for st, count in status.iteritems()])
msg_title = "%s %s on %s" % (image, status_text, self.hostname)
msg_body = (
"%%%\n"
"{image_name} {status} on {hostname}\n"
"```\n{status_changes}\n```\n"
"%%%"
).format(
image_name=image,
status=status_text,
hostname=self.hostname,
status_changes="\n".join(
["%s \t%s" % (change[1].upper(), change[0]) for change in status_change])
)
if any(error in status_text for error in ERROR_ALERT_TYPE):
alert_type = "error"
else:
alert_type = None
return {
'timestamp': max_timestamp,
'host': self.hostname,
'event_type': EVENT_TYPE,
'msg_title': msg_title,
'msg_text': msg_body,
'source_type_name': EVENT_TYPE,
'event_object': 'docker:%s' % image,
'tags': list(c_tags),
'alert_type': alert_type,
'priority': priority
}
def _report_disk_stats(self):
"""Report metrics about the volume space usage"""
stats = {
'docker.data.used': None,
'docker.data.total': None,
'docker.data.free': None,
'docker.metadata.used': None,
'docker.metadata.total': None,
'docker.metadata.free': None
# these two are calculated by _calc_percent_disk_stats
# 'docker.data.percent': None,
# 'docker.metadata.percent': None
}
info = self.docker_client.info()
driver_status = info.get('DriverStatus', [])
if not driver_status:
self.log.warning('Disk metrics collection is enabled but docker info did not'
' report any. Your storage driver might not support them, skipping.')
return
for metric in driver_status:
# only consider metrics about disk space
if len(metric) == 2 and 'Space' in metric[0]:
# identify Data and Metadata metrics
mtype = 'data'
if 'Metadata' in metric[0]:
mtype = 'metadata'
if 'Used' in metric[0]:
stats['docker.{0}.used'.format(mtype)] = metric[1]
elif 'Space Total' in metric[0]:
stats['docker.{0}.total'.format(mtype)] = metric[1]
elif 'Space Available' in metric[0]:
stats['docker.{0}.free'.format(mtype)] = metric[1]
stats = self._format_disk_metrics(stats)
stats.update(self._calc_percent_disk_stats(stats))
tags = self._get_tags()
for name, val in stats.iteritems():
if val is not None:
self.gauge(name, val, tags)
def _format_disk_metrics(self, metrics):
"""Cast the disk stats to float and convert them to bytes"""
for name, raw_val in metrics.iteritems():
if raw_val:
match = DISK_STATS_RE.search(raw_val)
if match is None or len(match.groups()) != 2:
self.log.warning('Can\'t parse value %s for disk metric %s. Dropping it.' % (raw_val, name))
metrics[name] = None
val, unit = match.groups()
# by default some are uppercased others lowercased. That's error prone.
unit = unit.lower()
try:
val = int(float(val) * UNIT_MAP[unit])
metrics[name] = val
except KeyError:
self.log.error('Unrecognized unit %s for disk metric %s. Dropping it.' % (unit, name))
metrics[name] = None
return metrics
def _calc_percent_disk_stats(self, stats):
"""Calculate a percentage of used disk space for data and metadata"""
mtypes = ['data', 'metadata']
percs = {}
for mtype in mtypes:
used = stats.get('docker.{0}.used'.format(mtype))
total = stats.get('docker.{0}.total'.format(mtype))
free = stats.get('docker.{0}.free'.format(mtype))
if used and total and free and ceil(total) < free + used:
self.log.debug('used, free, and total disk metrics may be wrong, '
'used: %s, free: %s, total: %s',
used, free, total)
total = used + free
try:
if isinstance(used, int):
percs['docker.{0}.percent'.format(mtype)] = round(100 * float(used) / float(total), 2)
elif isinstance(free, int):
percs['docker.{0}.percent'.format(mtype)] = round(100 * (1.0 - (float(free) / float(total))), 2)
except ZeroDivisionError:
self.log.error('docker.{0}.total is 0, calculating docker.{1}.percent'
' is not possible.'.format(mtype, mtype))
return percs
# Cgroups
def _get_cgroup_from_proc(self, cgroup, pid, filename):
"""Find a specific cgroup file, containing metrics to extract."""
