ceph-ansible

Ansible playbook for Ceph!

Clone me:

git clone https://github.com/ceph/ceph-ansible.git

What does it do?

General support for:

• Monitors
• OSDs
• MDSs
• RGW

More details:

• Authentication (cephx), this can be disabled.
• Supports cluster public and private network.
• Monitors deployment. You can easily start with one monitor and then progressively add new nodes. So can deploy one monitor for testing purpose. For production, I recommend to always use an odd number of monitors, 3 tends to be the standard.
• Object Storage Daemons. Like the monitors you can start with a certain amount of nodes and then grow this number. The playbook either supports a dedicated device for storing the journal or both journal and OSD data on the same device (using a tiny partition at the beginning of the device).
• Metadata daemons.
• Collocation. The playbook supports collocating Monitors, OSDs and MDSs on the same machine.
• The playbook was validated on Debian Wheezy, Ubuntu 12.04 LTS and CentOS 6.4.
• Tested on Ceph Dumpling and Emperor.
• A rolling upgrade playbook was written, an upgrade from Dumpling to Emperor was performed and worked.

Configuring Ceph

The supported method for defining your ceph.conf is to use the ceph_conf_overrides variable. This allows you to specify configuration options using an INI format. This variable can be used to override sections already defined in ceph.conf (see: roles/ceph-common/templates/ceph.conf.j2) or to provide new configuration options. The following sections in ceph.conf are supported: [global], [mon], [osd], [mds] and [rgw].

An example:

ceph_conf_overrides:
   global:
     foo: 1234
     bar: 5678
   osd:
     osd mkfs type: ext4

Note: we will no longer accept pull requests that modify the ceph.conf template unless it helps the deployment. For simple configuration tweaks please use the ceph_conf_overrides variable.

Special notes

If you are looking at deploying a Ceph version older than Jewel. It is highly recommended that you apply the following settings to your group_vars/all file on the ceph_conf_overrides variable:

ceph_conf_overrides:
  osd:
    osd recovery max active: 5
    osd max backfills: 2
    osd recovery op priority: 2
    osd recovery threads: 1

ceph#694 removed all the default options that were part of the repo. The goal is to keep the default from Ceph. Below you will find the configuration that was applied prior to the PR in case you want to keep using them:

Setting	ceph-ansible	ceph
cephx require signatures	true	false
cephx cluster require signatures	true	false
osd pool default pg num	128	8
osd pool default pgp num	128	8
rbd concurrent management ops	20	10
rbd default map options	rw	''
rbd default format	2	1
mon osd down out interval	600	300
mon osd min down reporters	7	1
mon clock drift allowed	0.15	0.5
mon clock drift warn backoff	30	5
mon osd report timeout	900	300
mon pg warn max per osd	0	300
mon osd allow primary affinity	true	false
filestore merge threshold	40	10
filestore split multiple	8	2
osd op threads	8	2
filestore op threads	8	2
osd recovery max active	5	15
osd max backfills	2	10
osd recovery op priority	2	63
osd recovery max chunk	1048576	8 << 20
osd scrub sleep	0.1	0
osd disk thread ioprio class	idle	''
osd disk thread ioprio priority	0	-1
osd deep scrub stride	1048576	524288
osd scrub chunk max	5	25

If you want to use them, just use the ceph_conf_overrides variable as explained above.

FAQ

1. I want to have OSD numbers seriallized between hosts, so the first OSD node has osd 1,2,3 and the second has osd 4,5,6 etc. How can I do this? Simply add serial: 1 after the osd section - hosts: osds in your site.yml file.

Setup with Vagrant using virtualbox provider

• Create vagrant_variables.yml

$ cp vagrant_variables.yml.sample vagrant_variables.yml

• Create site.yml

$ cp site.yml.sample site.yml

• Create VMs

$ vagrant up --no-provision --provider=virtualbox
$ vagrant provision
...
...
...
 ____________
< PLAY RECAP >
 ------------
        \   ^__^
         \  (oo)\_______
            (__)\       )\/\
                ||----w |
                ||     ||


mon0                       : ok=16   changed=11   unreachable=0    failed=0
mon1                       : ok=16   changed=10   unreachable=0    failed=0
mon2                       : ok=16   changed=11   unreachable=0    failed=0
osd0                       : ok=19   changed=7    unreachable=0    failed=0
osd1                       : ok=19   changed=7    unreachable=0    failed=0
osd2                       : ok=19   changed=7    unreachable=0    failed=0
rgw                        : ok=20   changed=17   unreachable=0    failed=0

Check the status:

$ vagrant ssh mon0 -c "sudo ceph -s"
    cluster 4a158d27-f750-41d5-9e7f-26ce4c9d2d45
     health HEALTH_OK
     monmap e3: 3 mons at {ceph-mon0=192.168.0.10:6789/0,ceph-mon1=192.168.0.11:6789/0,ceph-mon2=192.168.0.12:6789/0}, election epoch 6, quorum 0,1,2 ceph-mon0,ceph-mon1,ceph-mon
     mdsmap e6: 1/1/1 up {0=ceph-osd0=up:active}, 2 up:standby
     osdmap e10: 6 osds: 6 up, 6 in
      pgmap v17: 192 pgs, 3 pools, 9470 bytes data, 21 objects
            205 MB used, 29728 MB / 29933 MB avail
                 192 active+clean

To re-run the Ansible provisioning scripts:

$ vagrant provision

Specifying fsid and secret key in production

The Vagrantfile specifies an fsid for the cluster and a secret key for the monitor. If using these playbooks in production, you must generate your own fsid in group_vars/all and monitor_secret in group_vars/mons. Those files contain information about how to generate appropriate values for these variables.

