ceph-volume scans each host in the cluster from time to time in order
to determine which devices are present and whether they are eligible to be
used as OSDs.
To print a list of devices discovered by
cephadm, run this command:
ceph orch device ls [--hostname=...] [--wide] [--refresh]
Hostname Path Type Serial Size Health Ident Fault Available srv-01 /dev/sdb hdd 15P0A0YFFRD6 300G Unknown N/A N/A No srv-01 /dev/sdc hdd 15R0A08WFRD6 300G Unknown N/A N/A No srv-01 /dev/sdd hdd 15R0A07DFRD6 300G Unknown N/A N/A No srv-01 /dev/sde hdd 15P0A0QDFRD6 300G Unknown N/A N/A No srv-02 /dev/sdb hdd 15R0A033FRD6 300G Unknown N/A N/A No srv-02 /dev/sdc hdd 15R0A05XFRD6 300G Unknown N/A N/A No srv-02 /dev/sde hdd 15R0A0ANFRD6 300G Unknown N/A N/A No srv-02 /dev/sdf hdd 15R0A06EFRD6 300G Unknown N/A N/A No srv-03 /dev/sdb hdd 15R0A0OGFRD6 300G Unknown N/A N/A No srv-03 /dev/sdc hdd 15R0A0P7FRD6 300G Unknown N/A N/A No srv-03 /dev/sdd hdd 15R0A0O7FRD6 300G Unknown N/A N/A No
--wide option provides all details relating to the device,
including any reasons that the device might not be eligible for use as an OSD.
In the above example you can see fields named “Health”, “Ident”, and “Fault”.
This information is provided by integration with libstoragemgmt. By default,
this integration is disabled (because libstoragemgmt may not be 100%
compatible with your hardware). To make
cephadm include these fields,
enable cephadm’s “enhanced device scan” option as follows;
ceph config set mgr mgr/cephadm/device_enhanced_scan true
Although the libstoragemgmt library performs standard SCSI inquiry calls, there is no guarantee that your firmware fully implements these standards. This can lead to erratic behaviour and even bus resets on some older hardware. It is therefore recommended that, before enabling this feature, you test your hardware’s compatibility with libstoragemgmt first to avoid unplanned interruptions to services.
There are a number of ways to test compatibility, but the simplest may be
to use the cephadm shell to call libstoragemgmt directly -
lsmcli ldl. If your hardware is supported you should see something like
Path | SCSI VPD 0x83 | Link Type | Serial Number | Health Status ---------------------------------------------------------------------------- /dev/sda | 50000396082ba631 | SAS | 15P0A0R0FRD6 | Good /dev/sdb | 50000396082bbbf9 | SAS | 15P0A0YFFRD6 | Good
After you have enabled libstoragemgmt support, the output will look something like this:
# ceph orch device ls Hostname Path Type Serial Size Health Ident Fault Available srv-01 /dev/sdb hdd 15P0A0YFFRD6 300G Good Off Off No srv-01 /dev/sdc hdd 15R0A08WFRD6 300G Good Off Off No :
In this example, libstoragemgmt has confirmed the health of the drives and the ability to interact with the Identification and Fault LEDs on the drive enclosures. For further information about interacting with these LEDs, refer to device management.
The current release of libstoragemgmt (1.8.8) supports SCSI, SAS, and SATA based local disks only. There is no official support for NVMe devices (PCIe)
Listing Storage Devices¶
In order to deploy an OSD, there must be a storage device that is available on which the OSD will be deployed.
Run this command to display an inventory of storage devices on all cluster hosts:
ceph orch device ls
A storage device is considered available if all of the following conditions are met:
The device must have no partitions.
The device must not have any LVM state.
The device must not be mounted.
The device must not contain a file system.
The device must not contain a Ceph BlueStore OSD.
The device must be larger than 5 GB.
Ceph will not provision an OSD on a device that is not available.
