RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2024-12-26 21:43:10 +00:00
Code Issues Packages Projects Releases Wiki Activity
cd2cabecf1
ceph/doc/rados/configuration/mon-config-ref.rst
John Wilkins ca77aabbf1 doc: Rewrote monitor configuration section. 
Signed-off-by: John Wilkins <john.wilkins@inktank.com>
2013-03-29 17:34:45 -07:00

796 lines
27 KiB
ReStructuredText
Raw Blame History

==========================
Monitor Config Reference
==========================
Understanding how to configure a Ceph monitor is an important part of building a
reliable Ceph cluster. **All Ceph clusters have at least one monitor**. A
monitor configuration usually remains fairly consistent, but you can add,
remove or replace a monitor in a cluster. See `Adding/Removing a Monitor`_
for details.
Background
==========
Monitors maintain a "master copy" of the cluster map, which means a client can
determine the location of all monitors, OSDs, and metadata servers just by
connecting to one monitor and retrieving a current cluster map. Before Ceph
clients can read from or write to OSDs or metadata servers, they must connect to
a monitor first. With a current copy of the cluster map and the CRUSH algorithm,
a client can compute the location for any object. The ability to compute object
locations allows a client to talk directly to OSDs, which is a very important
aspect of Ceph's high scalability and performance.
The primary role of the monitor is to maintain a master copy of the cluster map.
Monitors also provide authentication and logging services. Ceph monitors write
all changes in the monitor services to a single Paxos instance, and Paxos writes
the changes to a key/value store for strong consistency. Ceph monitors can query
the most recent version of the cluster map during sync operations. Ceph monitors
leverage the key/value store's snapshots and iterators (using leveldb) to
perform store-wide synchronization.
.. ditaa::
/-------------\ /-------------\
| Monitor | Write Changes | Paxos |
| cCCC +-------------->+ cCCC |
| | | |
+-------------+ \------+------/
| Auth | |
+-------------+ | Write Changes
| Log | |
+-------------+ v
| Monitor Map | /------+------\
+-------------+ | Key / Value |
| OSD Map | | Store |
+-------------+ | cCCC |
| PG Map | \------+------/
+-------------+ ^
| MDS Map | | Read Changes
+-------------+ |
| cCCC |*---------------------+
\-------------/
.. deprecated:: version 0.58
In Ceph versions 0.58 and earlier, Ceph monitors use a Paxos instance for
each service and store the map as a file.
Cluster Maps
------------
The cluster map is a composite of maps, including the monitor map, the OSD map,
the placement group map and the metadata server map. The cluster map tracks a
number of important things: which processes are ``in`` the cluster; which
processes that are ``in`` the cluster are ``up`` and running or ``down``;
whether, the placement groups are ``active`` or ``inactive``, and ``clean`` or
in some other state; and, other details that reflect the current state of the
cluster such as the total amount of storage space, and the amount of storage
used.
When there is a significant change in the state of the cluster--e.g., an OSD
goes down, a placement group falls into a degraded state, etc.--the cluster map
gets updated to reflect the current state of the cluster. Additionally, the
monitor also maintains a history of the prior states of the cluster. The monitor
map, OSD map, placement group map and metadata server map each maintain a
history of their map versions. We call each version an "epoch."
When operating your cluster, keeping track of these states is an important
part of your system administration duties. See `Monitoring a Cluster`_ and
`Monitoring OSDs and PGs`_ for details.
Monitor Quorum
--------------
Our 5-minute Quick Start provides a trivial `Ceph configuration file`_ that
provides for one monitor in the test cluster. A cluster will run fine with a
single monitor; however, **a single monitor is a single-point-of-failure**. To
ensure high availability in a production cluster, you should run Ceph with
multiple monitors so that the failure of a single monitor **WILL NOT** bring
down your entire cluster.
When a cluster runs multiple monitors for high availability, Ceph monitors use
`Paxos`_ to establish consensus about the master cluster map. A consensus
requires a majority of monitors running to establish a quorum for consensus
about the cluster map (e.g., 1; 2 out of 3; 3 out of 5; 4 out of 6; etc.).
