Name: ceph-chef
Owner: Ceph
Description: Chef cookbooks for managing a Ceph cluster
Created: 2015-11-13 13:10:03.0
Updated: 2017-07-05 19:47:04.0
Pushed: 2018-02-14 20:31:10.0
Homepage: null
Size: 479
Language: Ruby
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
Installs and configures Ceph, a distributed network storage and filesystem designed to provide excellent performance, reliability, and scalability. Supports Hammer and higher releases (nothing below Hammer is supported in this repo).
Once Hammer support has stopped in Ceph then it will be removed from this cookbook as an option.
The current version is focused on installing and configuring Ceph for Ubuntu, CentOS and RHEL.
For documentation on how to use this cookbook, refer to the USAGE section.
Ceph-Chef works with ALL of the latest versions of Ceph and includes
Ceph-Mgr
recipe.
There are many Enterprises that use the cookbook to install and manage Ceph. Below are a few reference Chef Wrapper Repos and projects that use Ceph-Chef.
https://github.com/bloomberg/chef-bcs. That Chef App (repo) uses this repo for Bloomberg's large clusters. The chef-bcs repo is an S3 Object Store Cluster used in multiple data centers.
https://github.com/cepheus-io/cepheus or http://cepheus.io. This is a powerful and flexible example of use Ceph-Chef. Everything is data driven so you can simply change the data and use it to build your own Ceph cluster or as a reference. It also provides a full management stack for Ceph.
Note: The documentation is a WIP along with a few other features. This repo is actively managed.
For help, use Gitter chat, mailing-list or issues here in this repo.
The original ceph-cookbook will remain and may continue to be updated (see that repo for specifics). We tried to use some of the interesting features of ceph-cookbook but we added a lot of enhancements and simplifications. Simply replacing ceph-cookbook with ceph-chef will not work without a few modifications. Also, ceph-chef only works with Chef 12.8+ and Hammer and higher. Nothing below the Hammer release of Ceph is supported in this repo. In addition, only civitweb is used going forward (not Apache).
NOTE: The current LWRP are using the style prior to Chef version 12.5. There will be a new release shortly that will support the now recommended way of handling custom resources. To make that change easier we will be using a helper cookbook called Poise. Using Poise makes creating custom resources and common services very simple.
>= 12.8+
Tested as working:
The ceph cookbook requires the following cookbooks from Chef:
https://supermarket.chef.io/
Ceph cluster design and Ceph support are beyond the scope of this README, please turn to the:
public wiki, mailing lists, visit our IRC channel, or contact Red Hat:
http://ceph.com/docs/master http://ceph.com/resources/mailing-list-irc/
This cookbook can be used to implement a chosen cluster design. Most of the configuration is retrieved from node attributes, which can be set by an environment json file or by a wrapper cookbook that updates the attributes directly. A basic cluster configuration will need most of the following attributes:
node['ceph']['config']['fsid']
- the cluster UUIDnode['ceph']['config]'['global']['public network']
- a CIDR specification of the public networknode['ceph']['config]'['global']['cluster network']
- a CIDR specification of a separate cluster replication networknode['ceph']['config]'['global']['rgw dns name']
- the main domain of the radosgw daemonMost notably, the configuration does NOT need to set the mon initial members
, because the cookbook does a node search based on TAGS or ENVIRONMENTS to find other mons in the same environment. However, you can add them to node['ceph']['config']['global']['mon initial members'] = <whatever mon ip list you want>
The other set of attributes that this recipe needs is node['ceph']['osd']['devices']
, which is an array of OSD definitions, similar to the following:
ceph-disk-prepare
An example of finding a mon socket in a generic like environment.
ceph-conf --name mon.$(hostname -s) --show-config-value admin_socket
Pools - After creating it appears that some of the PGs are stuck 'creating+peering'. This can be caused by a number of things. Most likely an OSD is not blocking the creation. Do something like:
ceph pg ls-by-pool
Something like: ceph pg ls-by-pool .rgw | grep creating
It should show you the PG number as the first column. Use that to query it to see if something is blocking it.
ceph pg
Something like: ceph pg 175.f7 query
This will return a lot of json data. You can grep or look for 'blocking'. If found then restart the given OSD. You can find the host for the OSD with: ceph osd find <OSD number>
Once you're on the host simply restart the specific OSD with: sudo service ceph restart osd.<whatever the number>
To automate setting several of these node attributes, it is recommended to use a policy wrapper cookbook. This allows the ability to use Chef Server cookbook versions along with environment version restrictions to roll out configuration changes in an ordered fashion.
