Skip to content

Latest commit

 

History

History
219 lines (159 loc) · 7.56 KB

README.md

File metadata and controls

219 lines (159 loc) · 7.56 KB

Decom Note:

rpc-ceph is no longer being developed or tested. Please use the upstream ceph-ansible playbooks for any future deployments.

rpc-ceph and ceph-ansible

rpc-ceph deploys Ceph as an RPC stand-alone platform in a uniform, managed, and tested way to ensure version consistency and testing.

By adding automated tests, rpc-ceph provides a way to manage tested versions of ceph-ansible used in RPC deployments.

Current versions of ceph-ansible & Ansible

ceph-ansible version: v3.1.12

Ansible version: 2.4.4.0

What is rpc-ceph?

rpc-ceph is a thin wrapper around the ceph-ansible project. rpc-ceph manages the versions of ansible and ceph-ansible by providing:

  • RPC integration testing (MaaS/Logging and WIP-OpenStack).
  • Tested and versioned ceph-ansible and Ceph releases.
  • Default variables (still WIP) for base installs.
  • Standarized deployments.
  • Default playbooks for integration.
  • Benchmarking tools using fio.

Deploying rpc-ceph uses boostrap.sh, ceph-ansible, default group_vars, and a pre-created playbook.

NOTE: Anything that can be configured with ceph-ansible is configurable with rpc-ceph.

Deploying Ceph for multi-node and production environments

Architecture

We do not recommend or use containers for rpc-ceph production deployments. Containers are setup and used as part of the run_tests.sh (AIO) testing strategy only. The default playbooks are not set up to build containers or configure any of the required container specific roles.

The inventory should consist of the following:

  • 1-3+ mons hosts (perferably 3 or more), and an uneven number of them.
  • 1-3+ mgrs hosts (perferably 3 or more) - Ideally on the mon hosts (Since the Luminous release this is required).
  • 3+ osds hosts with storage drives.
  • 1+ repo_servers hosts to serve as apt repo servers for ceph version pinning.
  • OPTIONAL: 1-3+ rgws hosts - these will be load balanced.
  • rsyslog_all host, pointing to an existing or new rsyslog logging server.
  • OPTIONAL:benchmark_hosts - the host on which to run benchmarking (Read benchmark/README.md for more).

Configuring your deployment host

  1. Configure the following inventory:

    • ansible_host var for each host.
    • Devices, dedicated_devices for osd hosts.
  2. Configure a variables file including the following ceph-ansible vars:

    • monitor_interface
    • public_network
    • cluster_network
    • osd_scenario
    • repo_server_interface
    • Any other ceph-ansible settings you want to configure.
  3. Set any override vars in playbooks/group_vars/host_group/overrides.yml, this allows:

    • Defaults to remain, but be overriden if required (overrides.yml will take precedence).
    • Git will ignore the overrides.yml file, so the repo can be updated without clearing out all deploy specific vars.
  4. Override any variables from ceph.conf using ceph_conf_overrides_extra or ceph_conf_overrides_<group>_extra:

    • This allows the default group_vars to remain in place, and means you do not have to respecify any vars you aren't setting.
    • The ceph_conf_overrides_<group>_extra var will override only vars for only the hosts in that group, with currently supported groups:
      • ceph_conf_overrides_rgw_extra
      • ceph_conf_overrides_mon_extra
      • ceph_conf_overrides_mgr_extra
      • ceph_conf_overrides_osd_extra
    • The overrides will merge with the existing settings and take precedence but not squash them.
  5. Run the bootstrap-ansible.sh inside the scripts directory:

    ./scripts/bootstrap-ansible.sh
  6. This configures ansible at a pre-tested version, creates a ceph-ansible binary that points to the appropriate ansible-playbook binary, and clones the required role repositories:

    • ceph-ansible
    • rsyslog_client
    • openstack-ansible-plugins (ceph-ansible uses the config template plugin from here).
    • haproxy_server
    • rsyslog_server
  7. Run the ceph-ansible playbook from the playbooks directory:

    ceph-ansible-playbook -i <link to your inventory file> playbooks/add-repo.yml -e @<link to your vars file>
    ceph-ansible-playbook -i <link to your inventory file> playbooks/deploy-ceph.yml -e @<link to your vars file>
  8. Run any additional playbooks from the playbook directory:

    • ceph-setup-logging.ymlwill setup rsyslog client, ensure you have the appropriate rsyslog server setup, or other log shipping location, refer to: https://docs.openstack.org/openstack-ansible-rsyslog_client/latest/ for more details
    • ceph-keystone-rgw.yml will setup required keystone users and endpoints for Ceph.
    • ceph-rgw-haproxy.yml will setup the HAProxy VIP for Ceph Rados GW. Ensure you specify haproxy_all group in your inventory with the HAProxy hosts.
    • ceph-rsyslog-server.yml will setup rsyslog server on the rsyslog_all hosts specified. NB If there is already an existing rsyslog server that you are connecting into, you should not run this.

Your deployment should be successful.

NOTE: If there are any errors, troubleshoot as a standard ceph-ansible deployment.

Deploying Ceph as an AIO

Virtual Machine requirements for AIO

  • RAX Public Cloud general-8 (or equivalent) using:
    • Ubuntu 16.04 (xenial)
    • CentOS7

To run an AIO build

For MaaS integration, perform the following export commands. Otherwise just use ./run_tests.sh to build the AIO.

export PUBCLOUD_USERNAME=<username>
export PUBCLOUD_API_KEY=<api_key>

AIO Scenarios

To run an AIO scenario for Ceph you can run the following export on a general1-8 or perf2-15 flavor instance, unless otherwise noted: #export RE_JOB_SCENARIO="name of scenario from below"

build_releasenotes This will build the project releae notes using sphinx and place it in the directory rpc-ceph/release/build/

functional: This is a base AIO for Ceph, includes MaaS testing, this runs on each commit, with the following components:

  • 2 x rgw hosts
  • 3 x osd hosts
  • 3 x mon hosts
  • 3 x mgr hosts
  • 1 x rsyslog server
  • HAproxy configured on localhost

This job does not run the benchmarking playbooks.

bluestore: This is the same as the functional job but runs using bluestore, and 3 collocated OSD devices per osd host.

rpco_newton: An RPC-O newton integration test, that will deploy an RPC-O AIO, and integrate it with Ceph, followed by Tempest tests. This runs daily, as it takes a long time to build.

  • RPC-O AIO @ newton
    • Keystone
    • Glance
    • Cinder
    • Nova
    • Neutron
    • Tempest
  • 2 x rgw hosts
  • 3 x osd hosts
  • 3 x mon hosts
  • 3 x mgr hosts

NB: This requires a perf2-15 instance.

rpco_pike This is the same as the rpco_newton job but built against the pike branch of RPC-O.

rpco_queens This is the same as the rpco_newton and rpco-pike jobs but built against the queens branch of RPC-O.

rpco_rocky This is the same as the rpco_newton and rpco-pike jobs but built against the rocky branch of RPC-O.

keystone_rgw: A basic keystone integration test, that will run on each commit. Utilizing the swift client to ensure Keystone integration is working.

  • Keystone deployed from OpenStack-Ansible role
  • 2 x rgw hosts
  • 3 x osd hosts
  • 3 x mon hosts
  • 3 x mgr hosts

Additionally this test runs the FIO and RGW benchmarking playbooks to ensure they work, but does not run the MaaS playbooks.

Currently not supported for AIO

  • Trusty deployments - due to changes in losetup Trusty will not work with the current method.
  • Different Ceph versions.
  • Upgrade testing.