Consistency
-----------
When you add monitor settings to your Ceph configuration file, you need to be
aware of some of the architectural aspects of Ceph monitors. **Ceph imposes
strict consistency requirements** for a Ceph monitor when discovering another
Ceph monitor within the cluster. Whereas, Ceph clients and other Ceph daemons
use the Ceph configuration file to discover monitors, monitors discover each
other using the monitor map (monmap), not the Ceph configuration file.
A monitor always refers to the local copy of the monmap when discovering other
monitors in the cluster. Using the monmap instead of the Ceph configuration file
avoids errors that could break the cluster (e.g., typos in ``ceph.conf`` when
specifying a monitor address or port). Since monitors use monmaps for discovery
and they share monmaps with clients and other Ceph daemons, **the monmap
provides monitors with a strict guarantee that their consensus is valid.**
Strict consistency also applies to updates to the monmap. As with any other
updates on the monitor, changes to the monmap always run through a distributed
consensus algorithm called `Paxos`_. The monitors must agree on each update to
the monmap, such as adding or removing a monitor, to ensure that each monitor in
the quorum has the same version of the monmap. Updates to the monmap are
incremental so that monitors have the latest agreed upon version, and a set of
previous versions. Maintaining a history enables a monitor that has an older
version of the monmap to catch up with the current state of the cluster.
If monitors discovered each other through the Ceph configuration file instead of
through the monmap, it would introduce additional risks because the Ceph
configuration files aren't updated and distributed automatically. Monitors might
inadvertently use an older Ceph configuration file, fail to recognize a monitor,
fall out of a quorum, or develop a situation where `Paxos`_ isn't able to
determine the current state of the system accurately.
Bootstrapping Monitors
----------------------
In most configuration and deployment cases, tools that deploy Ceph may help
bootstrap the monitors by generating a monitor map for you (e.g., ``mkcephfs``,
``ceph-deploy``, etc). A monitor requires four explicit settings:
- **Filesystem ID**: The ``fsid`` is the unique identifier for your object
store. Since you can run multiple clusters on the same hardware, you must
specify the unique ID of the object store when bootstrapping a monitor.
Deployment tools usually do this for you (e.g., ``mkcephfs`` or
``ceph-deploy`` can call a tool like ``uuidgen``), but you may specify the
``fsid`` manually too.
- **Monitor ID**: A monitor ID is a unique ID assigned to each monitor within
the cluster. It is an alphanumeric value, and by convention the identifier
usually follows an alphabetical increment (e.g., ``a``, ``b``, etc.). This
can be set in a Ceph configuration file (e.g., ``[mon.a]``, [mon.b]``, etc.),
by a deployment tool, or using the ``ceph`` commandline.
- **Keys**: The monitor must have secret keys. A deployment tool such as
``mkcephfs`` or ``ceph-deploy`` usually does this for you, but you may
perform this step manually too. See `Monitor Keyrings`_ for details.
For additional details on bootstrapping, see `Bootstrapping a Monitor`_.
Configuring Monitors
====================
To apply configuration settings to the entire cluster, enter the configuration
settings under ``[global]``. To apply configuration settings to all monitors in
your cluster, enter the configuration settings under ``[mon]``. To apply
configuration settings to specific monitors, specify the monitor instance
(e.g., ``[mon.a]``). By convention, monitor instance names use alpha notation.
.. code-block:: ini
[global]
[mon]
[mon.a]
[mon.b]
[mon.c]
Minimum Configuration
---------------------
The bare minimum monitor settings for a Ceph monitor via the Ceph configuration
file include a hostname and a monitor address for each monitor. You can configure
these under ``[mon]`` or under the entry for a specific monitor.
.. code-block:: ini
[mon]
mon host = hostname1,hostname2,hostname3
mon addr = 10.0.0.10:6789,10.0.0.11:6789,10.0.0.12:6789
.. code-block:: ini
[mon.a]
host = hostname1
mon addr = 10.0.0.10:6789
See the `Network Configuration Reference`_ for details.