Creating New OSDs¶
There are a few ways to create new OSDs:
Tell Ceph to consume any available and unused storage device:
ceph orch apply osd --all-available-devices
Create an OSD from a specific device on a specific host:
ceph orch daemon add osd *<host>*:*<device-path>*
ceph orch daemon add osd host1:/dev/sdb
You can use Advanced OSD Service Specifications to categorize device(s) based on their properties. This might be useful in forming a clearer picture of which devices are available to consume. Properties include device type (SSD or HDD), device model names, size, and the hosts on which the devices exist:
ceph orch apply -i spec.yml
--dry-run flag causes the orchestrator to present a preview of what
will happen without actually creating the OSDs.
ceph orch apply osd --all-available-devices --dry-runNAME HOST DATA DB WAL all-available-devices node1 /dev/vdb - - all-available-devices node2 /dev/vdc - - all-available-devices node3 /dev/vdd - -
The effect of
ceph orch apply is persistent. This means that drives that
are added to the system after the
ceph orch apply command completes will be
automatically found and added to the cluster. It also means that drives that
become available (by zapping, for example) after the
ceph orch apply
command completes will be automatically found and added to the cluster.
We will examine the effects of the following command:
ceph orch apply osd --all-available-devices
After running the above command:
If you add new disks to the cluster, they will automatically be used to create new OSDs.
If you remove an OSD and clean the LVM physical volume, a new OSD will be created automatically.
To disable the automatic creation of OSD on available devices, use the
If you want to avoid this behavior (disable automatic creation of OSD on available devices), use the
ceph orch apply osd --all-available-devices --unmanaged=true
Keep these three facts in mind:
The default behavior of
ceph orch applycauses cephadm constantly to reconcile. This means that cephadm creates OSDs as soon as new drives are detected.
unmanaged: Truedisables the creation of OSDs. If
unmanaged: Trueis set, nothing will happen even if you apply a new OSD service.
ceph orch daemon addcreates OSDs, but does not add an OSD service.
For cephadm, see also Disabling automatic deployment of daemons.
Remove an OSD¶
Removing an OSD from a cluster involves two steps:
evacuating all placement groups (PGs) from the cluster
removing the PG-free OSD from the cluster
The following command performs these two steps:
ceph orch osd rm <osd_id(s)> [--replace] [--force]
ceph orch osd rm 0
Scheduled OSD(s) for removal
OSDs that are not safe to destroy will be rejected.
Monitoring OSD State¶
You can query the state of OSD operation with the following command:
ceph orch osd rm status
OSD_ID HOST STATE PG_COUNT REPLACE FORCE STARTED_AT 2 cephadm-dev done, waiting for purge 0 True False 2020-07-17 13:01:43.147684 3 cephadm-dev draining 17 False True 2020-07-17 13:01:45.162158 4 cephadm-dev started 42 False True 2020-07-17 13:01:45.162158
When no PGs are left on the OSD, it will be decommissioned and removed from the cluster.
After removing an OSD, if you wipe the LVM physical volume in the device used by the removed OSD, a new OSD will be created.
For more information on this, read about the
unmanaged parameter in Declarative State.
Stopping OSD Removal¶
It is possible to stop queued OSD removals by using the following command:
ceph orch osd rm stop <osd_id(s)>
ceph orch osd rm stop 4
Stopped OSD(s) removal
This resets the initial state of the OSD and takes it off the removal queue.
Replacing an OSD¶
orch osd rm <osd_id(s)> --replace [--force]
ceph orch osd rm 4 --replace
Scheduled OSD(s) for replacement
This follows the same procedure as the procedure in the “Remove OSD” section, with one exception: the OSD is not permanently removed from the CRUSH hierarchy, but is instead assigned a ‘destroyed’ flag.
The new OSD that will replace the removed OSD must be created on the same host as the OSD that was removed.
Preserving the OSD ID
The ‘destroyed’ flag is used to determine which OSD ids will be reused in the next OSD deployment.
If you use OSDSpecs for OSD deployment, your newly added disks will be assigned the OSD ids of their replaced counterparts. This assumes that the new disks still match the OSDSpecs.
--dry-run flag to make certain that the
ceph orch apply osd
command does what you want it to. The
--dry-run flag shows you what the
outcome of the command will be without making the changes you specify. When
you are satisfied that the command will do what you want, run the command
The name of your OSDSpec can be retrieved with the command
ceph orch ls
Alternatively, you can use your OSDSpec file:
ceph orch apply osd -i <osd_spec_file> --dry-run
NAME HOST DATA DB WAL <name_of_osd_spec> node1 /dev/vdb - -
When this output reflects your intention, omit the
--dry-run flag to
execute the deployment.
Erasing Devices (Zapping Devices)¶
Erase (zap) a device so that it can be reused.
zap on the remote host.
ceph orch device zap <hostname> <path>
ceph orch device zap my_hostname /dev/sdx
If the unmanaged flag is unset, cephadm automatically deploys drives that
match the OSDSpec. For example, if you use the
all-available-devices option when creating OSDs, when you
device the cephadm orchestrator automatically creates a new OSD in the
device. To disable this behavior, see Declarative State.
Automatically tuning OSD memory¶
OSD daemons will adjust their memory consumption based on the
osd_memory_target config option (several gigabytes, by
default). If Ceph is deployed on dedicated nodes that are not sharing
memory with other services, cephadm can automatically adjust the per-OSD
memory consumption based on the total amount of RAM and the number of deployed
This option is enabled globally with:
ceph config set osd osd_memory_target_autotune true
Cephadm will start with a fraction
mgr/cephadm/autotune_memory_target_ratio, which defaults to
.7) of the total RAM in the system, subtract off any memory
consumed by non-autotuned daemons (non-OSDs, for OSDs for which
osd_memory_target_autotune is false), and then divide by the
The final targets are reflected in the config database with options like:
WHO MASK LEVEL OPTION VALUE osd host:foo basic osd_memory_target 126092301926 osd host:bar basic osd_memory_target 6442450944
Both the limits and the current memory consumed by each daemon are visible from
ceph orch ps output in the
MEM LIMIT column:
NAME HOST PORTS STATUS REFRESHED AGE MEM USED MEM LIMIT VERSION IMAGE ID CONTAINER ID osd.1 dael running (3h) 10s ago 3h 72857k 117.4G 17.0.0-3781-gafaed750 7015fda3cd67 9e183363d39c osd.2 dael running (81m) 10s ago 81m 63989k 117.4G 17.0.0-3781-gafaed750 7015fda3cd67 1f0cc479b051 osd.3 dael running (62m) 10s ago 62m 64071k 117.4G 17.0.0-3781-gafaed750 7015fda3cd67 ac5537492f27
To exclude an OSD from memory autotuning, disable the autotune option for that OSD and also set a specific memory target. For example,
ceph config set osd.123 osd_memory_target_autotune false ceph config set osd.123 osd_memory_target 16G
Advanced OSD Service Specifications¶
Service Specifications of type
osd are a way to describe a
cluster layout, using the properties of disks. Service specifications give the
user an abstract way to tell Ceph which disks should turn into OSDs with which
configurations, without knowing the specifics of device names and paths.
Service specifications make it possible to define a yaml or json file that can be used to reduce the amount of manual work involved in creating OSDs.
For example, instead of running the following command:
ceph orch daemon add osd *<host>*:*<path-to-device>*
for each device and each host, we can define a yaml or json file that allows us to describe the layout. Here’s the most basic example.
Create a file called (for example)
service_type: osd service_id: default_drive_group # custom name of the osd spec placement: host_pattern: '*' # which hosts to target spec: data_devices: # the type of devices you are applying specs to all: true # a filter, check below for a full list
This means :
Turn any available device (ceph-volume decides what ‘available’ is) into an OSD on all hosts that match the glob pattern ‘*’. (The glob pattern matches against the registered hosts from host ls) A more detailed section on host_pattern is available below.
