def _stats_lookup(self):
"""
Grab the disk stats from /proc
"""
now = int(time.time())
interval = now - self.timestamp
self.timestamp = now
for perf_entry in freadlines('/proc/diskstats'):
field = perf_entry.split()
dev_name = field[2]
if dev_name in self.osd.keys():
new_stats = field[3:]
device = self.osd[dev_name]
if device.perf._current:
device.perf._previous = device.perf._current
device.perf._current = new_stats
else:
device.perf._current = new_stats
device.perf.compute(interval)
device.refresh()
def _stats_lookup(self):
"""
Grab the disk stats from /proc
"""
now = time.time()
interval = int(now) - self.timestamp
self.timestamp = int(now)
for perf_entry in freadlines('/proc/diskstats'):
field = perf_entry.split()
dev_name = field[2]
if dev_name in self.osd.keys():
new_stats = field[3:]
device = self.osd[dev_name]
if device.perf._current:
device.perf._previous = device.perf._current
device.perf._current = new_stats
else:
device.perf._current = new_stats
device.perf.compute(interval)
device.refresh()
end = time.time()
self.elapsed_log_msg("disk performance stats generation", (end - now))
def _dev_to_osd(self):
"""
Look at the system to determine which disks are acting as OSD's
"""
osd_indicators = {'var', 'lib', 'osd'}
for mnt in freadlines('/proc/mounts'):
items = mnt.split(' ')
dev_path, path_name = items[:2]
if path_name.startswith('/var/lib'):
# take a close look since this is where ceph osds usually
# get mounted
dirs = set(path_name.split('/'))
if dirs.issuperset(osd_indicators):
osd_id = path_name.split('-')[-1]
device = filter(lambda ch: ch.isalpha(),
dev_path.split('/')[-1])
if device not in self.osd:
disk = Disk()
disk._name = device
disk._path_name = path_name
disk.osd_id = osd_id
disk.rotational = disk._get_rota()
disk.perf = IOstat()
disk.disk_size = disk._get_size()
disk.refresh()
self.osd[device] = disk
def probe(self):
"""
set up which collector(s) to use, based on what types of sockets we
find in /var/run/ceph
"""
mon_socket = '/var/run/ceph/{}-mon.{}.asok'.format(
self.cluster_name, self.host_name)
if os.path.exists(mon_socket):
self.mon_socket = mon_socket
self.mon = Mon(self.cluster_name, admin_socket=mon_socket)
rgw_socket_list = glob.glob('/var/run/ceph/{}-client.rgw.{}.'
'*.asok'.format(self.cluster_name,
self.host_name))
if rgw_socket_list:
rgw_socket = rgw_socket_list[0]
self.rgw = RGW(self.cluster_name, admin_socket=rgw_socket)
osd_socket_list = glob.glob('/var/run/ceph/{}-osd.*'
'.asok'.format(self.cluster_name))
mounted = freadlines('/proc/mounts')
osds_mounted = [
mnt for mnt in mounted
if mnt.split()[1].startswith('/var/lib/ceph')
]
if osd_socket_list or osds_mounted:
self.osd = OSDs(self.cluster_name)
collectd.info("{}: Roles detected - mon:{} "
"osd:{} rgw:{}".format(__name__,
isinstance(self.mon, Mon),
isinstance(self.osd, OSDs),
isinstance(self.rgw, RGW)))
def _stats_lookup(self):
"""
Grab the disk stats from /proc/diskstats, and the key osd perf dump
counters
"""
now = time.time()
interval = int(now) - self.timestamp
self.timestamp = int(now)
# Fetch diskstats from the OS
for perf_entry in freadlines('/proc/diskstats'):
field = perf_entry.split()
dev_name = field[2]
device = None
if self.dev_lookup.get(dev_name, None) == 'osd':
device = self.osd[dev_name]
elif self.dev_lookup.get(dev_name, None) == 'jrnl':
device = self.jrnl[dev_name]
if device:
new_stats = field[3:]
if device.perf._current:
device.perf._previous = device.perf._current
device.perf._current = new_stats
else:
device.perf._current = new_stats
device.perf.compute(interval)
end = time.time()
self.logger.debug("OS disk stats calculated in "
"{:.4f}s".format(end - now))
# fetch stats from each osd daemon
osd_stats_start = time.time()
for osd_id in self.osd_id_list:
osd_type = self.osd[osd_id]._osd_type
if osd_type in OSDs.supported_object_stores:
osd_stats = self._fetch_osd_stats(osd_id, osd_type)
osd_device = self.osd[osd_id]
osd_device.update(osd_stats)
else:
self.logger.warning("Unknown OSD type encountered for "
"osd.{}".format(osd_id))
osd_stats_end = time.time()
self.logger.debug(
"OSD perf dump stats collected for {} OSDs "
"in {:.3f}s".format(len(self.osd_id_list),
(osd_stats_end - osd_stats_start)))
def _dev_to_osd(self):
"""
Look at the system to determine which disks are acting as OSD's
"""