.. note:: This minimum configuration for monitors assumes that a deployment
tool generates the ``fsid`` and the ``mon.`` key for you.
Once you deploy a Ceph cluster, you **SHOULD NOT** change the IP address of
the monitors. However, if you decide to change the monitor's IP address, you
must follow a specific procedure. See `Changing a Monitor's IP Address`_ for
details.
Cluster ID
----------
Each Ceph cluster has a unique identifier (``fsid``). If specified, it usually
appears under the ``[global]`` section of the configuration file. Deployment
tools usually generate the ``fsid`` and store it in the monitor map, so the
value may not appear in a configuration file. The ``fsid`` makes it possible to
run daemons for multiple clusters on the same hardware.
``fsid``
:Description: The cluster ID. One per cluster.
:Type: UUID
:Required: Yes.
:Default: N/A. May be generated by a deployment tool if not specified.
.. note:: Do not set this value if you use a deployment tool that does
it for you.
Initial Members
---------------
We recommend running a production cluster with at least three monitors to ensure
high availability. When you run multiple monitors, you may specify the initial
monitors that must be members of the cluster in order to establish a quorum.
This may reduce the time it takes for your cluster to come online.
.. code-block:: ini
[mon]
mon initial members = a,b,c
``mon initial members``
:Description: The IDs of initial monitors in a cluster during startup. If
specified, Ceph requires an odd number of monitors to form an
initial quorum (e.g., 3).
:Type: String
:Default: None
.. note:: A *majority* of monitors in your cluster must be able to reach
each other in order to establish a quorum. You can decrease the initial
number of monitors to establish a quorum with this setting.
Data
----
Ceph provides a default path where monitors store data. For optimal performance
in a production cluster, we recommend running monitors on separate hosts and
drives from OSDs. Monitors do lots of ``fsync()``, which can interfere with OSD
workloads.
In Ceph versions 0.58 and earlier, monitors store their data in files. This
approach allows users to inspect monitor data with common tools like ``ls``
and ``cat``. However, it doesn't provide strong consistency.
In Ceph versions 0.59 and later, monitors store their data as key/value pairs.
Monitors require `ACID`_ transactions. Using a data store prevents recovering
monitors from running corrupted versions through Paxos, and it enables multiple
modification operations in one single atomic batch, among other advantages.
Generally, we do not recommend changing the default data location. If you modify
the default location, we recommend that you make it uniform across monitors by
setting it in the ``[mon]`` section of the configuration file.
``mon data``
:Description: The monitor's data location.
:Type: String
:Default: ``/var/lib/ceph/mon/$cluster-$id``
Storage Capacity
----------------
When a Ceph cluster gets close to its maximum capacity (i.e., ``mon osd full
ratio``), Ceph prevents you from writing to or reading from OSDs as a safety
measure to prevent data loss. Therefore, letting a production cluster approach
its full ratio is not a good practice, because it sacrifices high availability.
The default full ratio is ``.95``, or 95% of capacity. This a very aggressive
setting for a test cluster with a small number of OSDs.
.. tip:: When monitoring your cluster, be alert to warnings related to the
``nearfull`` ratio. This means that a failure of some OSDs could result
in a temporary service disruption if one or more OSDs fails. Consider adding
more OSDs to increase storage capacity.
A common scenario for test clusters involves a system administrator removing an
OSD from the cluster to watch the cluster rebalance; then, removing another OSD,
and so on until the cluster eventually reaches the full ratio and locks up. We
recommend a bit of capacity planning even with a test cluster so that you can
gauge how much spare capacity you will need to maintain for high availability.
Ideally, you want to plan for a series of OSD failures where the cluster can
recover to an ``active + clean`` state without replacing those OSDs immediately.
You can run a cluster in an ``active + degraded`` state, but this is not ideal
for normal operating conditions.
The following diagram depicts a simplistic Ceph cluster containing 33 hosts with
one OSD per host, each OSD having a 3TB capacity. So this exemplary cluster has
a maximum actual capacity of 99TB. With a ``mon osd full ratio`` of ``0.95``, if
the cluster falls to 5TB of remaining capacity, the cluster will not allow Ceph
clients to read and write data. So its operating capacity is 95TB, not 99TB.
