Pure Storage iSCSI, Fibre Channel and NVMe volume drivers

The Pure Storage FlashArray volume drivers for OpenStack Block Storage interact with configured Pure Storage arrays and support various operations.

Support for iSCSI storage protocol is available with the PureISCSIDriver Volume Driver class, Fibre Channel with the PureFCDriver and NVMe-ROCE or NVMe-TCP with the PureNVMEDriver.

iSCSI, Fibre Channel and NVMe-RoCE drivers are compatible with FlashArrays that support the REST API version 1.16 and higher (Purity 5.2.0 and newer). The NVMe-TCP driver is compatible with FlashArrays that are running Purity 6.4.2 and higher. Some features may require newer versions of Purity.

Limitations and known issues

If you do not set up the nodes hosting instances to use multipathing, all network connectivity will use a single physical port on the array. In addition to significantly limiting the available bandwidth, this means you do not have the high-availability and non-disruptive upgrade benefits provided by FlashArray. Multipathing must be used to take advantage of these benefits.

Supported operations

  • Create, delete, attach, detach, retype, clone, and extend volumes.

  • Create a volume from snapshot.

  • Create, list, and delete volume snapshots.

  • Create, list, update, and delete consistency groups.

  • Create, list, and delete consistency group snapshots.

  • Revert a volume to a snapshot.

  • Manage and unmanage a volume.

  • Manage and unmanage a snapshot.

  • Get volume statistics.

  • Create a thin provisioned volume.

  • Replicate volumes to remote Pure Storage array(s)

QoS support for the Pure Storage drivers include the ability to set the following capabilities in the OpenStack Block Storage API cinder.api.contrib.qos_spec_manage qos specs extension module:

  • maxIOPS - Maximum number of IOPs allowed for volume. Range: 100 - 100M

  • maxBWS - Maximum bandwidth limit in MB/s. Range: 1 - 524288 (512GB/s)

The qos keys above must be created and asscoiated to a volume type. For information on how to set the key-value pairs and associate them with a volume type see the volume qos section in the OpenStack Client command list.

Configure OpenStack and Purity

You need to configure both your Purity array and your OpenStack cluster.

Note

These instructions assume that the cinder-api and cinder-scheduler services are installed and configured in your OpenStack cluster.

Configure the OpenStack Block Storage service

In these steps, you will edit the cinder.conf file to configure the OpenStack Block Storage service to enable multipathing and to use the Pure Storage FlashArray as back-end storage.

  1. Install Pure Storage PyPI module. A requirement for the Pure Storage driver is the installation of the Pure Storage Python SDK version 1.4.0 or later from PyPI.

    $ pip install purestorage

  2. Retrieve an API token from Purity. The OpenStack Block Storage service configuration requires an API token from Purity. Actions performed by the volume driver use this token for authorization. Also, Purity logs the volume driver’s actions as being performed by the user who owns this API token.

    If you created a Purity user account that is dedicated to managing your OpenStack Block Storage volumes, copy the API token from that user account.

    Use the appropriate create or list command below to display and copy the Purity API token:

    • To create a new API token:

      $ pureadmin create --api-token USER

      The following is an example output:

      $ pureadmin create --api-token pureuser
Name      API Token                             Created
pureuser  902fdca3-7e3f-d2e4-d6a6-24c2285fe1d9  2014-08-04 14:50:30

    • To list an existing API token:

      $ pureadmin list --api-token --expose USER

      The following is an example output:

      $ pureadmin list --api-token --expose pureuser
Name      API Token                             Created
pureuser  902fdca3-7e3f-d2e4-d6a6-24c2285fe1d9  2014-08-04 14:50:30

  3. Copy the API token retrieved (902fdca3-7e3f-d2e4-d6a6-24c2285fe1d9 from the examples above) to use in the next step.

  4. Edit the OpenStack Block Storage service configuration file. The following sample /etc/cinder/cinder.conf configuration lists the relevant settings for a typical Block Storage service using a single Pure Storage array:

    [DEFAULT]
enabled_backends = puredriver-1
default_volume_type = puredriver-1

[puredriver-1]
volume_backend_name = puredriver-1
volume_driver = PURE_VOLUME_DRIVER
san_ip = IP_PURE_MGMT
pure_api_token = PURE_API_TOKEN
use_multipath_for_image_xfer = True

    Replace the following variables accordingly:

    PURE_VOLUME_DRIVER

    Use cinder.volume.drivers.pure.PureISCSIDriver for iSCSI, cinder.volume.drivers.pure.PureFCDriver for Fibre Channel or cinder.volume.drivers.pure.PureNVMEDriver for NVME connectivity.