Then pass it to osd create like this:
ceph orch apply osd -i /path/to/osd_spec.yml
This instruction will be issued to all the matching hosts, and will deploy these OSDs.
Setups more complex than the one specified by the
allfilter are possible. See Filters for details.
--dry-runflag can be passed to the
apply osdcommand to display a synopsis of the proposed layout.
ceph orch apply osd -i /path/to/osd_spec.yml --dry-run
Filters are applied using an AND gate by default. This means that a drive
must fulfill all filter criteria in order to get selected. This behavior can
be adjusted by setting
filter_logic: OR in the OSD specification.
Filters are used to assign disks to groups, using their attributes to group them.
The attributes are based off of ceph-volume’s disk query. You can retrieve information about the attributes with this command:
ceph-volume inventory </path/to/disk>
Vendor or Model¶
Specific disks can be targeted by vendor or model:
Specific disks can be targeted by Size:
Size specifications can be of the following forms:
To include disks of an exact size
To include disks within a given range of size:
To include disks that are less than or equal to 10G in size:
To include disks equal to or greater than 40G in size:
Sizes don’t have to be specified exclusively in Gigabytes(G).
Other units of size are supported: Megabyte(M), Gigabyte(G) and Terrabyte(T).
Appending the (B) for byte is also supported:
This operates on the ‘rotational’ attribute of the disk.
rotational: 0 | 1
1 to match all disks that are rotational
0 to match all disks that are non-rotational (SSD, NVME etc)
This will take all disks that are ‘available’
This is exclusive for the data_devices section.
If you have specified some valid filters but want to limit the number of disks that they match, use the
For example, if you used vendor to match all disks that are from VendorA but want to use only the first two, you could use limit:
data_devices: vendor: VendorA limit: 2
limit is a last resort and shouldn’t be used if it can be avoided.
There are multiple optional settings you can use to change the way OSDs are deployed. You can add these options to the base level of an OSD spec for it to take effect.
This example would deploy all OSDs with encryption enabled.
service_type: osd service_id: example_osd_spec placement: host_pattern: '*' spec: data_devices: all: true encrypted: true
See a full list in the DriveGroupSpecs
DriveGroupSpec(*args: Any, **kwargs: Any)¶
Describe a drive group in the same form that ceph-volume understands.
block_db_size: Optional[Union[int, str]]¶
Set (or override) the “bluestore_block_db_size” value, in bytes
block_wal_size: Optional[Union[int, str]]¶
Set (or override) the “bluestore_block_wal_size” value, in bytes
A list of strings, containing paths which should back OSDs
How many OSDs per DB device
The logic gate we use to match disks with filters. defaults to ‘AND’
journal_size: Optional[Union[int, str]]¶
set journal_size in bytes
Optional: mapping of host -> List of osd_ids that should be replaced See OSD Replacement
Number of osd daemons per “DATA” device. To fully utilize nvme devices multiple osds are required.
If this should be treated as a ‘preview’ spec
How many OSDs per WAL device
The simple case¶
All nodes with the same setup
20 HDDs Vendor: VendorA Model: HDD-123-foo Size: 4TB 2 SSDs Vendor: VendorB Model: MC-55-44-ZX Size: 512GB
This is a common setup and can be described quite easily:
service_type: osd service_id: osd_spec_default placement: host_pattern: '*' spec: data_devices: model: HDD-123-foo # Note, HDD-123 would also be valid db_devices: model: MC-55-44-XZ # Same here, MC-55-44 is valid
However, we can improve it by reducing the filters on core properties of the drives:
service_type: osd service_id: osd_spec_default placement: host_pattern: '*' spec: data_devices: rotational: 1 db_devices: rotational: 0
Now, we enforce all rotating devices to be declared as ‘data devices’ and all non-rotating devices will be used as shared_devices (wal, db)
If you know that drives with more than 2TB will always be the slower data devices, you can also filter by size:
service_type: osd service_id: osd_spec_default placement: host_pattern: '*' spec: data_devices: size: '2TB:' db_devices: size: ':2TB'
All of the above OSD specs are equally valid. Which of those you want to use depends on taste and on how much you expect your node layout to change.
