race in backup manager remove_export call

[Eric Harney]

In _detach_device in the backup manager, rpcapi.remove_export() is called to detach a volume. This then quickly allows subsequent operations to occur, but since remove_export() is called with a "cast" and may take a moment to complete, it can interfere with operations happening shortly after it.

This manifests in the lio-barbican CI job, where a call to "cinder-rtstool add-initiator" is made and the target is not found. It presumably affects other volume drivers in similar ways, maybe with varying frequency.

[Eric Harney (eharney)]

https://review.opendev.org/c/openstack/cinder/+/781788

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/cinder 18.0.0.0rc1 release candidate.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/cinder/+/785970
Committed: https://opendev.org/openstack/cinder/commit/34120d49808537d0523375b173db9c657114a4cb
Submitter: "Zuul (22348)"
Branch: stable/ussuri

commit 34120d49808537d0523375b173db9c657114a4cb
Author: Eric Harney <email address hidden>
Date: Fri Mar 19 16:01:33 2021 +0000

    Backup manager: Synchronously call remove_export

    Call the volume's rpcapi remove_export method with
    a "call" instead of a "cast" when detaching volumes
    in the backup manager.

    This prevents _detach_volume() from returning before
    the remove_export() has completed, which helps prevent
    problems where remove_export() races against a
    subsequent operation.

    Closes-Bug: #1920237

    Change-Id: I482c3d5520a7bbb9971942cdb4c4208052106d70
    (cherry picked from commit d6895f9d2aab84c116efc2f853bb5f9e57fac539)
    (cherry picked from commit 5402ac8e2d30872b526c3ac7d9f10645452fbf66)

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/train
Review: https://review.opendev.org/c/openstack/cinder/+/789859

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/cinder/+/789859
Committed: https://opendev.org/openstack/cinder/commit/3bb28c637e98078a4da2b6930e56bfce0cce9f73
Submitter: "Zuul (22348)"
Branch: stable/train

commit 3bb28c637e98078a4da2b6930e56bfce0cce9f73
Author: Eric Harney <email address hidden>
Date: Fri Mar 19 16:01:33 2021 +0000

    Backup manager: Synchronously call remove_export

    Call the volume's rpcapi remove_export method with
    a "call" instead of a "cast" when detaching volumes
    in the backup manager.

    This prevents _detach_volume() from returning before
    the remove_export() has completed, which helps prevent
    problems where remove_export() races against a
    subsequent operation.

    Closes-Bug: #1920237

    Change-Id: I482c3d5520a7bbb9971942cdb4c4208052106d70
    (cherry picked from commit d6895f9d2aab84c116efc2f853bb5f9e57fac539)
    (cherry picked from commit 5402ac8e2d30872b526c3ac7d9f10645452fbf66)
    (cherry picked from commit 34120d49808537d0523375b173db9c657114a4cb)

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/cinder 15.6.0 release.

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/cinder 16.4.0 release.

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/cinder 17.2.0 release.