openstack/devstack-plugin-ceph - devstack-plugin-ceph - OpenDev: Free Software Needs Free Tools

The backing disk currently created is 8GB. devstack-gate
sets that to 24GB when running Tempest. We're seeing ceph
job failures due to NoValidHost in the cinder scheduler
because 8GB isn't enough capacity for Tempest runs. So this
change uses the same backing disk size for the ceph jobs as
we get in the default devstack setup, which uses LVM.
Depends-On: I71be308c8373e9ac429b901c374100c6b3c1e59d
Change-Id: I788eefa6c1d427bf51d2d3d40be4abe0336443e7
Closes-Bug: #1645530 

Add config file argument to ceph commands in cleanup_ceph_remote
function so that it will cleanup properly if CEPH_CONF_FILE has
a non-default value.
Change-Id: I0c6dc3494bcf700364af0c578cc9a60b082427d5
Closes-Bug: #1643513 

This commit introduces the support of deploying a containerized ceph
cluster. It relies on the ceph/demo container image available as part of
the ceph-docker project (https://github.com/ceph/ceph-docker).
To enable this scenario, just set CEPH_CONTAINERIZED to True, all the
previous available options will work. So settings like
CEPH_LOOPBACK_DISK_SIZE continue to work.
Change-Id: Ie10155140448c04b88b9920381b54505f7359585
Signed-off-by: Sébastien Han <seb@redhat.com>

Xenial currently ships with Ceph 10.2.2 which has a known issue that
can result in updated images failing to be refreshed when the cluster
is overloaded and drops update notifications [1]. This is most likely
the cause for recent Ceph gate failures [2].
Disabling the new Jewel+ default image feature of exclusive-lock
(revert to Hammer default behavior) should be a workaround for this
issue.
The RBD default features can be overridden via CEPH_RBD_DEFAULT_FEATURES
[1] http://tracker.ceph.com/issues/16404
[2] https://bugs.launchpad.net/cinder/+bug/1627220
Change-Id: Id7fab306e918ba7566ebe2a765d4345fea6c5653
Closes-Bug: #1627878
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

This patch introduce new option GLANCE_RGW_BACKEND, if it set to
true and radosgw enabled, glance is configured with swift backend
options.
Also create glance-swift user with ResellerAdmin role to interact
with radosgw.
CEPH_RGW_KEYSTONE_API_VERSION option was added to specify which
keystone API versioni radosgw will use, default v2.0.
Change-Id: I2cacf511e48720896aa621201d6188bd1d370b6e

Since Ceph supports more architectures such as Aarch64.
Change-Id: I3cb05c4a3eb929e9648ecc05c4f0ba042f9e86ad
Signed-off-by: Kevin Zhao <kevin.zhao@linaro.org>

Signed-off-by: Kevin Zhao <kevin.zhao@linaro.org>
Change-Id: Ifc75a97fa7795647494ab92e94b4dec49064eb87

we remove the download.ceph.com repo during cleanup.
Change-Id: Ia2a62a6307b61623b72dbd7ee391d11cbd27f660
Signed-off-by: Sébastien Han <seb@redhat.com>

http://ceph.com/debian-hammer redirects to
http://download.ceph.com/debian-hammer/
However while setting: CEPH_RELEASE=jewel http://ceph.com/debian-jewel
does not redirect to http://download.ceph.com/debian-jewel and stack
fails. So fixing the repo url by directly using
http://download.ceph.com
Change-Id: Ib8e80e43674c6d8ffd01bc6af56e66e9e254213b
Signed-off-by: Sébastien Han <seb@redhat.com>

We just want to set new defaults, not force the
variables. This way we can mix ceph and other
backends with like:
CINDER_ENABLED_BACKENDS=lvm:lvm-driver,ceph:ceph-driver
Change-Id: I99cc047bce2f584c2b54196a9c84267d2c686231