params = {
"file": filename,
}
return DockerUtil.find_cgroup_from_proc(self._mountpoints, pid, cgroup, self.docker_util._docker_root) % (params)
def _parse_cgroup_file(self, stat_file):
"""Parse a cgroup pseudo file for key/values."""
self.log.debug("Opening cgroup file: %s" % stat_file)
try:
with open(stat_file, 'r') as fp:
if 'blkio' in stat_file:
return self._parse_blkio_metrics(fp.read().splitlines())
elif 'cpuacct.usage' in stat_file:
return dict({'usage': str(int(fp.read())/10000000)})
else:
return dict(map(lambda x: x.split(' ', 1), fp.read().splitlines()))
except IOError:
# It is possible that the container got stopped between the API call and now.
# Some files can also be missing (like cpu.stat) and that's fine.
self.log.debug("Can't open %s. Its metrics will be missing." % stat_file)
def _parse_blkio_metrics(self, stats):
"""Parse the blkio metrics."""
metrics = {
'io_read': 0,
'io_write': 0,
}
for line in stats:
if 'Read' in line:
metrics['io_read'] += int(line.split()[2])
if 'Write' in line:
metrics['io_write'] += int(line.split()[2])
return metrics
def _is_container_cgroup(self, line, selinux_policy):
if line[1] not in ('cpu,cpuacct', 'cpuacct,cpu', 'cpuacct') or line[2] == '/docker-daemon':
return False
if 'docker' in line[2]: # general case
return True
if 'docker' in selinux_policy: # selinux
return True
if line[2].startswith('/') and re.match(CONTAINER_ID_RE, line[2][1:]): # kubernetes
return True
if line[2].startswith('/') and re.match(CONTAINER_ID_RE, line[2].split('/')[-1]): # kube 1.6+ qos hierarchy
return True
return False
# proc files
def _crawl_container_pids(self, container_dict, custom_cgroups=False):
"""Crawl `/proc` to find container PIDs and add them to `containers_by_id`."""
proc_path = os.path.join(self.docker_util._docker_root, 'proc')
pid_dirs = [_dir for _dir in os.listdir(proc_path) if _dir.isdigit()]
if len(pid_dirs) == 0:
self.warning("Unable to find any pid directory in {0}. "
"If you are running the agent in a container, make sure to "
'share the volume properly: "/proc:/host/proc:ro". '
"See https://github.com/DataDog/docker-dd-agent/blob/master/README.md for more information. "
"Network metrics will be missing".format(proc_path))
self._disable_net_metrics = True
return container_dict
self._disable_net_metrics = False
for folder in pid_dirs:
try:
path = os.path.join(proc_path, folder, 'cgroup')
with open(path, 'r') as f:
content = [line.strip().split(':') for line in f.readlines()]
selinux_policy = ''
path = os.path.join(proc_path, folder, 'attr', 'current')
if os.path.exists(path):
with open(path, 'r') as f:
selinux_policy = f.readlines()[0]
except IOError, e:
# Issue #2074
self.log.debug("Cannot read %s, process likely raced to finish : %s", path, e)
except Exception as e:
self.warning("Cannot read %s : %s" % (path, str(e)))
continue
try:
for line in content:
if self._is_container_cgroup(line, selinux_policy):
cpuacct = line[2]
break
else:
continue
matches = re.findall(CONTAINER_ID_RE, cpuacct)
if matches:
container_id = matches[-1]
if container_id not in container_dict:
self.log.debug(
"Container %s not in container_dict, it's likely excluded", container_id
)
continue
container_dict[container_id]['_pid'] = folder
container_dict[container_id]['_proc_root'] = os.path.join(proc_path, folder)
elif custom_cgroups: # if we match by pid that should be enough (?) - O(n) ugh!
for _, container in container_dict.iteritems():
if container.get('_pid') == int(folder):
container['_proc_root'] = os.path.join(proc_path, folder)
break
except Exception, e:
self.warning("Cannot parse %s content: %s" % (path, str(e)))
continue