def get_mesos_task_count_by_slave(mesos_state, pool=None):
"""Get counts of running tasks per mesos slave. Also include separate count of chronos tasks
:param mesos_state: mesos state dict
:param pool: pool of slaves to return (None means all)
:returns: dict of {<slave_id>:{'count': <count>, 'slave': <slave_object>, 'chronos_count':<chronos_count>}}
"""
all_mesos_tasks = get_running_tasks_from_active_frameworks('') # empty string matches all app ids
if pool:
slaves = {
slave['id']: {'count': 0, 'slave': slave, 'chronos_count': 0} for slave in mesos_state.get('slaves', [])
if slave['attributes'].get('pool', 'default') == pool
}
else:
slaves = {
slave['id']: {'count': 0, 'slave': slave, 'chronos_count': 0} for slave in mesos_state.get('slaves', [])
}
for task in all_mesos_tasks:
if task.slave['id'] not in slaves:
log.error("Task associated with %s but slave not in masters list of slaves" % task.slave['id'])
continue
else:
slaves[task.slave['id']]['count'] += 1
log.debug("Task framework: {0}".format(task.framework.name))
if task.framework.name == 'chronos':
slaves[task.slave['id']]['chronos_count'] += 1
slaves = {slave_pid: SlaveTaskCount(**slave_counts) for slave_pid, slave_counts in slaves.items()}
for slave in slaves.values():
log.info("Slave: {0}, running {1} tasks, including {2} chronos tasks".format(slave.slave['pid'],
slave.count,
slave.chronos_count))
return slaves
def get_mesos_leader():
"""Get the current mesos-master leader's hostname.
Attempts to determine this by using mesos.cli to query ZooKeeper.
:returns: The current mesos-master hostname"""
try:
url = master.CURRENT.host
except MesosMasterConnectionError:
log.debug('mesos.cli failed to provide the master host')
raise
log.debug("mesos.cli thinks the master host is: %s" % url)
hostname = urlparse(url).hostname
log.debug("The parsed master hostname is: %s" % hostname)
# This check is necessary, as if we parse a value such as 'localhost:5050',
# it won't have a hostname attribute
if hostname:
try:
host = socket.gethostbyaddr(hostname)[0]
fqdn = socket.getfqdn(host)
except (socket.error, socket.herror, socket.gaierror, socket.timeout):
log.debug("Failed to convert mesos leader hostname to fqdn!")
raise
log.debug("Mesos Leader: %s" % fqdn)
return fqdn
else:
raise ValueError('Expected to receive a valid URL, got: %s' % url)
def get_mesos_leader():
"""Get the current mesos-master leader's hostname.
Attempts to determine this by using mesos.cli to query ZooKeeper.
:returns: The current mesos-master hostname"""
try:
url = master.CURRENT.host
except MesosMasterConnectionError:
log.debug('mesos.cli failed to provide the master host')
raise
log.debug("mesos.cli thinks the master host is: %s" % url)
hostname = urlparse(url).hostname
log.debug("The parsed master hostname is: %s" % hostname)
# This check is necessary, as if we parse a value such as 'localhost:5050',
# it won't have a hostname attribute
if hostname:
try:
host = socket.gethostbyaddr(hostname)[0]
fqdn = socket.getfqdn(host)
except (socket.error, socket.herror, socket.gaierror, socket.timeout):
log.debug("Failed to convert mesos leader hostname to fqdn!")
raise
log.debug("Mesos Leader: %s" % fqdn)
return fqdn
else:
raise ValueError('Expected to receive a valid URL, got: %s' % url)
def get_mesos_task_count_by_slave(mesos_state, pool=None):
"""Get counts of running tasks per mesos slave. Also include separate count of chronos tasks
:param mesos_state: mesos state dict
:param pool: pool of slaves to return (None means all)
:returns: dict of {<slave_id>:{'count': <count>, 'slave': <slave_object>, 'chronos_count':<chronos_count>}}
"""
all_mesos_tasks = get_running_tasks_from_active_frameworks(
'') # empty string matches all app ids
if pool:
slaves = {
slave['id']: {
'count': 0,
'slave': slave,
'chronos_count': 0
}
for slave in mesos_state.get('slaves', [])
if slave['attributes'].get('pool', 'default') == pool
}
else:
slaves = {
slave['id']: {
'count': 0,
'slave': slave,
'chronos_count': 0
}
for slave in mesos_state.get('slaves', [])
}
for task in all_mesos_tasks:
if task.slave['id'] not in slaves:
log.debug(
"Task associated with slave {0} not found in {1} pool".format(
task.slave['id'], pool))
continue
else:
slaves[task.slave['id']]['count'] += 1
log.debug("Task framework: {0}".format(task.framework.name))
if task.framework.name == 'chronos':
slaves[task.slave['id']]['chronos_count'] += 1
slaves = {
slave_counts['slave']['hostname']: SlaveTaskCount(**slave_counts)
for slave_counts in slaves.values()
}
for slave in slaves.values():
log.debug("Slave: {0}, running {1} tasks, including {2} chronos tasks".
format(slave.slave['hostname'], slave.count,
slave.chronos_count))
return slaves