.. ditaa::
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| Rack 1 | | Rack 2 | | Rack 3 | | Rack 4 | | Rack 5 | | Rack 6 |
| cCCC | | cF00 | | cCCC | | cCCC | | cCCC | | cCCC |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| OSD 1 | | OSD 7 | | OSD 13 | | OSD 19 | | OSD 25 | | OSD 31 |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| OSD 2 | | OSD 8 | | OSD 14 | | OSD 20 | | OSD 26 | | OSD 32 |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| OSD 3 | | OSD 9 | | OSD 15 | | OSD 21 | | OSD 27 | | OSD 33 |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| OSD 4 | | OSD 10 | | OSD 16 | | OSD 22 | | OSD 28 | | Spare |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| OSD 5 | | OSD 11 | | OSD 17 | | OSD 23 | | OSD 29 | | Spare |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
| OSD 6 | | OSD 12 | | OSD 18 | | OSD 24 | | OSD 30 | | Spare |
+--------+ +--------+ +--------+ +--------+ +--------+ +--------+
It is normal in such a cluster for one or two OSDs to fail. A less frequent but
reasonable scenario involves a rack's router or power supply failing, which
brings down multiple OSDs simultaneously (e.g., OSDs 7-12). In such a scenario,
you should still strive for a cluster that can remain operational and achieve an
``active + clean`` state--even if that means adding a few hosts with additional
OSDs in short order. If your capacity utilization is too high, you may not lose
data, but you could still sacrifice data availability while resolving an outage
within a failure domain if capacity utilization of the cluster exceeds the full
ratio. For this reason, we recommend at least some rough capacity planning.
Identify two numbers for your cluster:
#. The number of OSDs.
#. The total capacity of the cluster
If you divide the total capacity of your cluster by the number of OSDs in your
cluster, you will find the mean average capacity of an OSD within your cluster.
Consider multiplying that number by the number of OSDs you expect will fail
simultaneously during normal operations (a relatively small number). Finally
multiply the capacity of the cluster by the full ratio to arrive at a maximum
operating capacity; then, subtract the number of amount of data from the OSDs
you expect to fail to arrive at a reasonable full ratio. Repeat the foregoing
process with a higher number of OSD failures (e.g., a rack of OSDs) to arrive at
a reasonable number for a near full ratio.
.. code-block:: ini
[global]
mon osd full ratio = .80
mon osd nearfull ratio = .70
``mon osd full ratio``
:Description: The percentage of disk space used before an OSD is
considered ``full``.
:Type: Float
:Default: ``.95``
``mon osd nearfull ratio``
:Description: The percentage of disk space used before an OSD is
considered ``nearfull``.
:Type: Float
:Default: ``.85``
.. tip:: If some OSDs are nearfull, but others have plenty of capacity, you
may have a problem with the CRUSH weight for the nearfull OSDs.
Heartbeat
---------
Ceph monitors know about the cluster by requiring reports from each OSD, and by
receiving reports from OSDs about the status of their neighboring OSDs. Ceph
provides reasonable default settings for monitor/OSD interaction; however, you
may modify them as needed. See `Monitor/OSD Interaction`_ for details.
Monitor Store Synchronization
-----------------------------
When you run a production cluster with multiple monitors (recommended), each
monitor checks to see if a neighboring monitor has a more recent version of the
cluster map (e.g., a map in a neighboring monitor with one or more epoch numbers
higher than the most current epoch in the map of the instant monitor).
Periodically, one monitor in the cluster may fall behind the other monitors to
the point where it must leave the quorum, synchronize to retrieve the most
current information about the cluster, and then rejoin the quorum. For the
purposes of synchronization, monitors may assume one of three roles:
#. **Leader**: The `Leader` is the first monitor to achieve the most recent
Paxos version of the cluster map.
#. **Provider**: The `Provider` is a monitor that has the most recent version
of the cluster map, but wasn't the first to achieve the most recent version.
