-.. _ceph-volume-lvm-batch_bluestore:
-
-``bluestore``
--------------
-The :term:`bluestore` objectstore (the default) is used when creating multiple OSDs
-with the ``batch`` sub-command. It allows a few different scenarios depending
-on the input of devices:
+Automatic sorting of disks
+--------------------------
+If ``batch`` receives only a single list of data devices and the ``--no-auto`` option
+ is *not* passed (currently the default), ``ceph-volume`` will auto-sort disks by its rotational
+ property and use non-rotating disks for ``block.db`` or ``journal`` depending
+ on the objectstore used.
+This default behavior is now DEPRECATED and will be removed in future releases. Instead
+ an ``auto`` option is introduced to retain this behavior.
+It is recommended to make use of the explicit device lists for ``block.db``,
+ ``block.wal`` and ``journal``.
+For example assuming :term:`bluestore` is used and ``--no-auto`` is not passed,
+ the deprecated behavior would deploy the following, depending on the devices
+ passed:
#. Devices are all spinning HDDs: 1 OSD is created per device
#. Devices are all SSDs: 2 OSDs are created per device
def ensure_disjoint_device_lists(data, db=[], wal=[], journal=[]):
# check that all device lists are disjoint with each other
- if not(set(data).isdisjoint(set(db)) and
- set(data).isdisjoint(set(wal)) and
- set(data).isdisjoint(set(journal)) and
- set(db).isdisjoint(set(wal))):
+ if not all([set(data).isdisjoint(set(db)),
+ set(data).isdisjoint(set(wal)),
+ set(data).isdisjoint(set(journal)),
+ set(db).isdisjoint(set(wal))]):
raise Exception('Device lists are not disjoint')
+def separate_devices_from_lvs(devices):
+ phys = []
+ lvm = []
+ for d in devices:
+ phys.append(d) if d.is_device else lvm.append(d)
+ return phys, lvm
+
+
def get_physical_osds(devices, args):
'''
Goes through passed physical devices and assigns OSDs
dev_size = dev.vg_size[0]
abs_size = disk.Size(b=int(dev_size * rel_data_size))
free_size = dev.vg_free[0]
- if abs_size < 419430400:
- mlogger.error('Data LV on {} would be too small (<400M)'.format(dev.path))
- continue
for _ in range(args.osds_per_device):
if abs_size > free_size:
break
if args.osd_ids:
osd_id = args.osd_ids.pop()
ret.append(Batch.OSD(dev.path,
- rel_data_size,
- abs_size,
- args.osds_per_device,
- osd_id))
+ rel_data_size,
+ abs_size,
+ args.osds_per_device,
+ osd_id))
return ret
if args.osd_ids:
osd_id = args.osd_ids.pop()
osd = Batch.OSD("{}/{}".format(lv.vg_name, lv.lv_name),
- 100.0,
- disk.Size(b=int(lv.lv_size)),
- 1,
- osd_id)
+ 100.0,
+ disk.Size(b=int(lv.lv_size)),
+ 1,
+ osd_id)
ret.append(osd)
return ret
dev_size = dev.vg_size[0]
abs_size = disk.Size(b=int(dev_size / requested_slots))
free_size = dev.vg_free[0]
- # if abs_size < 419430400:
- # mlogger.error('{} LV on {} would be too small (<400M)'.format(
- # type_, dev.path))
- # continue
relative_size = int(abs_size) / dev_size
if requested_size:
if requested_size <= abs_size:
help='Devices to provision OSDs journal volumes',
)
parser.add_argument(
- '--no-auto',
+ '--auto',
action='store_true',
+ help=('deploy multi-device OSDs if rotational and non-rotational drives '
+ 'are passed in DEVICES'),
+ default=True
+ )
+ parser.add_argument(
+ '--no-auto',
+ action='store_false',
+ dest='auto',
help=('deploy standalone OSDs if rotational and non-rotational drives '
'are passed in DEVICES'),
)
if self.args.data_slots and self.args.osds_per_device:
if self.args.data_slots < self.args.osds_per_device:
raise ValueError('data_slots is smaller then osds_per_device')
- # TODO this needs more thought.
- # for slot in ['block_db_slots', 'block_wal_slots', 'journal_slots']:
- # slot_value = getattr(self.args, slot, None)
- # if slot_value:
- # if slot_value < len(self.args.devices):
- # raise ValueError('{} is smaller then osds_per_device')
def _sort_rotational_disks(self):
'''
'''
mlogger.warning('DEPRECATION NOTICE')
mlogger.warning('You are using the legacy automatic disk sorting behavior')
- mlogger.warning('The Pacific release will change the default to --notauto')
+ mlogger.warning('The Pacific release will change the default to --no-auto')
rotating = []
ssd = []
for d in self.args.devices:
- if d.rotational:
- rotating.append(d)
- else:
- ssd.append(d)
+ rotating.append(d) if d.rotational else ssd.append(d)
self.args.devices = rotating
if self.args.filestore:
self.args.journal_devices = ssd
if not self.args.bluestore and not self.args.filestore:
self.args.bluestore = True
- if (not self.args.no_auto and not self.args.db_devices and not
+ if (self.args.auto and not self.args.db_devices and not
self.args.wal_devices and not self.args.journal_devices):
self._sort_rotational_disks()
very_fast_devices=[]):
'''
The methods here are mostly just organization, error reporting and
- setting up of (default) args. The hravy lifting code for the deployment
- layout can be found in the static get_*_osds and get_*:fast_allocs
+ setting up of (default) args. The heavy lifting code for the deployment
+ layout can be found in the static get_*_osds and get_*_fast_allocs
functions.
'''
plan = []
- phys_devs = [d for d in devices if d.is_device]
- lvm_devs = [d.lvs[0] for d in list(set(devices) -
- set(phys_devs))]
+ phys_devs, lvm_devs = separate_devices_from_lvs(devices)
mlogger.debug(('passed data devices: {} physical,'
' {} LVM').format(len(phys_devs), len(lvm_devs)))
type_=fast_type)
if very_fast_devices and args.bluestore:
osd.add_very_fast_device(*very_fast_allocations.pop(),
- type_='block.wal')
+ type_='block.wal')
return plan
def fast_allocations(self, devices, requested_osds, new_osds, type_):
ret = []
if not devices:
return ret
- phys_devs = [d for d in devices if d.is_device]
- lvm_devs = [d.lvs[0] for d in list(set(devices) -
- set(phys_devs))]
+ phys_devs, lvm_devs = separate_devices_from_lvs(devices)
mlogger.debug(('passed {} devices: {} physical,'
' {} LVM').format(type_, len(phys_devs), len(lvm_devs)))
projects / ceph.git / commitdiff