Matt Riedemann [Tue, 10 Nov 2015 20:55:46 +0000 (12:55 -0800)]
Test for object version hash changes
Use the ObjectVersionChecker fixture from oslo.versionedobjects to check
when Cinder's registered objects have a change (new remotable method,
new field, etc) that require a version bump.
Fixes some registration problems along the way:
1. CinderObject doesn't need to be registered.
2. TestObject should be registered where it's used, not globally.
Removes deps=requirements.txt from tox.ini since it's not necessary and
prevents us from pip installing oslo.versionedobjects[fixtures].
Matt Riedemann [Tue, 10 Nov 2015 23:45:29 +0000 (15:45 -0800)]
Fix cinder objects unit test registration
The base test case was extending test code from oslo_versionedobjects
which is bad because (1) it was adding test objects from the o.vo
registry into the cinder object registry and (2) it was extending
internal private test classes from o.vo which are subject to change and
would break cinder.
This cleans up the base test case and object registration to basically
do it the same way that nova does.
It also moves the duplicate _compare() method into the base class and
fixes a bug where db obj fields that were lists were compared
incorrectly against object lists.
This will allow us to verify object version hashes using o.vo's
ObjectVersionChecker fixture in a follow on change without the
extraneous object registrations from o.vo messing with the results.
Kendall Nelson [Mon, 9 Nov 2015 16:52:43 +0000 (10:52 -0600)]
Removal of deprecated NPIV option in Storwize
The storwize_svc_npiv_compatibility_mode option was deprecated and
set to be removed during the M release. This patch removes the
option, the associated code, and the test cases for the associated
code. This patch also removes the unused import versionutils.
Shay Halsband [Tue, 3 Nov 2015 14:33:03 +0000 (16:33 +0200)]
XtremIO fix create CG from src flow
Fix a bug during volume creation with CG,
that prevented the volume from being added to the CG on the array.
Query array snapshots names for CG snapshot before copying them
to a new volume.
Add unittest for create_cg_from_src.
Kurt Martin [Fri, 6 Nov 2015 16:42:50 +0000 (08:42 -0800)]
3PAR drivers volume size conversion is incorrect
The 3PAR drivers conversion of the volume size to gibibytes (GiBs) is
incorrect, resulting in slightly smaller volumes than expected. The
drivers were assuming the input was gigabytes (GBs) when in fact the
input was gibibytes (GiBs) as documented by the cinder API.
Michał Dulko [Fri, 6 Nov 2015 11:37:23 +0000 (12:37 +0100)]
Use Service object instead of DB API directly
In cmd.manage and volume.api we had code accessing DB API directly.
This should be done through Service versioned object instead and this
commit fixes that.
Victor Stinner [Thu, 5 Nov 2015 16:20:41 +0000 (17:20 +0100)]
Port test_tintri to Python 3
* Set configuration.nfs_oversub_ratio, otherwise tests
fail with TypeError on comparison between mock.Mock and int.
* tox.ini: add test_tintri to Python 3.4.
After introducing the commit 93d26e40098c966e7d791a9fb6152244a9f9409d
unit test TestCinderAllCmd fails if a test environment has
/etc/cinder/cinder.conf with proper backend configuration.
We should mock out some variables not to depend on test environment.
Tom Patzig [Mon, 26 Oct 2015 19:28:05 +0000 (20:28 +0100)]
Execute mount.nfs check with absolute path
Currently the existence of mount.nfs is checked by executing
the relative binary 'mount.nfs' with a non-root user,
in this case cinder. This results, for example on SUSE, in the error:
NfsException: mount.nfs is not installed
Because mount.nfs is located under /sbin, unprivileged users
do not have /sbin in their PATH to search for executables.
The change runs the mount.nfs check by using the absolute binary
path /sbin/mount.nfs. This seems to be common for most distributions
(SUSE, RedHat, CentOS, Ubuntu, Debian). The check can still be executed
as non privileged user, by not relying on correctly set PATH variable
and using the absolute path.
Sonia Ghanekar [Mon, 2 Nov 2015 21:58:42 +0000 (13:58 -0800)]
Using extra-specs in cloned vols for Nimble driver
This patch uses the set extra-specs while cloning a volume. All
existing extra-specs: 'nimble:encryption', 'nimble:perfpol-name'
and 'nimble:multi-initiator' are being reflected in the cloned
volume.
In the existing code, the extra-specs were not being considered
while cloning volumes; the parent volume attributes were used
directly in the cloned volume. This patch uses the extra-spec
values if they are set or if not then uses the default values
for the same.
This patch improves get_volume_stats method to invoke
dog command in SheepdogClient Class method instead of
in SheepdogDriver class method directly. Here are two
benefits we can realize as a result:
1.The current implementation can only get volume status
by local sheepdog node, but the SheepdogClient Class also
supports the method to run dog command with remote sheepdog
node.
2.SheepdogClient Class methods are implemented to run dog
command with fine grained Error handling. So it can improve
the robustness and is more readable to run dog command in
SheepdogClient Class method.
This patch adds the CORS support middleware to Cinder, allowing a
deployer to optionally configure rules under which a javascript client
may break the single-origin policy and access the API directly.
