All About Cinder Drivers

General Considerations

Cinder allows you to integrate various storage solutions into your OpenStack cloud. It does this by providing a stable interface for hardware providers to write drivers that allow you to take advantage of the various features that their solutions offer.

“Supported” drivers

In order to make it easier for you to assess the stability and quality of a particular vendor’s driver, The Cinder team has introduced the concept of a supported driver. These are drivers that:

  • have an identifiable driver maintainer

  • are included in the Cinder source code repository

  • use the upstream Cinder bug tracking mechanism

  • support the Cinder Required Driver Functions

  • maintain a third-party Continuous Integration system that runs the OpenStack Tempest test suite against their storage devices

    • this must be done for every Cinder commit, and the results must be reported to the OpenStack Gerrit code review interface

    • for details, see Driver Testing

In summary, there are two important aspects to a driver being considered as supported:

  • the code meets the Cinder driver specifications (so you know it should integrate properly with Cinder)

  • the driver code is continually tested against changes to Cinder (so you know that the code actually does integrate properly with Cinder)

The second point is particularly important because changes to Cinder can impact the drivers in two ways:

  • A Cinder change may introduce a bug that only affects a particular driver or drivers (this could be because many drivers implement functionality well beyond the Required Driver Functions). With a properly running and reporting third-party CI system, such a bug can be detected at the code review stage.

  • A Cinder change may exercise a new code path that exposes a driver bug that had previously gone undetected. A properly running third-party CI system will detect this and alert the driver maintainer that there is a problem.

Driver Compliance

The current policy for CI compliance is:

  • CIs must report on every patch, whether the code change is in their own driver code or not

  • The CI comments must be properly formatted to show up in the CI summary in Gerrit

Non-compliant drivers will be tagged as unsupported if:

  • No CI success reporting occurs within a two week span

  • The CI is found to not be testing the expected driver (CI runs using the default LVM driver, etc.)

  • Other issues are found but failed to be addressed in a timely manner

CI results are reviewed on a regular basis and if found non-compliant, a driver patch is submitted flagging it as ‘unsupported’. This can occur at any time during the development cycle. A driver can be returned to ‘supported’ status as soon as the CI problem is corrected.

We do a final compliance check around the third milestone of each release. If a driver is marked as ‘unsupported’, vendors have until the time of the first Release Candidate tag (two weeks after the third milestone) to become compliant, in which case the patch flagging the driver as ‘unsupported’ can be reverted. Otherwise, the driver will be considered ‘unsupported’ in the release.

The CI results are currently posted here: http://cinderstats.ivehearditbothways.com/cireport.txt

“Unsupported” drivers

A driver is marked as ‘unsupported’ when it is out of compliance.

Such a driver will log a warning message to be logged in the cinder-volume log stating that it is unsupported and deprecated for removal.

In order to use an unsupported driver, an operator must set the configuration option enable_unsupported_driver=True in the driver’s configuration section of cinder.conf or the Cinder service will fail to load.

If the issue is not corrected before the next release, the driver will be eligible for removal from the Cinder code repository per the standard OpenStack deprecation policy.

If the issue is corrected before the next release and the team maintaining the driver in question submits a patch marking the driver as ‘supported’, that patch is eligible (at the discretion of the cinder stable maintenance team) for backport to the most recent stable branch.

Note

The idea behind backporting ‘supported’ status is that reinstatement should happen very early in the next development cycle after the driver has been marked ‘unsupported’. For example, a driver is marked ‘unsupported’ in the Victoria release but CI issues are addressed early in the Wallaby development cycle; the patch marking the driver may then be proposed to stable/victoria. Thus the patch will be included in the first stable release of Victoria, and operators upgrading from Ussuri to this release will not have to change their configuration files.

Note the “at the discretion of the cinder stable maintenance team” qualification. One reason for this is that the third party CI systems typically run only on changes to the development branch. Thus if a driver’s CI is restored early in the development cycle when there have not been many code changes yet, the CI passing in the development branch can be interpreted as a proxy for CI in the most recent stable branch. Obviously, this interpretation becomes increasingly invalid as the development cycle progresses. Further, this interpretation does not extend to older stable branches.

Driver Removal

(Added January 2020)

As stated above, an unsupported driver is eligible for removal during the development cycle following the release in which it was marked ‘unsupported’. (For example, a driver marked ‘unsupported’ in the Ussuri release is eligible for removal during the development cycle leading up to the Victoria release.)

During the Ussuri development cycle, the Cinder team decided that drivers eligible for removal, at the discretion of the team, may remain in the code repository as long as they continue to pass OpenStack CI testing. When such a driver blocks the CI check or gate, it will be removed immediately. (This does not violate the OpenStack deprecation policy because such a driver’s deprecation period began when it was marked as ‘unsupported’.)

Note

Why the “at the discretion of the team” qualification? Some vendors may announce that they have no intention of continuing to support a driver. In that case, the Cinder team reserves the right to remove the driver as soon as the deprecation period has passed.

Thus, unsupported drivers may remain in the code repository for multiple releases following their declaration as ‘unsupported’. Operators should therefore take into account the length of time a driver has been marked ‘unsupported’ when deciding to deploy an unsupported driver. This is because as an unmaintained driver ages, updates and bugfixes to libraries and other software it depends on may cause the driver to fail unit and functional tests, making it subject to immediate removal.

The intent of this policy revision is twofold. First, it gives vendors a longer grace period in which to make the necessary changes to have their drivers reinstated as ‘supported’. Second, keeping these drivers in-tree longer should make life easier for operators who have deployed storage backends with drivers that have been marked as ‘unsupported’. Operators should keep the above points in mind, however, when deploying such a driver.

Current Cinder Drivers

The Cinder team maintains a page of the current drivers and what exactly they support in the Driver Support Matrix.

You may find more details about the current drivers on the Available Drivers page.

Additionally, the configuration reference for each driver provides even more information. See Volume drivers.