Specifying package origin

By default, ceph-common installs from Ceph repository. However, you can set ceph_origin to "distro" to install Ceph from your default repository.

Moreover for people looking to install any version of Ceph prior to the Jewel release on a Red Hat based system you would have to set use_server_package_split: false.

Setup for Vagrant using libvirt provider

• Create vagrant_variables.yml

$ cp vagrant_variables.yml.sample vagrant_variables.yml

• Edit vagrant_variables.yml and setup the following variables:

memory: 1024
disks: "[ '/dev/vdb', '/dev/vdc' ]"
vagrant_box: centos/7

• Create site.yml

$ cp site.yml.sample site.yml

• Create VMs

$ sudo vagrant up --no-provision --provider=libvirt
$ sudo vagrant provision

Setup for Vagrant using parallels provider

• Create vagrant_variables.yml

$ cp vagrant_variables.yml.sample vagrant_variables.yml

• Edit vagrant_variables.yml and setup the following variables:

vagrant_box: parallels/ubuntu-14.04

• Create site.yml

$ cp site.yml.sample site.yml

• Create VMs

$ vagrant up --no-provision --provider=parallels
$ vagrant provision

For Debian based systems

If you want to use "backports", you can set "true" to ceph_use_distro_backports. Attention, ceph-common doesn't manage backports repository, you must add it yourself.

For Atomic systems

If you want to run containerized deployment on Atomic systems (RHEL/CentOS Atomic), please copy vagrant.yml.atomic to vagrant_variables.yml, and copy group_vars/all.docker to group_vars/all.

Since centos/atomic-host VirtualBox box doesn't have spare storage controller to attach more disks, it is likely the first time vagrant up --provider=virtualbox runs, it will fail to attach to a storage controller. In such case, run the following command:

VBoxManage storagectl `VBoxManage list vms |grep ceph-ansible_osd0|awk '{print $1}'|tr \" ' '` --name "SATA" --add sata

then run vagrant up --provider=virtualbox again.

Setup for Vagrant using OpenStack provider

Install the Vagrant plugin for the openstack provider: vagrant plugin install vagrant-openstack-provider.

$ cp site.yml.sample site.yml
$ cp group_vars/all.docker.sample group_vars/all
$ cp vagrant_variables.yml.openstack vagrant_variables.yml

• Edit vagrant_variables.yml: Set mon_vms and osd_vms to the numbers you want. If you are using an Atomic image, un-comment out the skip_tags line. Un-comment the os_ lines. Set os_ssh_username to 'centos' for Centos and 'cloud-user' for RHEL images. Set os_ssh_private_key_path to '~/.ssh/id_rsa' Set os_openstack_auth_url to the auth url of your Open Stack cloud Set os_username and os_password to what you provided for Open Stack registration or leave them as ENV vars if you have set the corresponding env vars for your user. Set os_tenant_name to your Open Stack cloud project name. Set os_region to your Open Stack cloud region name. Set os_flavor to 'm3.medium'. This size has ephemeral storage that will be used by the OSD for the /dev/vdb disk Set the os_image to an image found in the Images list in the Open Stack cloud Dashboard (i.e. 'centos-atomic-host'). Set the os_keypair_name to the keypair name you used when you did the Open Stack registration.

$ vagrant up --provider=openstack

Once the playbook is finished, you should be able to do vagrant ssh mon0 or vagrant ssh osd0 to get to the VMs. sudo docker ps should show the running containers When you are done, use vagrant destroy to get rid of the VMs. You should also remove the associated entries in .ssh/known_hosts so that if the IP addresses get reused by future Open Stack Cloud instances there will not be old known_hosts entries.

Want to contribute?

Read this carefully then :). The repository centralises all the Ansible roles. The roles are all part of the Galaxy. We love contribution and we love giving visibility to our contributors, this is why all the commits must be signed-off.

Submit a patch

To start contributing just do:

$ git checkout -b my-working-branch
$ # do your changes #
$ git add -p

One more step, before pushing your code you should run a syntax check:

$ ansible-playbook -i dummy-ansible-hosts test.yml --syntax-check

If your change impacts a variable file in a role such as roles/ceph-common/defaults/main.yml, you need to generate a group_vars file:

$ ./generate_group_vars_sample.sh

You are finally ready to push your changes on Github:

$ git commit -s
$ git push origin my-working-branch

Worked on a change and you don't want to resend a commit for a syntax fix?

$ # do your syntax change #
$ git commit --amend
$ git push -f origin my-working-branch

Testing PR

Go on the github interface and submit a PR.

Now we have 2 online CIs:

• Travis, simply does a syntax check
• Jenkins Ceph: bootstraps one monitor, one OSD, one RGW

If Jenkins detects that your commit broke something it will turn red. You can then check the logs of the Jenkins by clicking on "Testing Playbooks" button in your PR and go to "Console Output". You can now submit a new commit/change that will update the CI system to run a new play.

It might happen that the CI does not get reloaded so you can simply leave a comment on your PR with "test this please" and it will trigger a new CI build.

Vagrant Demo

Bare metal demo

Deployment from scratch on bare metal machines: