juju deploy ceph-osd
Discuss this charm
Share your thoughts on this charm with the community on discourse.
Ceph is a unified, distributed storage system designed for excellent performance, reliability, and scalability.
The ceph-osd charm deploys the Ceph object storage daemon (OSD) and manages its volumes. It is used in conjunction with the ceph-mon charm. Together, these charms can scale out the amount of storage available in a Ceph cluster.
This section covers common and/or important configuration options. See file
config.yaml for the full list of options, along with their descriptions and
default values. A YAML file (e.g.
ceph-osd.yaml) is often used to store
configuration options. See the Juju documentation for
details on configuring applications.
customize-failure-domain option determines how a Ceph CRUSH map is
A value of 'false' (the default) will lead to a map that will replicate data across hosts (implemented as Ceph bucket type 'host'). With a value of 'true' all MAAS-defined zones will be used to generate a map that will replicate data across Ceph availability zones (implemented as bucket type 'rack').
This option is also supported by the ceph-mon charm. Its value must be the same for both charms.
osd-devices option lists what block devices can be used for OSDs across
the cluster. See section 'Storage devices' for an elaboration on this
source option states the software sources. A common value is an OpenStack
UCA release (e.g. 'cloud:xenial-queens' or 'cloud:bionic-ussuri'). See Ceph
and the UCA. The underlying host's existing apt sources
will be used if this option is not specified (this behaviour can be explicitly
chosen by using the value of 'distro').
A storage device is destined as an OSD (Object Storage Device). There can be multiple OSDs per storage node (ceph-osd unit).
The list of all possible storage devices for the cluster is defined by the
osd-devices option (default value is '/dev/vdb'). The following examples can
be used in the
ceph-osd.yaml configuration file:
- Block devices (regular)
ceph-osd: options: osd-devices: /dev/vdb /dev/vdc /dev/vdd
Each regular block device must be an absolute path to a device node.
- Block devices (Juju storage)
ceph-osd: storage: osd-devices: cinder,20G
See the Juju documentation for guidance on implementing Juju storage.
- Directory-backed OSDs
ceph-osd: storage: osd-devices: /var/tmp/osd-1
Note: OSD directories can no longer be created starting with Ceph Nautilus. Existing OSD directories will continue to function after an upgrade to Nautilus.
The list defined by option
osd-devices may affect newly added ceph-osd units
as well as existing units (the option may be modified after units have been
added). The charm will attempt to activate as Ceph storage any listed device
that is visible by the unit's underlying machine. To prevent the activation of
volumes on existing units the
blacklist-add-disk action may be used.
The configuration option is modified in the usual way. For instance, to have it consist solely of devices '/dev/sdb' and '/dev/sdc':
juju config ceph-osd osd-devices='/dev/sdb /dev/sdc'
The charm will go into a blocked state (visible in
juju status output) if it
detects pre-existing data on a device. In this case the operator can either
instruct the charm to ignore the disk (action
blacklist-add-disk) or to have
it purge all data on the disk (action
Important: The recommended minimum number of OSDs in the cluster is three and this is what the ceph-mon charm expects (the cluster will not form with a lesser number). See option
expected-osd-countin the ceph-mon charm to overcome this but beware that going below three is not a supported configuration.
A cloud with three MON nodes is a typical design whereas three OSDs are considered the minimum. For example, to deploy a Ceph cluster consisting of three OSDs (one per ceph-osd unit) and three MONs:
juju deploy -n 3 --config ceph-osd.yaml ceph-osd juju deploy -n 3 --to lxd:0,lxd:1,lxd:2 ceph-mon juju add-relation ceph-osd:mon ceph-mon:osd
Here, a containerised MON is running alongside each storage node. We've assumed that the machines spawned in the first command are assigned IDs of 0, 1, and 2.
Note: Refer to the Install OpenStack page in the OpenStack Charms Deployment Guide for instructions on installing the ceph-osd application for use with OpenStack.
For each ceph-osd unit, the ceph-osd charm will scan for all the devices
configured via the
osd-devices option and attempt to assign to it all of the
ones it finds. The cluster's initial pool of available storage is the "sum" of
all these assigned devices.
This charm supports the use of Juju network spaces (Juju
v.2.0). This feature optionally allows specific types of the application's
network traffic to be bound to subnets that the underlying hardware is
Note: Spaces must be configured in the backing cloud prior to deployment.
