Lightbits LightOS Cinder Driver

The Lightbits(TM) LightOS(R) OpenStack driver enables OpenStack clusters to use LightOS clustered storage servers. This documentation explains how to configure Cinder for use with the Lightbits LightOS storage backend system.

Supported operations

  • Create volume

  • Delete volume

  • Attach volume

  • Detach volume

  • Create image from volume

  • Live migration

  • Volume replication

  • Thin provisioning

  • Multi-attach

  • Supported vendor driver

  • Extend volume

  • Create snapshot

  • Delete snapshot

  • Create volume from snapshot

  • Create volume from volume (clone)

  • Active active deployment support

LightOS OpenStack Driver Components

The LightOS OpenStack driver has three components: - Cinder driver - Nova libvirt volume driver - os_brick initiator connector

In addition, it requires the LightOS discovery-client, provided with LightOS. The os_brick connector uses the LightOS discovery-client to communicate with LightOS NVMe/TCP discovery services.

The Cinder Driver

The Cinder driver integrates with Cinder and performs REST operations against the LightOS cluster. To enable the driver, add the following to Cinder’s configuration file

enabled_backends = lightos,<any other storage backend you use>

and

[lightos]
volume_driver = cinder.volume.drivers.lightos.LightOSVolumeDriver
volume_backend_name = lightos
lightos_api_address = <TARGET_ACCESS_IPS>
lightos_api_port = 443
lightos_jwt=<LIGHTOS_JWT>
lightos_default_num_replicas = 3
lightos_default_compression_enabled = False
lightos_api_service_timeout=30

  • TARGET_ACCESS_IPS are the LightOS cluster nodes access IPs. Multiple nodes should be separated by commas. For example: lightos_api_address = 192.168.67.78,192.168.34.56,192.168.12.17. These IPs are where the driver looks for the LightOS clusters REST API servers.

  • LIGHTOS_JWT is the JWT (JSON Web Token) that is located at the LightOS installation controller. You can find the jwt at ~/lightos-default-admin-jwt.

  • The default number of replicas for volumes is 3, and valid values for lightos_default_num_replicas are 1, 2, or 3.

  • The default compression setting is False (i.e., data is uncompressed). The default compression setting can also be True to indicate that new volumes should be created compressed, assuming no other compression setting is specified via the volume type. To control compression on a per-volume basis, create volume types for compressed and uncompressed, and use them as appropriate.

  • The default time to wait for API service response is 30 seconds per API endpoint.

Creating volumes with non-default compression and number of replicas settings can be done through the volume types mechanism. To create a new volume type with compression enabled:

$ openstack volume type create --property compression='<is> True' volume-with-compression

To create a new volume type with one replica:

$ openstack volume type create --property lightos:num_replicas=1 volume-with-one-replica

To create a new type for a compressed volume with three replicas:

$ openstack volume type create --property compression='<is> True' --property lightos:num_replicas=3 volume-with-three-replicas-and-compression

Then create a new volume with one of these volume types:

$ openstack volume create --size <size> --type <type name> <vol name>

NVNe/TCP and Asymmetric Namespace Access (ANA)

The LightOS clusters expose their volumes using NVMe/TCP Asynchronous Namespace Access (ANA). ANA is a relatively new feature in the NVMe/TCP stack in Linux but it is fully supported in Ubuntu 20.04. Each compute host in the OpenStack cluster needs to be ANA-capable to provide OpenStack VMs with LightOS volumes over NVMe/TCP. For more information on how to set up the compute nodes to use ANA, see the CentOS Linux Cluster Client Software Installation section of the Lightbits(TM) LightOS(R) Cluster Installation and Initial Configuration Guide.

Note

In the current version, if any of the cluster nodes changes its access IPs, the Cinder driver’s configuration file should be updated with the cluster nodes access IPs and restarted. As long as the Cinder driver can access at least one cluster access IP it will work, but will be susceptible to cluster node failures.

Driver options

The following table contains the configuration options supported by the Lightbits LightOS Cinder driver.

Description of Lightbits LightOS configuration options

Configuration option = Default value

Description

lightos_api_address = None

(List of IPAddress) The IP addresses of the LightOS API servers separated by commas.

lightos_api_port = 443

(Port(min=0, max=65535)) The TCP/IP port at which the LightOS API endpoints listen. Port 443 is used for HTTPS and other values are used for HTTP.

lightos_api_service_snapshots_max_calls = 5

(Integer) The maximum number of calls to the LightOS when creating snapshots. The default is 5 calls.

lightos_api_service_timeout = 30

(Integer) The default amount of time (in seconds) to wait for an API endpoint response.

lightos_default_compression_enabled = False

(Boolean) Set to True to create new volumes compressed assuming no other compression setting is specified via the volumes type.

lightos_default_num_replicas = 3

(Integer(min=1, max=3)) The default number of replicas to create for each volume.

lightos_jwt = None

(String) JWT to be used for volume and snapshot operations with the LightOS cluster. Do not set this parameter if the cluster is installed with multi-tenancy disabled.

lightos_use_ipacl = True

(Boolean) IPACL work in conjunction with the standard NVME ACL. A host must be in both the IPACL and the ACL of a volume to access that volume. Cinder always sets the volume`s ACL. If lightos_use_ipacl is set to True, Cinder will also add the host`s IP addresses to a volume IPACL. If set to False, any IP address may access the volume. The default is True.

Active active deployment support

To enable active-active deployment, follow these steps:

  1. Activate the active-active mode by setting the “cluster” option in the “DEFAULT” section.

  2. Configure the Distributed Lock Manager (DLM) such as Redis or etcd in the “coordination” section.

These options should be added to the cinder.conf file:

[DEFAULT]
cluster = <cluster_name>

[coordination]
backend_url = <coordination_backend_url>

For more detailed instructions, please refer to the guidelines at:: https://docs.openstack.org/cinder/latest/contributor/high_availability.html