OpenStack CI would prefer to not have the ceph repo package mirror
enabled by default for all jobs. Instead the mirror config will be
provided in a directory that apt won't read and can by copied/linked
into the dir that apt will read.
This gives testing greater flexibility and prevents growing dependencies
on non standard repos in jobs that don't intend on growing those
dependencies.
Change-Id: Icd87013fa74601268c19a8b79abf6b7baeaca9db
Depends-On: I410f45064080e1bb93684899028e9af34bd555a1

Change-Id: Ia1adcad17c447c5ee03b9e79a9df341f877ce189

Currently the devstack ceph plugin uses installation,
configuration, startup, and shutdown commands that do not always
work correctly for Ceph releases after hammer or with systemd
init system.
Check the ceph version and the init system and use appropriate
commands.
Co-Authored-By: Ramana Raja <rraja@redhat.com>
Change-Id: I25d8816740c6008b076986f3677c299ad9843c30

Radosgw relies on shared secret token admin, passed via SERVICE_TOKEN
variable, but token admin was deprecated, and devstack doesn't
configure shared secret.
This change switch on using service credentials instead, creates
'radosgw' service user with admin privileges and configure radosgw
with them.
Change-Id: I12d4ea4021b93be73358f0aa68a32879af8fda53

Shell scripts are whitespace sensitive, in particular
a= $(echo A)
a=\
 $(echo A)
tries to run the 'A' executable somewhere in $PATH with environment variable
`a' set to an empty string.
Change-Id: I57648e597a2875061f38c8669370598348e2c6f9

In a lab setting it can be useful to allow multiple manila devstacks,
each with its own CephFS filesystem, to share a common REMOTE_CEPH
cluster.
Parameterize the name of the CephFS filesystem that devstack creates
and enable creation of multiple filesystems if appropriate variables
are set in local.conf.
Note that since separate filesystems require separate data and metadata
pools, one also needs to set values for the extant CEPHFS_DATA_POOL and
CEPHFS_METADATA_POOL parameters.
Change-Id: I67a88b75d06f04621cb0dac726f307a06ee6113b

Although the devstack ceph plugin uses variables - CEPHFS_DATA_POOL and
CEPHFS_METADATA_POOL - as arguments to the 'osd pool create' commands
that provision the pools required for manila to use the CephFS Native
backend, the overrides to the default values for these variables sets
both of them to $CEPHFS_CEPH_POOL. This setting is counter-intuitive,
and worse, forces use of a single pool for both data and metadata in the
case both default values are overridden.
Use CEPHFS_DATA_POOL and CEPHFS_METADATA_POOL as the override
variables for the data pool and metadata pool, respectively.
Change-Id: Iae2859388ce5e74b6f532cf2cf7809aaa09746cb
Closes-Bug: 1593829

The CEPH_DATA_DIR variable holds the location of Ceph drives and
objects. It defaults to '/var/lib/ceph'.
Replace hardcoded occurrences of '/var/lib/ceph' with
'${CEPH_DATA_DIR}'.
Change-Id: I2570e37f9135d853854600acfbfff3b768e4f0e3

By default bashate checks for lines longer than 80
characters and issues E006 warnings for these.
Suppress the E006 warnings since we don't enforce
this line length standard in OpenStack shell
scripts so that other warnings will be more evident.
Change-Id: Icdf32cb7e3d810d2f8644674d3516ed808c873a5

Restacking with ceph plugin fails when trying to install
the cinder user's ceph authorization key in libvirt's list
of secrets because there is already a secret in that list
with the same uuid. The function _undefine_virsh_secret
is run on cleanup and should have removed that key from
the libvirt secret list.
Fix a typo in the variable name for the uuid that identifies
the authorization key in the secret list so that the
cleanup routine removes the key as intended.
Closes-Bug: #1570665
Change-Id: I19d2c20979dca49d73c427c5b1413229ff39f394

The rule_id line is formatted as:
 "rule_id": 0,
Which means the id integer is the second column of the parsed output,
not the third.
Since a third column doesn't exist, the call to cut returns nothing,
leaving $RULE_ID invalid, which then causes the subsequent ruleset
commands to fail.
We don't see this normally because $CEPH_REPLICAS is set to 1 by
default.
Change-Id: I0cc57f5abc023cb8b2faeea940f28ab54cd290ff
Related-Bug: #1581159 

