def get_fs(self):
fs_map = self.get('fs_map')
servers = self.get_service_list()
self.log.debug('standbys: {}'.format(fs_map['standbys']))
# export standby mds metadata, default standby fs_id is '-1'
for standby in fs_map['standbys']:
id_ = standby['name']
host_version = servers.get((id_, 'mds'), ('', ''))
self.metrics['mds_metadata'].set(1, (
'mds.{}'.format(id_), '-1',
host_version[0], standby['addr'],
standby['rank'], host_version[1]
))
for fs in fs_map['filesystems']:
# collect fs metadata
data_pools = ",".join([str(pool)
for pool in fs['mdsmap']['data_pools']])
self.metrics['fs_metadata'].set(1, (
data_pools,
fs['id'],
fs['mdsmap']['metadata_pool'],
fs['mdsmap']['fs_name']
))
self.log.debug('mdsmap: {}'.format(fs['mdsmap']))
for gid, daemon in fs['mdsmap']['info'].items():
id_ = daemon['name']
host_version = servers.get((id_, 'mds'), ('', ''))
self.metrics['mds_metadata'].set(1, (
'mds.{}'.format(id_), fs['id'],
host_version[0], daemon['addr'],
daemon['rank'], host_version[1]
))
def get_quorum_status(self):
mon_status = json.loads(self.get('mon_status')['json'])
servers = self.get_service_list()
for mon in mon_status['monmap']['mons']:
rank = mon['rank']
id_ = mon['name']
host_version = servers.get((id_, 'mon'), ('', ''))
self.metrics['mon_metadata'].set(
1, ('mon.{}'.format(id_), host_version[0],
mon['public_addr'].split(':')[0], rank, host_version[1]))
in_quorum = int(rank in mon_status['quorum'])
self.metrics['mon_quorum_status'].set(in_quorum,
('mon.{}'.format(id_), ))
def get_quorum_status(self):
mon_status = json.loads(self.get('mon_status')['json'])
servers = self.get_service_list()
for mon in mon_status['monmap']['mons']:
rank = mon['rank']
id_ = mon['name']
host_version = servers.get((id_, 'mon'), ('', ''))
self.metrics['mon_metadata'].set(1, (
'mon.{}'.format(id_), host_version[0],
mon['public_addr'].split(':')[0], rank,
host_version[1]
))
in_quorum = int(rank in mon_status['quorum'])
self.metrics['mon_quorum_status'].set(in_quorum, (
'mon.{}'.format(id_),
))
def get_mgr_status(self):
mgr_map = self.get('mgr_map')
servers = self.get_service_list()
active = mgr_map['active_name']
standbys = [s.get('name') for s in mgr_map['standbys']]
all_mgrs = list(standbys)
all_mgrs.append(active)
all_modules = {module.get('name'):module.get('can_run') for module in mgr_map['available_modules']}
ceph_release = None
for mgr in all_mgrs:
host_version = servers.get((mgr, 'mgr'), ('', ''))
if mgr == active:
_state = 1
ceph_release = host_version[1].split()[-2] # e.g. nautilus
else:
_state = 0
self.metrics['mgr_metadata'].set(1, (
'mgr.{}'.format(mgr), host_version[0],
host_version[1]
))
self.metrics['mgr_status'].set(_state, (
'mgr.{}'.format(mgr),
))
always_on_modules = mgr_map['always_on_modules'].get(ceph_release, [])
active_modules = list(always_on_modules)
active_modules.extend(mgr_map['modules'])
for mod_name in all_modules.keys():
if mod_name in always_on_modules:
_state = 2
elif mod_name in active_modules:
_state = 1
else:
_state = 0
_can_run = 1 if all_modules[mod_name] else 0
self.metrics['mgr_module_status'].set(_state, (mod_name,))
self.metrics['mgr_module_can_run'].set(_can_run, (mod_name,))
def get_fs(self):
fs_map = self.get('fs_map')
servers = self.get_service_list()
for fs in fs_map['filesystems']:
# collect fs metadata
data_pools = ",".join([str(pool)
for pool in fs['mdsmap']['data_pools']])
self.metrics['fs_metadata'].set(1, (
data_pools,
fs['id'],