Multiple OSD specs for a single host¶
Here we have two distinct setups
20 HDDs Vendor: VendorA Model: HDD-123-foo Size: 4TB 12 SSDs Vendor: VendorB Model: MC-55-44-ZX Size: 512GB 2 NVMEs Vendor: VendorC Model: NVME-QQQQ-987 Size: 256GB
20 HDDs should share 2 SSDs
10 SSDs should share 2 NVMes
This can be described with two layouts.
service_type: osd service_id: osd_spec_hdd placement: host_pattern: '*' spec: data_devices: rotational: 0 db_devices: model: MC-55-44-XZ limit: 2 # db_slots is actually to be favoured here, but it's not implemented yet --- service_type: osd service_id: osd_spec_ssd placement: host_pattern: '*' spec: data_devices: model: MC-55-44-XZ db_devices: vendor: VendorC
This would create the desired layout by using all HDDs as data_devices with two SSD assigned as dedicated db/wal devices. The remaining SSDs(8) will be data_devices that have the ‘VendorC’ NVMEs assigned as dedicated db/wal devices.
Multiple hosts with the same disk layout¶
Assuming the cluster has different kinds of hosts each with similar disk layout, it is recommended to apply different OSD specs matching only one set of hosts. Typically you will have a spec for multiple hosts with the same layout.
The sevice id as the unique key: In case a new OSD spec with an already applied service id is applied, the existing OSD spec will be superseeded. cephadm will now create new OSD daemons based on the new spec definition. Existing OSD daemons will not be affected. See Declarative State.
20 HDDs Vendor: Intel Model: SSD-123-foo Size: 4TB 2 SSDs Vendor: VendorA Model: MC-55-44-ZX Size: 512GB
5 NVMEs Vendor: Intel Model: SSD-123-foo Size: 4TB 20 SSDs Vendor: VendorA Model: MC-55-44-ZX Size: 512GB
You can use the ‘placement’ key in the layout to target certain nodes.
service_type: osd service_id: disk_layout_a placement: label: disk_layout_a spec: data_devices: rotational: 1 db_devices: rotational: 0 --- service_type: osd service_id: disk_layout_b placement: label: disk_layout_b spec: data_devices: model: MC-55-44-XZ db_devices: model: SSD-123-foo
This applies different OSD specs to different hosts depending on the placement key. See Daemon Placement
Assuming each host has a unique disk layout, each OSD spec needs to have a different service id
Dedicated wal + db¶
All previous cases co-located the WALs with the DBs. It’s however possible to deploy the WAL on a dedicated device as well, if it makes sense.
20 HDDs Vendor: VendorA Model: SSD-123-foo Size: 4TB 2 SSDs Vendor: VendorB Model: MC-55-44-ZX Size: 512GB 2 NVMEs Vendor: VendorC Model: NVME-QQQQ-987 Size: 256GB
The OSD spec for this case would look like the following (using the model filter):
service_type: osd service_id: osd_spec_default placement: host_pattern: '*' spec: data_devices: model: MC-55-44-XZ db_devices: model: SSD-123-foo wal_devices: model: NVME-QQQQ-987
It is also possible to specify directly device paths in specific hosts like the following:
service_type: osd service_id: osd_using_paths placement: hosts: - Node01 - Node02 spec: data_devices: paths: - /dev/sdb db_devices: paths: - /dev/sdc wal_devices: paths: - /dev/sdd
This can easily be done with other filters, like size or vendor as well.
Activate existing OSDs¶
In case the OS of a host was reinstalled, existing OSDs need to be activated again. For this use case, cephadm provides a wrapper for activate that activates all existing OSDs on a host.
ceph cephadm osd activate <host>...
This will scan all existing disks for OSDs and deploy corresponding daemons.