Hammer LTS 0.94.6 is currently available for Fedora 22 and 23 [1].
Fedora 21 is EOL and should be removed from the list [2].
[1] https://admin.fedoraproject.org/pkgdb/package/rpms/ceph/
[2] https://fedoraproject.org/wiki/Releases
Change-Id: I5fc7f25ed145c29cadbe03d945511aaa12c54310

In commit bb773e7285 the key for
verifying release packages from ceph.com was removed when support for
the ceph package mirrors was added. However, it is still needed for
non-gate case of fetching packages from ceph.com, so this adds it
back in that case.
Closes-Bug: #1578494
Change-Id: I18f8927f42710e9365c246c55f26f53403965ee1

For the ceph job to be voting, infra requires that we pull packages from
a pre-configured repository that's mirrored and reliable for gate jobs.
They have arranged for ceph.com packages to be mirrored and will store a
link to those packages in /etc/apt/sources.list.d/ceph.list. If this
file is present, we should use it. If not, then we're free to update
the sources list to fetch the latest packages.
This approach is required to regain voting status in the gate.
Change-Id: I5f811c917e6964c33440ed644597c9e485963626

The changes in Ceph that were needed to make manila's cephfs_native
driver work are available in the Jewel branch. It's no longer
necessary to use the 'wip-manila' branch.
Also, with Jewel branch, manila's libcephfs client needs to be root
to perform certain manila operations. So set manila's client as root.
Change-Id: I715f79a6801b12522c047b1df4d950c99a1ffad5

The cephfs_native driver CI does not need the service image that is
being uploaded. Disable the upload until a need arises, for example,
when the CI does scenario testing.
Change-Id: I1ed0af5fd61829a075a41df2939f0377deb351b3

This patch elaborates on fine-grained control of each service by listing
examples instead of using pseudocode - which could be confusing to
newcomers.
Change-Id: I9ba3c7b4460aa840e88c2c8c82aa70ba06c8302c

If we want to put the disk image into a tmpfs, changing CEPH_DATA_DIR
won't work because some file operation needed by ceph are not allowed
on tmpfs.
So, this change allows to change the disk image path only instead of the
whole CEPH_DATA_DIR.
Change-Id: If92f0bdca16f866a475820cf71a30ded28fb2572

So if the directory exists the stack does not fail.
Change-Id: Ibdd2883b5328d58d7c2a74d97dfe2e0ea87d65ee
Signed-off-by: Sébastien Han <seb@redhat.com>

Since we're about to be voting and I expect a few more people to be
interested in the code, I think now is the best time to fix the
indenting. It's a large patch, but there should be no functional
changes.
I removed one unused function and renamed internal-only function to be
prefixed with '_'.
Change-Id: Ibb9b26f0cc8719be4fec121c1b9aa279938aebd7

It seems to be a duplicate and I don't think we need both.
Change-Id: Iaf7b6fe2b633fd6dd2d228960e48de238c8b1c33

This was renamed from drivers.cephfs to
drivers.cephfs.cephfs_native at the
request of the Manila maintainers.
Change-Id: I7ac2299ad323229418761bb06ee7891a48d87877

Currently glance v2 doesn't allow to specify custom
locations for image by default, it returns 403.
To enable this 'show_multiple_locations' param must
be set to True.
When Nova is doing snapshots with Ceph it provides
location to activate image, which fails when we use
Glance v2.
To fix it we change the default param to True.
'show_image_direct_url' will be overridden.
Related to blueprint use-glance-v2-api
Change-Id: Id0f1c398b8b48f2ffc2488b29bc7cbd279069337

manila's cephfs-native CI tests cephfs driver, which does not support
glusterfs protocol. The glusterfs protocol is enabled by a manila's
tempest config option by default. Modify the config option to disable
glusterfs protocol.
Change-Id: Ia7a226d6a8b0ba5aeedcdf34e8a6f9a8dbbfda40
Closes-Bug: #1541377