PowerFlex: OpenStack Volume Creation Fails Intermittently

OpenStack creates and deletes thousands of small volumes during normal operations. During peak workload hours, creating and deleting volumes may not work.

Errors and screenshots were observed in this scenario where the 2.5 ratio was reached:

Impact

Unable to create new volumes.

[flexos]
backend_host=hostgroup
volume_driver=cinder.volume.drivers.dell_emc.powerflex.driver.PowerFlexDriver
san_login=admin
san_password=********
san_ip=X.X.X.X
powerflex_storage_pools=PD01:PD01_FG01
volume_backend_name=flexos
powerflex_allow_migration_during_rebuild=False
powerflex_allow_non_padded_volumes=False
powerflex_max_over_subscription_ratio=2.5
powerflex_rest_server_port=443
powerflex_round_volume_capacity=True
powerflex_server_api_version=True
powerflex_unmap_volume_before_deletion=False
san_thin_provision=True
driver_ssl_cert_verify=False
driver_ssl_cert_path=

The over-provisioning setting is causing the volume errors. Changing the value in the cinder.conf from 2.5 to a higher value, for example 3.5, allows the operations to work smoothly during peak hours.

The powerflex_max_over_subscription_ratio key was first introduced in Cinder version 17.4.1.dev13 (Victoria).

Fix:

Working as expected.

Impacted versions:

• ScaleIO 2.0.x and later
• VxFlex OS 3.0.x and later
• PowerFlex 3.5.x and later
• PowerFlex 4.x and later