#. **Requester:** A `Requester` is a monitor that has fallen behind the leader
and must synchronize in order to retrieve the most recent information about
the cluster before it can rejoin the quorum.
These roles enable a leader to delegate synchronization duties to a provider,
which prevents synchronization requests from overloading the leader--improving
performance. In the following diagram, the requester has learned that it has
fallen behind the other monitors. The requester asks the leader to synchronize,
and the leader tells the requester to synchronize with a provider.
.. ditaa:: +-----------+ +---------+ +----------+
| Requester | | Leader | | Provider |
+-----------+ +---------+ +----------+
| | |
| | |
| Ask to Synchronize | |
|------------------->| |
| | |
|<-------------------| |
| Tell Requester to | |
| Sync with Provider | |
| | |
| Synchronize |
|--------------------+-------------------->|
| | |
|<-------------------+---------------------|
| Send Chunk to Requester |
| (repeat as necessary) |
| Requester Acks Chuck to Provider |
|--------------------+-------------------->|
| |
| Sync Complete |
| Notification |
|------------------->|
| |
|<-------------------|
| Ack |
| |
Synchronization always occurs when a new monitor joins the cluster. During
runtime operations, monitors may receive updates to the cluster map at different
times. This means the leader and provider roles may migrate from one monitor to
another. If this happens while synchronizing (e.g., a provider falls behind the
leader), the provider can terminate synchronization with a requester.
Once synchronization is complete, Ceph requires trimming across the cluster.
Trimming requires that the placement groups are ``active + clean``.
``mon sync trim timeout``
:Description:
:Type: Double
:Default: ``30.0``
``mon sync heartbeat timeout``
:Description:
:Type: Double
:Default: ``30.0``
``mon sync heartbeat interval``
:Description:
:Type: Double
:Default: ``5.0``
``mon sync backoff timeout``
:Description:
:Type: Double
:Default: ``30.0``
``mon sync timeout``
:Description:
:Type: Double
:Default: ``30.0``
``mon sync max retries``
:Description:
:Type: Integer
:Default: ``5``
``mon sync max payload size``
:Description: The maximum size for a sync payload.
:Type: 32-bit Integer
:Default: ``1045676``
``mon accept timeout``
:Description: Number of seconds the Leader will wait for the Requester(s) to
accept a Paxos update. It is also used during the Paxos recovery
phase for similar purposes.
:Type: Float
:Default: ``10.0``
``paxos propose interval``
:Description: Gather updates for this time interval before proposing a map update.
:Type: Double
:Default: ``1.0``
``paxos min wait``
:Description: The minimum amount of time to gather updates after a period of
inactivity.
:Type: Double
:Default: ``0.05``
``paxos trim tolerance``
:Description: The number of extra proposals tolerated before trimming.
:Type: Integer
:Default: ``30``
``paxos trim disabled max versions``
:Description: The maximimum number of version allowed to pass without trimming.
:Type: Integer
:Default: ``100``
``mon lease``
:Description: The length (in seconds) of the lease on the monitor's versions.
:Type: Float
:Default: ``5``
``mon lease renew interval``
:Description: The interval (in seconds) for the Leader to renew the other
monitor's leases.
:Type: Float
:Default: ``3``
``mon lease ack timeout``
:Description: The number of seconds the Leader will wait for the Providers to
acknowledge the lease extension.
:Type: Float
:Default: ``10.0``
``mon min osdmap epochs``
:Description: Minimum number of OSD map epochs to keep at all times.
:Type: 32-bit Integer
:Default: ``500``
``mon max pgmap epochs``
:Description: Maximum number of PG map epochs the monitor should keep.
:Type: 32-bit Integer
:Default: ``500``
``mon max log epochs``
:Description: Maximum number of Log epochs the monitor should keep.
:Type: 32-bit Integer
:Default: ``500``
Slurp
-----
In Ceph version 0.58 and earlier, when a Paxos service drifts beyond a given
number of versions, Ceph triggers the `slurp` mechanism, which establishes a
connection with the quorum Leader and obtains every single version the Leader
has for every service that has drifted. In Ceph versions 0.59 and later, slurp
will not work, because there is a single Paxos instance for all services.