The ceph-osd charm exposes the following Ceph traffic types (bindings):
- 'public' (front-side)
- 'cluster' (back-side)
For example, providing that spaces 'data-space' and 'cluster-space' exist, the deploy command above could look like this:
juju deploy --config ceph-osd.yaml -n 3 ceph-osd \ --bind "public=data-space cluster=cluster-space"
Alternatively, configuration can be provided as part of a bundle:
ceph-osd: charm: cs:ceph-osd num_units: 1 bindings: public: data-space cluster: cluster-space
Refer to the Ceph Network Reference to learn about the implications of segregating Ceph network traffic.
Note: Existing ceph-osd units configured with the
ceph-cluster-networkoptions will continue to honour them. Furthermore, these options override any space bindings, if set.
Although AppArmor is not enabled for Ceph by default, an AppArmor profile can
be generated by the charm by assigning a value of 'complain', 'enforce', or
'disable' (the default) to option
Caution: Enabling an AppArmor profile is disruptive to a running Ceph cluster as all ceph-osd processes must be restarted.
The new profile has a narrow supported use case, and it should always be verified in pre-production against the specific configurations and topologies intended for production.
The profiles generated by the charm should not be used in the following scenarios:
- On any version of Ubuntu older than 16.04
- On any version of Ceph older than Luminous
- When OSD journal devices are in use
- When Ceph BlueStore is enabled
Block device encryption
The ceph-osd charm supports encryption for OSD volumes that are backed by block
devices. To use Ceph's native key management framework, available since Ceph
Jewel, set option
osd-encrypt for the ceph-osd charm:
ceph-osd: options: osd-encrypt: True
Here, dm-crypt keys are stored in the MON sub-cluster.
Alternatively, since Ceph Luminous, encryption keys can be stored in Vault,
which is deployed and initialised via the vault charm. Set
osd-encrypt-keymanager for the ceph-osd charm:
ceph-osd: options: osd-encrypt: True osd-encrypt-keymanager: vault
Important: Post deployment configuration will only affect block devices associated with new ceph-osd units.
This section covers Juju actions supported by the charm.
Actions allow specific operations to be performed on a per-unit basis. To
display action descriptions run
juju actions ceph-osd. If the charm is not
deployed then see file
Working with OSDs
Set OSDs to 'out'
osd-out action to set all OSD volumes on a unit to 'out'.
Warning: This action has the potential of impacting your cluster significantly. The Ceph documentation on this topic is considered essential reading.
osd-out action sets all OSDs on the unit as 'out'. Unless the cluster
itself is set to 'noout' this action will cause Ceph to rebalance data by
migrating PGs out of the unit's OSDs and onto OSDs available on other units.
The impact is twofold:
- The available space on the remaining OSDs is reduced. Not only is there less space for future workloads but there is a danger of exceeding the cluster's storage capacity.
- The traffic and CPU load on the cluster is increased.
Note: It has been reported that setting OSDs as 'out' may cause some PGs to get stuck in the 'active+remapped' state. This is an upstream issue.
The ceph-mon charm has an action called
set-noout that sets
'noout' for the cluster.
It may be perfectly fine to have data rebalanced. The decisive factor is whether the OSDs are being paused temporarily (e.g. the underlying machine is scheduled for maintenance) or whether they are being removed from the cluster completely (e.g. the storage hardware is reaching EOL).
juju run-action --wait ceph-osd/4 osd-out
Set OSDs to 'in'
osd-in action to set all OSD volumes on a unit to 'in'.
osd-in action is reciprocal to the
osd-out action. The OSDs are set to
'in'. It is typically used when the
osd-out action was used in conjunction
with the cluster 'noout' flag.
juju run-action --wait ceph-osd/4 osd-in
Working with disks
list-disks action to list disks known to a unit.
The action lists the unit's block devices by categorising them in three ways:
disks: visible (known by udev), unused (not mounted), and not designated as an OSD journal (via the
disksbut blacklisted (see action
disksbut not eligible for use due to the presence of existing data
juju run-action --wait ceph-osd/4 list-disks
Add a disk
add-disk action to add a disk to a unit.
A ceph-osd unit is automatically assigned OSD volumes based on the current
value of the
osd-devices application option. The
add-disk action allows the
operator to manually add OSD volumes (for disks that are not listed by
osd-devices) to an existing unit.