ceph-mgr administrator’s guide

Manual setup

Usually, you would set up a ceph-mgr daemon using a tool such as ceph-ansible. These instructions describe how to set up a ceph-mgr daemon manually.

First, create an authentication key for your daemon:

ceph auth get-or-create mgr.$name mon 'allow profile mgr' osd 'allow *' mds 'allow *'

Place that key as file named keyring into mgr data path, which for a cluster “ceph” and mgr $name “foo” would be /var/lib/ceph/mgr/ceph-foo respective /var/lib/ceph/mgr/ceph-foo/keyring.

Start the ceph-mgr daemon:

ceph-mgr -i $name

Check that the mgr has come up by looking at the output of ceph status, which should now include a mgr status line:

mgr active: $name

Client authentication

The manager is a new daemon which requires new CephX capabilities. If you upgrade a cluster from an old version of Ceph, or use the default install/deploy tools, your admin client should get this capability automatically. If you use tooling from elsewhere, you may get EACCES errors when invoking certain ceph cluster commands. To fix that, add a “mgr allow *” stanza to your client’s cephx capabilities by Modifying User Capabilities.

High availability

In general, you should set up a ceph-mgr on each of the hosts running a ceph-mon daemon to achieve the same level of availability.

By default, whichever ceph-mgr instance comes up first will be made active by the monitors, and the others will be standbys. There is no requirement for quorum among the ceph-mgr daemons.

If the active daemon fails to send a beacon to the monitors for more than mon_mgr_beacon_grace, then it will be replaced by a standby.

If you want to preempt failover, you can explicitly mark a ceph-mgr daemon as failed using ceph mgr fail <mgr name>.

Performance and Scalability

All the mgr modules share a cache that can be enabled with ceph config set mgr mgr_ttl_cache_expire_seconds <seconds>, where seconds is the time to live of the cached python objects.

It is recommended to enable the cache with a 10 seconds TTL when there are 500+ osds or 10k+ pgs as internal structures might increase in size, and cause latency issues when requesting large structures. As an example, an OSDMap with 1000 osds has a aproximate size of 4MiB. With heavy load, on a 3000 osd cluster there has been a 1.5x improvement enabling the cache.

Furthermore, you can run ceph daemon mgr.${MGRNAME} perf dump to retrieve perf counters of a mgr module. In mgr.cache_hit and mgr.cache_miss you’ll find the hit/miss ratio of the mgr cache.

Using modules

Use the command ceph mgr module ls to see which modules are available, and which are currently enabled. Use ceph mgr module ls --format=json-pretty to view detailed metadata about disabled modules. Enable or disable modules using the commands ceph mgr module enable <module> and ceph mgr module disable <module> respectively.

If a module is enabled then the active ceph-mgr daemon will load and execute it. In the case of modules that provide a service, such as an HTTP server, the module may publish its address when it is loaded. To see the addresses of such modules, use the command ceph mgr services.

Some modules may also implement a special standby mode which runs on standby ceph-mgr daemons as well as the active daemon. This enables modules that provide services to redirect their clients to the active daemon, if the client tries to connect to a standby.

Consult the documentation pages for individual manager modules for more information about what functionality each module provides.

Here is an example of enabling the Dashboard module:

$ ceph mgr module ls
{
        "enabled_modules": [
                "restful",
                "status"
        ],
        "disabled_modules": [
                "dashboard"
        ]
}

$ ceph mgr module enable dashboard
$ ceph mgr module ls
{
        "enabled_modules": [
                "restful",
                "status",
                "dashboard"
        ],
        "disabled_modules": [
        ]
}

$ ceph mgr services
{
        "dashboard": "http://myserver.com:7789/",
        "restful": "https://myserver.com:8789/"
}

The first time the cluster starts, it uses the mgr_initial_modules setting to override which modules to enable. However, this setting is ignored through the rest of the lifetime of the cluster: only use it for bootstrapping. For example, before starting your monitor daemons for the first time, you might add a section like this to your ceph.conf:

[mon]
    mgr_initial_modules = dashboard balancer

Module Pool

The manager creates a pool for use by its module to store state. The name of this pool is .mgr (with the leading . indicating a reserved pool name).

Note

Prior to Quincy, the devicehealth module created a device_health_metrics pool to store device SMART statistics. With Quincy, this pool is automatically renamed to be the common manager module pool.

Calling module commands

Where a module implements command line hooks, the commands will be accessible as ordinary Ceph commands. Ceph will automatically incorporate module commands into the standard CLI interface and route them appropriately to the module.:

ceph <command | help>

Configuration

mgr_module_path

Path to load modules from

type

str

mgr_initial_modules

This list of module names is read by the monitor when the cluster is first started after installation, to populate the list of enabled manager modules. Subsequent updates are done using the ‘mgr module [enable|disable]’ commands. List may be comma or space separated.

type

str

default

restful iostat nfs

mgr_disabled_modules

A comma delimited list of module names. This list is read by manager when it starts. By default, manager loads all modules found in specified ‘mgr_module_path’, and it starts the enabled ones as instructed. The modules in this list will not be loaded at all.

type

str

see also

mgr_module_path

mgr_standby_modules

By default, the standby modules will answer incoming requests with a HTTP redirect to the active manager, allowing users to point their browser at any mgr node and find their way to an active mgr. However, this mode is problematic when using a load balancer because (1) the redirect locations are usually private IPs and (2) the load balancer can’t identify which mgr is the right one to send traffic to. If a load balancer is being used, set this to false.

type

bool

default

true

mgr_data

Path to load daemon data (such as keyring)

type

str

default

/var/lib/ceph/mgr/$cluster-$id

mgr_tick_period

How many seconds between mgr beacons to monitors, and other periodic checks.

type

secs

default

2

mon_mgr_beacon_grace

Period in seconds from last beacon to monitor marking a manager daemon as failed

type

secs

default

30