    If using the NVME driver, specify the pure_nvme_transport value. Supported values are roce or tcp.

    IP_PURE_MGMT

    The IP address of the Pure Storage array’s management interface or a domain name that resolves to that IP address.

    PURE_API_TOKEN

    The Purity Authorization token that the volume driver uses to perform volume management on the Pure Storage array.

Note

The volume driver automatically creates Purity host objects for initiators as needed. If CHAP authentication is enabled via the use_chap_auth setting, you must ensure there are no manually created host objects with IQN’s that will be used by the OpenStack Block Storage service. The driver will only modify credentials on hosts that it manages.

Note

If using the PureFCDriver it is recommended to use the OpenStack Block Storage Fibre Channel Zone Manager.

Volume auto-eradication

To enable auto-eradication of deleted volumes, snapshots, and consistency groups on deletion, modify the following option in the cinder.conf file:

pure_eradicate_on_delete = true

By default, auto-eradication is disabled and all deleted volumes, snapshots, and consistency groups are retained on the Pure Storage array in a recoverable state for 24 hours from time of deletion.

Setting host personality

The host personality determines how the Purity system tunes the protocol used between the array and the initiator. To ensure the array works optimally with the host, set the personality to the name of the host operating or virtual memory system. Valid values are aix, esxi, hitachi-vsp, hpux, oracle-vm-server, solaris, and vms. If your system is not listed as one of the valid host personalities, do not set the option. By default, the host personality is not set.

To set the host personality, modify the following option in the cinder.conf file:

pure_host_personality = <personality>

Note

pure_host_personality is available from Purity REST API version 1.14, and affects only newly-created hosts.

SSL certification

To enable SSL certificate validation, modify the following option in the cinder.conf file:

driver_ssl_cert_verify = true

By default, SSL certificate validation is disabled.

To specify a non-default path to CA_Bundle file or directory with certificates of trusted CAs:

driver_ssl_cert_path = Certificate path

Note

This requires the use of Pure Storage Python SDK > 1.4.0.

Replication configuration

Add the following to the back-end specification to specify another Flash Array to replicate to:

[puredriver-1]
replication_device = backend_id:PURE2_NAME,san_ip:IP_PURE2_MGMT,api_token:PURE2_API_TOKEN,type:REPLICATION_TYPE

Where PURE2_NAME is the name of the remote Pure Storage system, IP_PURE2_MGMT is the management IP address of the remote array, and PURE2_API_TOKEN is the Purity Authorization token of the remote array.

The REPLICATION_TYPE value for the type key can be either sync or async

If the type is sync volumes will be created in a stretched Pod. This requires two arrays pre-configured with Active Cluster enabled. You can optionally specify uniform as true or false, this will instruct the driver that data paths are uniform between arrays in the cluster and data connections should be made to both upon attaching.

Note that more than one replication_device line can be added to allow for multi-target device replication.

To enable 3-site replication, ie. a volume that is synchronously replicated to one array and also asynchronously replicated to another then you must supply two, and only two, replication_device lines, where one has type of sync and one where type is async. Additionally, the parameter pure_trisync_enabled must be set True.

A volume is only replicated if the volume is of a volume-type that has the extra spec replication_enabled set to <is> True. You can optionally specify the replication_type key to specify <in> sync or <in> async or <in> trisync to choose the type of replication for that volume. If not specified it will default to async.

To create a volume type that specifies replication to remote back ends with async replication:

$ openstack volume type create ReplicationType
$ openstack volume type set --property replication_enabled='<is> True' ReplicationType
$ openstack volume type set --property replication_type='<in> async' ReplicationType

The following table contains the optional configuration parameters available for async replication configuration with the Pure Storage array.

Pure Storage replication configuration options

Option

Description

Default

pure_replica_interval_default

Snapshot replication interval in seconds.

3600

pure_replica_retention_short_term_default

Retain all snapshots on target for this time (in seconds).

14400

pure_replica_retention_long_term_per_day_default

Retain how many snapshots for each day.

3

pure_replica_retention_long_term_default

Retain snapshots per day on target for this time (in days).

7

pure_replication_pg_name

Pure Protection Group name to use for async replication (will be created if it does not exist).

cinder-group

pure_replication_pod_name

Pure Pod name to use for sync replication (will be created if it does not exist).

cinder-pod

Note

failover-host is only supported from the primary array to any of the multiple secondary arrays, but subsequent failover-host is only supported back to the original primary array.

