external ceph cinder volume config breaks volumes on ussuri upgrade

Train external ceph docs: https://docs.openstack.org/kolla-ansible/train/reference/storage/external-ceph-guide.html#cinder

[rbd-1]
rbd_ceph_conf=/etc/ceph/ceph.conf
rbd_user=cinder
backend_host=rbd:volumes
rbd_pool=volumes
volume_backend_name=rbd-1
volume_driver=cinder.volume.drivers.rbd.RBDDriver
rbd_secret_uuid = {{ cinder_rbd_secret_uuid }}

Ussuri made the integration simpler, adding the following to ceph.conf:

{% if cinder_backend_ceph | bool %}
[rbd-1]
volume_driver = cinder.volume.drivers.rbd.RBDDriver
volume_backend_name = rbd-1
rbd_pool = {{ ceph_cinder_pool_name }}
rbd_ceph_conf = /etc/ceph/ceph.conf
rbd_flatten_volume_from_snapshot = false
rbd_max_clone_depth = 5
rbd_store_chunk_size = 4
rados_connect_timeout = 5
rbd_user = {{ ceph_cinder_user }}
rbd_secret_uuid = {{ cinder_rbd_secret_uuid }}
report_discard_supported = True
image_upload_use_cinder_backend = True
{% endif %}

This is missing backend_host=rbd:volumes. There is a related Tripleo bug [1], which explains that this option is used to set the same host for all backends in an environment with multiple cinder-volume services representing a single storage cluster.

[1] https://bugs.launchpad.net/bugs/1753596

Actually, this OpenStack Ansible bug suggests that backend_host is not recommended: https://bugs.launchpad.net/cinder/+bug/1837403. We might need to set [DEFAULT] cluster to use active/active cinder-volume though.

OSA highlights the 'cinder-manage volume update_host' command in their release notes. It's not clear to me what the right solution is at this point though.

Fix proposed to branch: master
Review: https://review.opendev.org/763011

High level cinder active/active spec: https://specs.openstack.org/openstack/cinder-specs/specs/mitaka/cinder-volume-active-active-support.html

The relevant openstack-discuss ML thread: http://lists.openstack.org/pipermail/openstack-discuss/2020-November/018838.html (thank you all for answering our questions!)

Failure testing in https://review.opendev.org/c/openstack/kolla-ansible/+/767951

Related fix proposed to branch: stable/victoria
Review: https://review.opendev.org/c/openstack/kolla-ansible/+/808003

Related fix proposed to branch: stable/ussuri
Review: https://review.opendev.org/c/openstack/kolla-ansible/+/808004

Reviewed: https://review.opendev.org/c/openstack/kolla-ansible/+/808003
Committed: https://opendev.org/openstack/kolla-ansible/commit/f97e752018affcb81604230e7e9b0101960cec83
Submitter: "Zuul (22348)"
Branch: stable/victoria

commit f97e752018affcb81604230e7e9b0101960cec83
Author: Radosław Piliszek <email address hidden>
Date: Fri Dec 18 21:46:35 2020 +0100

    [CI] Cinder upgrade testing

    To gain visibility into how our upgrades affect existing Cinder
    volumes, a new testing path is required.
    This patch adds it.

    Additionally, it refactors the repeated actions and ensures that
    we wait for volume deletions as well.

    Change-Id: Ic08d461e6fdf91c378a87860765a489c2f86d690
    Related-Bug: #1904062
    (cherry picked from commit 62b8c6b68413330da032d14b45c6fbd340ec9e2d)

Change abandoned by "Radosław Piliszek <email address hidden>" on branch: stable/ussuri
Review: https://review.opendev.org/c/openstack/kolla-ansible/+/808004
Reason: ussuri going em, thus /us not extending CI

Hi,

can someone provide an update on this bug?

Because we hit this in real life deployments:

1. volume has a specific os controller node as `os-vol-host-attr:host`
2. that os controller node gets maintenance
3. vm instance with attached volume gets deleted
3. nova throws: openstack.nova nova-compute c84d9828-1277-457a-828d-db7dc3c03216 [instance: d5832d72-0b70-422e-ba94-12b24f1a75e1] Ignoring unknown cinder exception for volume 615d4759-bed3-4a84-91a8-8fce612bfb2a: Gateway Time-out (HTTP 504): cinderclient.exceptions.ClientException: Gateway Time-out (HTTP 504)

the volume than still exists and still claims to be attached to a nonexistant vm.

we can of course clean this up manually.

afaiu there needs to be added some active/active cinder deployment with a coordinator service like pacemaker or etcd to kolla-ansible?

would it be possible to mimic somehow what tripleo does? as far as I understand they have implemented active active cinder deployment with etcd coordinator?

Fix proposed to branch: stable/yoga
Review: https://review.opendev.org/c/openstack/kolla-ansible/+/847151

Change abandoned by "Radosław Piliszek <email address hidden>" on branch: stable/yoga
Review: https://review.opendev.org/c/openstack/kolla-ansible/+/847151
Reason: proper docs are the way to go

Change abandoned by "Michal Nasiadka <email address hidden>" on branch: stable/yoga
Review: https://review.opendev.org/c/openstack/kolla-ansible/+/847151