def info(self) -> ClusterStats: # pylint: disable=too-many-locals
"""Retrieve Kubernetes cluster statistics."""
pl_status = ClusterStats()
node_list = pykube.Node.objects(
self.api).filter(namespace=pykube.all).iterator()
node_dict = {}
# Get basic information from nodes
for node in node_list:
nss = NodeStats(node.name)
nss.cores_total = float(node.obj['status']['allocatable']['cpu']
[1]) ## Bug found on GKE
nss.memory_total = humanfriendly.parse_size(
node.obj['status']['allocatable']['memory'])
nss.labels = node.obj['metadata']['labels']
nss.status = 'online'
node_dict[str(socket.gethostbyname(node.name))] = nss
# Get information from all running pods, then accumulate to nodes
pod_list = pykube.Pod.objects(
self.api).filter(namespace=pykube.all).iterator()
for pod in pod_list:
try:
host_ip = pod.obj['status']['hostIP']
except KeyError:
continue
nss = node_dict[host_ip]
nss.container_count += 1
spec_cont = pod.obj['spec']['containers'][0]
if 'resources' in spec_cont:
if 'requests' in spec_cont['resources']:
if 'memory' in spec_cont['resources']['requests']:
memory = spec_cont['resources']['requests']['memory']
nss.memory_reserved = nss.memory_reserved + humanfriendly.parse_size(
memory)
if 'cpu' in spec_cont['resources']['requests']:
cpu = spec_cont['resources']['requests']['cpu']
# ex: cpu could be 100m or 0.1
cpu_splitted = cpu.split('m')
if len(cpu_splitted) > 1:
cpu_float = int(cpu_splitted[0]) / 1000
else:
cpu_float = int(cpu_splitted[0])
nss.cores_reserved = round(
nss.cores_reserved + cpu_float, 3)
for node_ip in node_dict:
pl_status.nodes.append(node_dict[node_ip])
return pl_status
def _host_subthread(self, host_config: DockerHostConfig):
log.info("Synchro thread for host {} started".format(host_config.name))
self.host_stats[host_config.name] = NodeStats(host_config.name)
while True:
time_start = time.time()
try:
my_engine = DockerClient(host_config)
container_list = my_engine.list(
only_label={
'zoe_deployment_name': get_conf().deployment_name
})
info = my_engine.info()
except ZoeException as e:
self.host_stats[host_config.name].status = 'offline'
log.error(str(e))
log.info('Node {} is offline'.format(host_config.name))
else:
if self.host_stats[host_config.name].status == 'offline':
log.info('Node {} is now online'.format(host_config.name))
self.host_stats[host_config.name].status = 'online'
self.host_stats[
host_config.name].container_count = info['Containers']
self.host_stats[host_config.name].cores_total = info['NCPU']
self.host_stats[
host_config.name].memory_total = info['MemTotal']
self.host_stats[host_config.name].labels = host_config.labels
if info['Labels'] is not None:
self.host_stats[host_config.name].labels.union(
set(info['Labels']))
self.host_stats[host_config.name].memory_allocated = sum([
cont['memory_soft_limit'] for cont in container_list
if cont['memory_soft_limit'] != info['MemTotal']
])
self.host_stats[host_config.name].cores_allocated = sum([
cont['cpu_quota'] / cont['cpu_period']
for cont in container_list if cont['cpu_period'] != 0
])
stats = {}
self.host_stats[host_config.name].memory_reserved = 0
self.host_stats[host_config.name].cores_reserved = 0
for cont in container_list:
service = self.state.services.select(
only_one=True,
backend_host=host_config.name,
backend_id=cont['id'])
if service is None:
log.warning(
'Container {} on host {} has no corresponding service'
.format(cont['name'], host_config.name))
if cont['state'] == Service.BACKEND_DIE_STATUS:
log.warning(
'Terminating dead and orphan container {}'.
format(cont['name']))
my_engine.terminate_container(cont['id'],
delete=True)
continue
self._update_service_status(service, cont)
self.host_stats[
host_config.
name].memory_reserved += service.resource_reservation.memory.min
self.host_stats[
host_config.
name].cores_reserved += service.resource_reservation.cores.min
stats[service.id] = {
'core_limit': cont['cpu_quota'] / cont['cpu_period'],
'mem_limit': cont['memory_soft_limit']
}
self.host_stats[host_config.name].service_stats = stats
self.host_stats[host_config.name].images = []
for dk_image in my_engine.list_images():
image = {
'id': dk_image.attrs['Id'],
'size': dk_image.attrs['Size'],
'names': dk_image.tags # type: list
}
for name in image['names']:
if name[-7:] == ':latest': # add an image with the name without 'latest' to fake Docker image lookup algorithm
image['names'].append(name[:-7])
break
self.host_stats[host_config.name].images.append(image)
sleep_time = CHECK_INTERVAL - (time.time() - time_start)
if sleep_time <= 0:
log.warning(
'synchro thread for host {} is late by {:.2f} seconds'.