Note

pure_replication_pg_name and pure_replication_pod_name should not be changed after volumes have been created in the Cinder backend, as this could have unexpected results in both replication and failover.

Automatic thin-provisioning/oversubscription ratio

This feature allows the driver to calculate the array oversubscription ratio as (total provisioned/actual used). By default this feature is enabled.

To disable this feature and honor the hard-coded configuration option max_over_subscription_ratio add the following option in the cinder.conf file:

[puredriver-1]
pure_automatic_max_oversubscription_ratio = False

Note

Arrays with very good data reduction rates (compression/data deduplication/thin provisioning) can get very large oversubscription rates applied.

Scheduling metrics

A large number of metrics are reported by the volume driver which can be useful in implementing more control over volume placement in multi-backend environments using the driver filter and weighter methods.

Metrics reported include, but are not limited to:

total_capacity_gb
free_capacity_gb
provisioned_capacity
total_volumes
total_snapshots
total_hosts
total_pgroups
writes_per_sec
reads_per_sec
input_per_sec
output_per_sec
usec_per_read_op
usec_per_read_op
queue_depth
replication_type

Note

All total metrics include non-OpenStack managed objects on the array.

In conjunction with QOS extra-specs, you can create very complex algorithms to manage volume placement. More detailed documentation on this is available in other external documentation.

Configuration Options

The following list all Pure driver specific configuration options that can be set in cinder.conf:

Description of Pure configuration options

Configuration option = Default value

Description

pure_api_token = None

(String) REST API authorization token.

pure_automatic_max_oversubscription_ratio = True

(Boolean) Automatically determine an oversubscription ratio based on the current total data reduction values. If used this calculated value will override the max_over_subscription_ratio config option.

pure_eradicate_on_delete = False

(Boolean) When enabled, all Pure volumes, snapshots, and protection groups will be eradicated at the time of deletion in Cinder. Data will NOT be recoverable after a delete with this set to True! When disabled, volumes and snapshots will go into pending eradication state and can be recovered.

pure_host_personality = None

(String(choices=[‘aix’, ‘esxi’, ‘hitachi-vsp’, ‘hpux’, ‘oracle-vm-server’, ‘solaris’, ‘vms’, None])) Determines how the Purity system tunes the protocol used between the array and the initiator.

pure_iscsi_cidr = 0.0.0.0/0

(String) CIDR of FlashArray iSCSI targets hosts are allowed to connect to. Default will allow connection to any IPv4 address. This parameter now supports IPv6 subnets. Ignored when pure_iscsi_cidr_list is set.

pure_iscsi_cidr_list = None

(List of String) Comma-separated list of CIDR of FlashArray iSCSI targets hosts are allowed to connect to. It supports IPv4 and IPv6 subnets. This parameter supersedes pure_iscsi_cidr.

pure_nvme_cidr = 0.0.0.0/0

(String) CIDR of FlashArray NVMe targets hosts are allowed to connect to. Default will allow connection to any IPv4 address. This parameter now supports IPv6 subnets. Ignored when pure_nvme_cidr_list is set.

pure_nvme_cidr_list = None

(List of String) Comma-separated list of CIDR of FlashArray NVMe targets hosts are allowed to connect to. It supports IPv4 and IPv6 subnets. This parameter supersedes pure_nvme_cidr.

pure_nvme_transport = roce

(String(choices=[‘roce’, ‘tcp’])) The NVMe transport layer to be used by the NVMe driver.

pure_replica_interval_default = 3600

(Integer) Snapshot replication interval in seconds.

pure_replica_retention_long_term_default = 7

(Integer) Retain snapshots per day on target for this time (in days.)

pure_replica_retention_long_term_per_day_default = 3

(Integer) Retain how many snapshots for each day.

pure_replica_retention_short_term_default = 14400

(Integer) Retain all snapshots on target for this time (in seconds.)

pure_replication_pg_name = cinder-group

(String) Pure Protection Group name to use for async replication (will be created if it does not exist).

pure_replication_pod_name = cinder-pod

(String) Pure Pod name to use for sync replication (will be created if it does not exist).

pure_trisync_enabled = False

(Boolean) When enabled and two replication devices are provided, one each of types sync and async, this will enable the ability to create a volume that is sync replicated to one array and async replicated to a separate array.

pure_trisync_pg_name = cinder-trisync

(String) Pure Protection Group name to use for trisync replication leg inside the sync replication pod (will be created if it does not exist).