It also can help with automating some settings. For example, a wrapper cookbook could peek at the list of hard drives that ohai
has found and populate node['ceph']['osd_devices'] accordingly, instead of manually typing them all in:
.override['ceph']['osd_devices'] = node['block_device'].each.reject{ |name, data| name !~ /^sd[b-z]/}.sort.map { |name, data| {'journal' => "/dev/#{name}"} }
For best results, the wrapper cookbook's recipe should be placed before the Ceph cookbook in the node's runlist. This will ensure that any attributes are in place before the Ceph cookbook runs and consumes those attributes.
Ceph monitor nodes should use the ceph-mon role.
Includes:
Ceph metadata server nodes should use the ceph-mds role. NB: Only required for Ceph-FS
Includes:
Ceph OSD nodes should use the ceph-osd role
Includes:
Ceph RGW nodes should use the ceph-radosgw role
node['ceph']['search_environment']
- a custom Chef environment to search when looking for mon nodes. The cookbook defaults to searching the current environment
node['ceph']['branch']
- selects whether to install the stable, testing, or dev version of Ceph
node['ceph']['version']
- install a version of Ceph that is different than the cookbook default. If this is changed in a wrapper cookbook, some repository urls may also need to be replaced, and they are found in attributes/repo.rb. If the branch attribute is set to dev, this selects the gitbuilder branch to install
node['ceph']['extras_repo']
- whether to install the ceph extras repo. The tgt recipe requires this
node['ceph']['config']['fsid']
- the cluster UUID
node['ceph']['config']['global']['public network']
- a CIDR specification of the public network
node['ceph']['config']['global']['cluster network']
- a CIDR specification of a separate cluster replication network
node['ceph']['config']['config-sections']
- add to this hash to add extra config sections to the ceph.conf
node['ceph']['user_pools']
- an array of pool definitions, with attributes name
, pg_num
and create_options
(optional), that are automatically created when a monitor is deployed
node['ceph']['config']['mon']
- a hash of settings to save in ceph.conf in the [mon] section, such as 'mon osd nearfull ratio' => '0.75'
node['ceph']['osd_devices']
- an array of OSD definitions for the current nodenode['ceph']['config']['osd']
- a hash of settings to save in ceph.conf in the [osd] section, such as 'osd max backfills' => 2
node['ceph']['config']['osd']['osd crush location']
- this attribute can be set on a per-node basis to maintain Crush map locationsnode['ceph']['config']['mds']
- a hash of settings to save in ceph.conf in the [mds] section, such as 'mds cache size' => '100000'
node['ceph']['cephfs_mount']
- where the cephfs recipe should mount CephFSnode['ceph']['cephfs_use_fuse']
- whether the cephfs recipe should use the fuse cephfs client. It will default to heuristics based on the kernel versionnode['ceph']['radosgw']['port']
- Port of the rgw. Defaults to 80node['ceph']['radosgw']['rgw_dns_name']
- the main domain of the radosgw daemon, to calculate the bucket name from a subdomainThe ceph_client LWRP provides an easy way to construct a Ceph client key. These keys are needed by anything that needs to talk to the Ceph cluster, including RGW, CephFS, and RBD access.
{ 'mon' => 'allow r', 'osd' => 'allow r' }
client.#{name}.#{hostname}
/etc/ceph/ceph.client.#{name}.#{hostname}.keyring
if as_keyring
and /etc/ceph/ceph.client.#{name}.#{hostname}.secret
if not as_keyring
The ceph_cephfs LWRP provides an easy way to mount CephFS. It will automatically create a Ceph client key for the machine and mount CephFS to the specified location. If the kernel client is used, instead of the fuse client, a pre-existing subdirectory of CephFS can be mounted instead of the root.
The ceph_pool LWRP provides an easy way to create and delete Ceph pools.
It assumes that connectivity to the cluster is setup and that admin credentials are available from default locations, e.g. /etc/ceph/ceph.client.admin.keyring.
This cookbook requires a style guide for all contributions. Travis will automatically verify that every Pull Request follows the style guide.
eval "$(chef shell-init bash)"
bundle install
bundle exec rake style
This cookbook uses Test Kitchen to verify functionality. A Pull Request can't be merged if it causes any of the test configurations to fail.
eval "$(chef shell-init bash)"
bundle install
bundle exec kitchen test aio-debian-74
bundle exec kitchen test aio-ubuntu-1204
bundle exec kitchen test aio-ubuntu-1404
We welcome ALL contributions:
Your changes will be tested and if all goes well it will be merged - Thanks!!