.. deprecated:: 0.58
``paxos max join drift``
:Description: The maximum Paxos iterations before we must first sync the
monitor data stores.
:Type: Integer
:Default: ``10``
``mon slurp timeout``
:Description: The number of seconds the monitor has to recover using slurp
before the process is aborted and the monitor bootstraps.
:Type: Double
:Default: ``10.0``
``mon slurp bytes``
:Description: Limits the slurp messages to the specified number of bytes.
:Type: 32-bit Integer
:Default: ``256 * 1024``
Clock
-----
``clock offset``
:Description: How much to offset the system clock. See ``Clock.cc`` for details.
:Type: Double
:Default: ``0``
.. deprecated:: 0.58
``mon tick interval``
:Description: A monitor's tick interval in seconds.
:Type: 32-bit Integer
:Default: ``5``
``mon clock drift allowed``
:Description: The clock drift in seconds allowed between monitors.
:Type: Float
:Default: ``.050``
``mon clock drift warn backoff``
:Description: Exponential backoff for clock drift warnings
:Type: Float
:Default: ``5``
``mon timecheck interval``
:Description: The time check interval (clock drift check) in seconds
for the leader.
:Type: Float
:Default: ``300.0``
Client
------
``mon client hung interval``
:Description: The client will try a new monitor every ``N`` seconds until it
establishes a connection.
:Type: Double
:Default: ``3.0``
``mon client ping interval``
:Description: The client will ping the monitor every ``N`` seconds.
:Type: Double
:Default: ``10.0``
``mon client max log entries per message``
:Description: The maximum number of log entries a monitor will generate
per client message.
:Type: Integer
:Default: ``1000``
``mon client bytes``
:Description: The amount of client message data allowed in memory (in bytes).
:Type: 64-bit Integer Unsigned
:Default: ``100ul << 20``
Miscellaneous
=============
``mon max osd``
:Description: The maximum number of OSDs allowed in the cluster.
:Type: 32-bit Integer
:Default: ``10000``
``mon globalid prealloc``
:Description: The number of global IDs to pre-allocate for clients and daemons in the cluster.
:Type: 32-bit Integer
:Default: ``100``
``mon sync fs threshold``
:Description: Synchronize with the filesystem when writing the specified number of objects. Set it to ``0`` to disable it.
:Type: 32-bit Integer
:Default: ``5``
``mon subscribe interval``
:Description: The refresh interval (in seconds) for subscriptions. The
subscription mechanism enables obtaining the cluster maps
and log information.
:Type: Double
:Default: ``300``
``mon stat smooth intervals``
:Description: Ceph will smooth statistics over the last ``N`` PG maps.
:Type: Integer
:Default: ``2``
``mon probe timeout``
:Description: Number of seconds the monitor will wait to find peers before bootstrapping.
:Type: Double
:Default: ``2.0``
``mon daemon bytes``
:Description: The message memory cap for metadata server and OSD messages (in bytes).
:Type: 64-bit Integer Unsigned
:Default: ``400ul << 20``
``mon max log entries per event``
:Description: The maximum number of log entries per event.
:Type: Integer
:Default: ``4096``
.. _Paxos: http://en.wikipedia.org/wiki/Paxos_(computer_science)
.. _Monitor Keyrings: ../../operations/authentication#monitor-keyrings
.. _Ceph configuration file: ../../../start/quick-start/#add-a-configuration-file
.. _Network Configuration Reference: ../network-config-ref
.. _ACID: http://en.wikipedia.org/wiki/ACID
.. _Adding/Removing a Monitor: ../../operations/add-or-rm-mons
.. _Monitoring a Cluster: ../../operations/monitoring
.. _Monitoring OSDs and PGs: ../../operations/monitoring-osd-pg
.. _Bootstrapping a Monitor: ../../../dev/mon-bootstrap
.. _Changing a Monitor's IP Address: ../../operations/add-or-rm-mons#changing-a-monitor-s-ip-address
.. _Monitor/OSD Interaction: ../mon-osd-interaction
Reference in New Issue View Git Blame Copy Permalink