puppet-ceph

About

This is a Puppet module to install a Ceph cluster.

Status

Developped/tested on Debian GNU/Linux Wheezy, targetting the Bobtail Ceph release.

Features

• Ceph package ✓
• Ceph MONs ✓
  • MON configuration ✓
  • MON service key ✓
  • MON filesystem creation ✓
  • MON service ✓
  • MON cluster ✓
  • admin key ✓
• Ceph OSDs ✓
  • OSD configuration ✓
  • OSD bootstrap key ✓
  • OSD device ✓
    • OSD device formatting ✓
    • OSD device mounting ✓
    • OSD filesystem creation ✓
    • OSD service key ✓
  • OSD service ✓
  • OSD registration ✓
  • Insert OSD into crushmap ✓
  • Working OSD ✓
• RadosGW ✓
  • Setup service ✓
  • User handling ✓

TODO

• Finish writing the rspec tests
• Better OSD device placement possibilities
• Test/finish MDS code

Contributing

Contributions are welcome, just fork on GitHub and send a pull-request !

• Please try to keep each new feature / change / fix in a separate pull-request, it will greatly help speeding the merge of the pull-requests ;)
• When adding features, don't forget to add unit tests. Use bundle exec rake spec to run the existing tests.
• puppet-lint (https://github.com/rodjek/puppet-lint) should not produce too much errors too :)
• Some repo-collaborators might ask you some questions or to update your code. All these questions/requests are open to discussion, but please answer, dont just drop your code & abandon it after !

Using

To install a Ceph cluster you'll need at least one host to act as a MON and with the current crushmap defaults two hosts to act as OSDs. (The MON might be the same as an OSD, but has not been tested yet). And of course one puppetmaster :-)

This module requires the puppet master to have storeconfigs = true set and a storage backend configured. On the puppet agents pluginsync = true is required too.

Minimum Puppet manifest for all members of the Ceph cluster

A Ceph cluster needs a cluster fsid : get one with uuidgen -r (Install it with apt-get install uuid-runtime):

$fsid ='some uuid from uuidgen -r'

The general configuration:

class { 'ceph::conf':
  fsid            => $fsid,
  auth_type       => 'cephx', # Currently only cephx is supported AND required
  cluster_network => '10.0.0.0/24', # The cluster's «internal» network
  public_network  => '192.168.0.0/24', # The cluster's «public» (where clients are) network
}

APT configuration to install from the official Ceph repositories:

include ceph::apt::ceph

Puppet manifest for a MON

A MON host also needs the MONs and client.admin secret : get them with ceph-authtool --gen-print-key:

$mon_secret   = 'AQD7kyJQQGoOBhAAqrPAqSopSwPrrfMMomzVdw=='
$admin_secret = 'AQD7kyJQQGoOBhAAqrPAqSopSwPrrfMMomzVdx=='

An Id:

$id = 0 # must be unique for each MON in the cluster

And the mon declaration:

ceph::mon { $id:
  monitor_secret      => $mon_secret,
  client_admin_secret => $admin_secret,
  mon_addr            => '192.168.0.10', # The host's «public» IP address
}

Puppet manifest for an OSD

An OSD host also needs the global host configuration for OSDs:

class { 'ceph::osd':
  client_admin_secret => $admin_secret,
  public_address      => '192.168.0.100',
  cluster_address     => '10.0.0.100',
}

And for each disk/device the path of the physical device to format:

ceph::osd::device { '/dev/sdb': }

WARNING: this previous step will trash all the data from your disk !!!

On an OSD, the puppet agent must be ran at least 4 times for the OSD to be formatted, registered on the OSDs and in the crushmap.

Testing

Using Vagrant

Clone the repo & enter the created directory

git clone git://github.com/enovance/puppet-ceph.git
cd puppet-ceph

Launch three MONs

vagrant up mon0
vagrant up mon1
vagrant up mon2

Run puppet one more time to update the ceph configuration (uses exported resources)

vagrant ssh mon0 -c 'sudo puppet agent -vt'
vagrant ssh mon1 -c 'sudo puppet agent -vt'
vagrant ssh mon2 -c 'sudo puppet agent -vt'

Ceph MONs should be up

vagrant ssh mon0 -c "sudo ceph mon stat"
    e3: 3 mons at {0=192.168.251.10:6789/0,1=192.168.251.11:6789/0,2=192.168.251.12:6789/0}, election epoch 4, quorum 0,1 0,1

Launch at least 2 OSDs

vagrant up osd0
vagrant up osd1
vagrant up osd2

Now login on mon0 (for example) & check ceph health

vagrant ssh mon0 -c 'sudo ceph -s'
   health HEALTH_OK
   monmap e2: 2 mons at {0=192.168.252.10:6789/0,1=192.168.252.11:6789/0}, election epoch 4, quorum 0,1 0,1
   osdmap e35: 6 osds: 6 up, 6 in
    pgmap v158: 192 pgs: 192 active+clean; 0 bytes data, 242 MB used, 23601 MB / 23844 MB avail