format(host_config.name, sleep_time * -1))
sleep_time = 0
if self.stop.wait(timeout=sleep_time):
break
log.info("Synchro thread for host {} stopped".format(host_config.name))
def _update_node_state(self, host_conf: DockerHostConfig, node_stats: NodeStats, get_usage_stats: bool):
node_stats.labels = host_conf.labels
try:
my_engine = DockerClient(host_conf)
except ZoeException as e:
log.error(str(e))
node_stats.status = 'offline'
log.info('Node {} is offline'.format(host_conf.name))
return
else:
node_stats.status = 'online'
try:
container_list = my_engine.list(only_label={'zoe_deployment_name': get_conf().deployment_name})
info = my_engine.info()
except ZoeException:
return
node_stats.container_count = len(container_list)
node_stats.cores_total = info['NCPU']
node_stats.memory_total = info['MemTotal']
if info['Labels'] is not None:
node_stats.labels += set(info['Labels'])
node_stats.memory_reserved = sum([cont['memory_soft_limit'] for cont in container_list if cont['memory_soft_limit'] != node_stats.memory_total])
node_stats.cores_reserved = sum([cont['cpu_quota'] / cont['cpu_period'] for cont in container_list if cont['cpu_period'] != 0])
stats = {}
for cont in container_list:
stats[cont['id']] = {}
stats[cont['id']]['core_limit'] = cont['cpu_quota'] / cont['cpu_period']
stats[cont['id']]['mem_limit'] = cont['memory_soft_limit']
node_stats.service_stats = stats
if get_usage_stats:
if get_conf().kairosdb_enable:
kdb = KairosDBInMetrics()
for cont in container_list:
stats[cont['id']].update(kdb.get_service_usage(cont['name']))
node_stats.memory_in_use = sum([stat['mem_usage'] for stat in stats.values()])
node_stats.cores_in_use = sum([stat['cpu_usage'] for stat in stats.values()])
else:
for cont in container_list:
try:
aux = my_engine.stats(cont['id'], stream=False) # this call is very slow (>~1sec)
if 'usage' in aux['memory_stats']:
stats[cont['id']]['mem_usage'] = aux['memory_stats']['usage']
else:
stats[cont['id']]['mem_usage'] = 0
stats[cont['id']]['cpu_usage'] = self._get_core_usage(aux)
except ZoeException:
continue
node_stats.memory_in_use = sum([stat['mem_usage'] for stat in stats.values()])
node_stats.cores_in_use = sum([stat['cpu_usage'] for stat in stats.values()])
else:
node_stats.memory_in_use = 0
node_stats.cores_in_use = 0
def info(self) -> ClusterStats:
"""Retrieve Swarm statistics. The Docker API returns a mess difficult to parse."""
info = self.cli.info()
pl_status = ClusterStats()
# SystemStatus is a list...
idx = 0 # Role, skip
idx += 1
assert 'Strategy' in info["SystemStatus"][idx][0]
pl_status.placement_strategy = info["SystemStatus"][idx][1]
idx += 1
assert 'Filters' in info["SystemStatus"][idx][0]
pl_status.active_filters = [x.strip() for x in info["SystemStatus"][idx][1].split(", ")]
idx += 1
assert 'Nodes' in info["SystemStatus"][idx][0]
node_count = int(info["SystemStatus"][idx][1])
idx += 1 # At index 4 the nodes begin
for node in range(node_count):
idx2 = 0
node_stats = NodeStats(info["SystemStatus"][idx + node][0].strip())
node_stats.docker_endpoint = info["SystemStatus"][idx + node][1]
idx2 += 1 # ID, skip
idx2 += 1 # Status
if info["SystemStatus"][idx + node + idx2][1] == 'Healthy':
node_stats.status = 'online'
else:
node_stats.status = 'offline'
idx2 += 1 # Containers
node_stats.container_count = int(info["SystemStatus"][idx + node + idx2][1].split(' ')[0])
idx2 += 1 # CPUs
node_stats.cores_reserved = int(info["SystemStatus"][idx + node + idx2][1].split(' / ')[0])
node_stats.cores_total = int(info["SystemStatus"][idx + node + idx2][1].split(' / ')[1])
idx2 += 1 # Memory
node_stats.memory_reserved = info["SystemStatus"][idx + node + idx2][1].split(' / ')[0]
node_stats.memory_total = info["SystemStatus"][idx + node + idx2][1].split(' / ')[1]
idx2 += 1 # Labels
node_stats.labels = info["SystemStatus"][idx + node + idx2][1].split(', ')
idx2 += 1 # Last update
node_stats.last_update = info["SystemStatus"][idx + node + idx2][1]
idx2 += 1 # Docker version
node_stats.server_version = info["SystemStatus"][idx + node + idx2][1]
node_stats.memory_reserved = humanfriendly.parse_size(node_stats.memory_reserved)
node_stats.memory_total = humanfriendly.parse_size(node_stats.memory_total)
pl_status.nodes.append(node_stats)
idx += idx2
pl_status.timestamp = time.time()
return pl_status