# the logic here uses the mount points to determine which OSD's are
# in the system - so the focus is on filestore (XFS) OSD's. Another
# approach could be to go directly to the admin socket (status cmd)
# to get the osd_fsid, and then lookup that in /dev/disk/by-partuuid
# to derive the osd device name...
osd_indicators = {'var', 'lib', 'osd'}
for mnt in freadlines('/proc/mounts'):
items = mnt.split(' ')
dev_path, path_name = items[:2]
if path_name.startswith('/var/lib'):
# take a close look since this is where ceph osds usually
# get mounted
dirs = set(path_name.split('/'))
if dirs.issuperset(osd_indicators):
osd_id = path_name.split('-')[-1]
osd_device = dev_path.split('/')[-1]
if osd_device not in self.osd:
self.osd[osd_device] = Disk(osd_device,
path_name=path_name,
osd_id=osd_id)
self.dev_lookup[osd_device] = 'osd'
if osd_id not in self.osd:
self.osd[osd_id] = OSDstats()
self.osd_id_list.append(osd_id)
journal_link = os.path.join(path_name, 'journal')
if os.path.exists(journal_link):
# this is a filestore based OSD
jrnl_path = os.path.realpath(journal_link)
jrnl_dev = jrnl_path.split('/')[-1]
if jrnl_dev not in self.osd:
self.jrnl[jrnl_dev] = Disk(jrnl_dev,
osd_id=osd_id)
self.dev_lookup[jrnl_dev] = 'jrnl'
else:
# No journal..?
pass
def _stats_lookup(self):
"""
Grab the disk stats from /proc
"""
now = time.time()
interval = int(now) - self.timestamp
self.timestamp = int(now)
# Fetch diskstats from the OS
for perf_entry in freadlines('/proc/diskstats'):
field = perf_entry.split()
dev_name = field[2]
device = None
if self.dev_lookup.get(dev_name, None) == 'osd':
device = self.osd[dev_name]
elif self.dev_lookup.get(dev_name, None) == 'jrnl':
device = self.jrnl[dev_name]
if device:
new_stats = field[3:]
if device.perf._current:
device.perf._previous = device.perf._current
device.perf._current = new_stats
else:
device.perf._current = new_stats
device.perf.compute(interval)
device.refresh()
# fetch stats from each osd daemon
for osd_id in self.osd_id_list:
osd_stats = self._fetch_osd_stats(osd_id)
self.logger.debug('stats : {}'.format(osd_stats))
osd_device = self.osd[osd_id]
osd_device.update(osd_stats)
end = time.time()
self.elapsed_log_msg("disk performance stats generation", (end - now))
def _dev_to_osd(self):
"""
Look at the system to determine which disks are acting as OSD's
"""
# the logic here uses the mount points to determine which OSD's are
# in the system. The encryption state is determine just by the use
# devicemapper (i.e. /dev/mapper prefixed devices) - since at this time
# this is all dm is used for.
osd_indicators = {'var', 'lib', 'osd'}
for mnt in freadlines('/proc/mounts'):
items = mnt.split(' ')
dev_path, path_name = items[:2]
if path_name.startswith('/var/lib'):
# take a close look since this is where ceph osds usually
# get mounted
dirs = set(path_name.split('/'))
if dirs.issuperset(osd_indicators):
# get the osd_id from the name is the most simple way
# to get the id, due to naming conventions. If this fails
# though, plan 'b' is the whoami file
osd_id = path_name.split('-')[-1]
if not osd_id.isdigit():
osd_id = fread(os.path.join(path_name, 'whoami'))
if osd_id not in self.osd:
osd_type = OSDs.get_osd_type(path_name)
self.osd[osd_id] = OSDstats(osd_type=osd_type)
self.osd_id_list.append(osd_id)
osd_type = self.osd[osd_id]._osd_type
if osd_type == 'filestore':
if dev_path.startswith('/dev/mapper'):
encrypted = 1
uuid = dev_path.split('/')[-1]
partuuid = '/dev/disk/by-partuuid/{}'.format(
uuid)
dev_path = os.path.realpath(partuuid)
osd_device = dev_path.split('/')[-1]
else:
encrypted = 0
osd_device = dev_path.split('/')[-1]
elif osd_type == 'bluestore':
block_link = os.path.join(path_name, 'block')
osd_path = os.path.realpath(block_link)
osd_device = osd_path.split('/')[-1]
encrypted = 0
else:
raise ValueError("Unknown OSD type encountered")
# if the osd_id hasn't been seem neither has the
# disk
self.osd[osd_device] = Disk(osd_device,
path_name=path_name,
osd_id=osd_id,
in_osd_type=osd_type,
encrypted=encrypted)
self.dev_lookup[osd_device] = 'osd'
self.osd_count += 1
if osd_type == 'filestore':
journal_link = os.path.join(path_name, 'journal')
else:
journal_link = os.path.join(path_name, 'block.wal')
if os.path.exists(journal_link):
link_tgt = os.readlink(journal_link)
if link_tgt.startswith('/dev/mapper'):
encrypted = 1
else:
encrypted = 0
partuuid_path = os.path.join(
'/dev/disk/by-partuuid',
link_tgt.split('/')[-1])
jrnl_path = os.path.realpath(partuuid_path)
jrnl_dev = jrnl_path.split('/')[-1]
if jrnl_dev not in self.osd:
self.jrnl[jrnl_dev] = Disk(
jrnl_dev,
osd_id=osd_id,
in_osd_type=osd_type,
encrypted=encrypted)
self.dev_lookup[jrnl_dev] = 'jrnl'
else:
# No journal or WAL link..?
pass