fs['mdsmap']['metadata_pool'],
fs['mdsmap']['fs_name']
))
self.log.debug('mdsmap: {}'.format(fs['mdsmap']))
for gid, daemon in fs['mdsmap']['info'].items():
id_ = daemon['name']
host_version = servers.get((id_, 'mds'), ('', ''))
self.metrics['mds_metadata'].set(1, (
'mds.{}'.format(id_), fs['id'],
host_version[0], daemon['addr'],
daemon['rank'], host_version[1]
))
def get_metadata_and_osd_status(self):
osd_map = self.get('osd_map')
osd_flags = osd_map['flags'].split(',')
for flag in OSD_FLAGS:
self.metrics['osd_flag_{}'.format(flag)].set(
int(flag in osd_flags)
)
osd_devices = self.get('osd_map_crush')['devices']
servers = self.get_service_list()
for osd in osd_map['osds']:
# id can be used to link osd metrics and metadata
id_ = osd['osd']
# collect osd metadata
p_addr = osd['public_addr'].split(':')[0]
c_addr = osd['cluster_addr'].split(':')[0]
if p_addr == "-" or c_addr == "-":
self.log.info(
"Missing address metadata for osd {0}, skipping occupation"
" and metadata records for this osd".format(id_)
)
continue
dev_class = None
for osd_device in osd_devices:
if osd_device['id'] == id_:
dev_class = osd_device.get('class', '')
break
if dev_class is None:
self.log.info("OSD {0} is missing from CRUSH map, "
"skipping output".format(id_))
continue
host_version = servers.get((str(id_), 'osd'), ('', ''))
# collect disk occupation metadata
osd_metadata = self.get_metadata("osd", str(id_))
if osd_metadata is None:
continue
obj_store = osd_metadata.get('osd_objectstore', '')
f_iface = osd_metadata.get('front_iface', '')
b_iface = osd_metadata.get('back_iface', '')
self.metrics['osd_metadata'].set(1, (
b_iface,
'osd.{}'.format(id_),
c_addr,
dev_class,
f_iface,
host_version[0],
obj_store,
p_addr,
host_version[1]
))
# collect osd status
for state in OSD_STATUS:
status = osd[state]
self.metrics['osd_{}'.format(state)].set(status, (
'osd.{}'.format(id_),
))
osd_dev_node = None
if obj_store == "filestore":
# collect filestore backend device
osd_dev_node = osd_metadata.get(
'backend_filestore_dev_node', None)
# collect filestore journal device
osd_wal_dev_node = osd_metadata.get('osd_journal', '')
osd_db_dev_node = ''
elif obj_store == "bluestore":
# collect bluestore backend device
osd_dev_node = osd_metadata.get(
'bluestore_bdev_dev_node', None)
# collect bluestore wal backend
osd_wal_dev_node = osd_metadata.get('bluefs_wal_dev_node', '')
# collect bluestore db backend
osd_db_dev_node = osd_metadata.get('bluefs_db_dev_node', '')
if osd_dev_node and osd_dev_node == "unknown":
osd_dev_node = None
osd_hostname = osd_metadata.get('hostname', None)
if osd_dev_node and osd_hostname:
self.log.debug("Got dev for osd {0}: {1}/{2}".format(
id_, osd_hostname, osd_dev_node))
self.metrics['disk_occupation'].set(1, (
"osd.{0}".format(id_),
osd_dev_node,
osd_db_dev_node,
osd_wal_dev_node,
osd_hostname
))
else:
self.log.info("Missing dev node metadata for osd {0}, skipping "
"occupation record for this osd".format(id_))
for pool in osd_map['pools']:
self.metrics['pool_metadata'].set(
1, (pool['pool'], pool['pool_name']))
# Populate other servers metadata
for key, value in servers.items():
service_id, service_type = key
if service_type == 'rgw':
hostname, version = value
self.metrics['rgw_metadata'].set(
1,
('{}.{}'.format(service_type, service_id),
hostname, version)
)
elif service_type == 'rbd-mirror':
mirror_metadata = self.get_metadata('rbd-mirror', service_id)
if mirror_metadata is None:
continue
mirror_metadata['ceph_daemon'] = '{}.{}'.format(service_type,
service_id)
self.metrics['rbd_mirror_metadata'].set(
1, (mirror_metadata.get(k, '')
for k in RBD_MIRROR_METADATA)
)
def get_metadata_and_osd_status(self):
osd_map = self.get('osd_map')
osd_flags = osd_map['flags'].split(',')
for flag in OSD_FLAGS:
self.metrics['osd_flag_{}'.format(flag)].set(
int(flag in osd_flags)
)
osd_devices = self.get('osd_map_crush')['devices']
servers = self.get_service_list()
for osd in osd_map['osds']:
# id can be used to link osd metrics and metadata
id_ = osd['osd']
# collect osd metadata
p_addr = osd['public_addr'].split(':')[0]
c_addr = osd['cluster_addr'].split(':')[0]
if p_addr == "-" or c_addr == "-":
self.log.info(
"Missing address metadata for osd {0}, skipping occupation"
" and metadata records for this osd".format(id_)
)
continue
dev_class = None
for osd_device in osd_devices:
if osd_device['id'] == id_:
dev_class = osd_device.get('class', '')
break
if dev_class is None:
self.log.info(
"OSD {0} is missing from CRUSH map, skipping output".format(
id_))
continue
host_version = servers.get((str(id_), 'osd'), ('', ''))
# collect disk occupation metadata
osd_metadata = self.get_metadata("osd", str(id_))
if osd_metadata is None:
continue
obj_store = osd_metadata.get('osd_objectstore', '')
f_iface = osd_metadata.get('front_iface', '')
b_iface = osd_metadata.get('back_iface', '')
self.metrics['osd_metadata'].set(1, (
b_iface,
'osd.{}'.format(id_),
c_addr,
dev_class,
f_iface,
host_version[0],
obj_store,
p_addr,
host_version[1]
))
# collect osd status
for state in OSD_STATUS:
status = osd[state]
self.metrics['osd_{}'.format(state)].set(status, (
'osd.{}'.format(id_),
))
if obj_store == "filestore":
# collect filestore backend device
osd_dev_node = osd_metadata.get(
'backend_filestore_dev_node', None)
# collect filestore journal device
osd_wal_dev_node = osd_metadata.get('osd_journal', '')
osd_db_dev_node = ''
elif obj_store == "bluestore":
# collect bluestore backend device
osd_dev_node = osd_metadata.get(
'bluestore_bdev_dev_node', None)
# collect bluestore wal backend
osd_wal_dev_node = osd_metadata.get('bluefs_wal_dev_node', '')
# collect bluestore db backend
osd_db_dev_node = osd_metadata.get('bluefs_db_dev_node', '')
if osd_dev_node and osd_dev_node == "unknown":
osd_dev_node = None
osd_hostname = osd_metadata.get('hostname', None)
if osd_dev_node and osd_hostname:
self.log.debug("Got dev for osd {0}: {1}/{2}".format(
id_, osd_hostname, osd_dev_node))
self.metrics['disk_occupation'].set(1, (
"osd.{0}".format(id_),
osd_dev_node,
osd_db_dev_node,
osd_wal_dev_node,
osd_hostname
))
else:
self.log.info("Missing dev node metadata for osd {0}, skipping "
"occupation record for this osd".format(id_))
pool_meta = []
for pool in osd_map['pools']:
self.metrics['pool_metadata'].set(
1, (pool['pool'], pool['pool_name']))
# Populate other servers metadata
for key, value in servers.items():
service_id, service_type = key
if service_type == 'rgw':
hostname, version = value
self.metrics['rgw_metadata'].set(
1,
('{}.{}'.format(service_type, service_id), hostname, version)
)
elif service_type == 'rbd-mirror':
mirror_metadata = self.get_metadata('rbd-mirror', service_id)
if mirror_metadata is None:
continue
mirror_metadata['ceph_daemon'] = '{}.{}'.format(service_type,
service_id)
self.metrics['rbd_mirror_metadata'].set(
1, (mirror_metadata.get(k, '')
for k in RBD_MIRROR_METADATA)
)