For cinder, the paste.ini method of deploying the middleware was
chosen, because it needs to be able to annotate responses created
by keystonemiddleware. If the middleware was explicitly included,
keystone would reject the request before the cross-domain headers
could be annotated, resulting in an error response that is
unreadable by the user agent.
Note that tests were explicitly not included. Cinder does not
possess a functional test suite, and thus constructing any
fake application with the CORS Middleware would not correctly
test the paste.ini initialization code path.
Fix calling delete_zones method with a wrong argument
Add a new zone with cisco zonemanager driver. If the added zone
name is the same as the existed name, the function delete_zones
will be called. But it fails because of passing a argument with
incorrect type. This patch fixes this error.
Co-Authored-By: Al Lau <alau2@cisco.com>
Closes-Bug: #1496283
Change-Id: Ib340b6828cfeb2bb318db23971268c7974ecc9f6
Nate Potter [Mon, 19 Oct 2015 16:24:29 +0000 (16:24 +0000)]
Return volume_type extra specs based on policy
Currently extra specs are only shown to the admin user
when showing volume types. This patch grants the flexibility
to grant other users privelages to see them based on policy.
It also hides the extra_specs entry in the dictionary from
users who can't see it rather than just showing "None" as
the value.
Ryan McNair [Tue, 27 Oct 2015 22:46:46 +0000 (22:46 +0000)]
Update register_opts hacking check to allow tuples
Update the hacking check for register_opts to allow tuples to be
passed in directly as an argument. Previously tuples were getting
flagged as invalid due to a missing pair of parenthesis in the
hacking check.
zhangsong [Sat, 24 Oct 2015 16:47:47 +0000 (00:47 +0800)]
optimize the copy_image_to_volume method of sheepdogdriver
the current copy_image_to_volume method of sheepdogdriver can only
write the image file to local sheepdog, with this patch, it supports
write image file to remote sheepdog node.
The rule "limits_extension:used_limits" is used by
"UsedLimitsController" which is a controller extension and it extends
"limits" api. Most of controller extensions authorize user's context
by applying the function "extensions.soft_extension_authorizer()",
but "UsedLimitsController" applies "extensions.extension_authorizer"
and this may cause 403 Forbidden error.
In this patch, I changed UsedLimitsController's authorizer to
"soft_extension_authorizer".
Fergal Mc Carthy [Tue, 27 Oct 2015 18:25:45 +0000 (14:25 -0400)]
lvconvert missing from cinder volume.filters
The LVM lvconvert command is invoked by the revert method of the
cinder.brick.local_dev.lvm.LVM class but was missing a matching
volume.filters CommandFilter entry which has now been added.
Wilson Liu [Tue, 13 Oct 2015 07:19:18 +0000 (15:19 +0800)]
Brocade driver add_zone optimization
Check whether the new zone we want to add/create
has the same zone name and members with the existing
zone on the switch. If they are same, no need to
delete and re-add it. This is significant if we
create volume from image or upload volume to image
in batch.
Michał Dulko [Fri, 23 Oct 2015 13:55:49 +0000 (15:55 +0200)]
Add missing cgsnapshot field to Snapshot object
SQLAlchemy model of snapshot specifies its relationship to cgsnapshot.
This wasn't reflected in Snapshot versioned object. The patch adds
cgsnapshot field to Snapshot object as well as required unit tests.
Michał Dulko [Fri, 23 Oct 2015 09:59:48 +0000 (11:59 +0200)]
Eager load snapshot_metadata in *snapshot_get_all
All methods returning snapshot lists in db.sqlalchemy.api are eager
loading snapshot_metadata - besides snapshot_get_all_by_project and
snapshot_get_active_by_window. In case of the latter that fact caused
unit tests to randomly fail because of SQLAlchemy Session sometimes
getting closed before the metadata got lazy loaded. This commit adds
missing options(joinedload('snapshot_metadata')) to these queries.
Yusuke Hayashi [Tue, 13 Oct 2015 23:53:49 +0000 (08:53 +0900)]
Case sensitivity problem in cinder scheduler
Volume creation fails when we use backend name which consists of
same characters as the name used before but different upper/lower case.
It happens in MySQL environment.
e.g.
1. configure 'lvm' as backend name and start cinder
2. reconfigure 'LVM' as backend name and restart cinder
It is caused by the difference of handling of case sensitivity
for backend name.
When checking the existence of cinder-volume/backend to judge
whether new cinder-volume service row in the DB should be created
or not, the search-logic is case insensitive for backend name.
On the other hand, when searching backend in create_volume processing,
the search-logic in cinder-scheduler is case sensitive for backend name.
That is why I change the former search-logic
from case insensitive one to case sensitive one.
Michał Dulko [Thu, 22 Oct 2015 19:49:13 +0000 (21:49 +0200)]
Add protocol to help of glance_api_servers option
Help for glance_api_servers option lacked information that it is
possible to specify protocol there (http or https) and suggested that
it isn't possible to use HTTPS in connection from Cinder to Glance.
This commit expands the message to include such information.
Lucian Petrut [Tue, 17 Feb 2015 16:28:07 +0000 (18:28 +0200)]
SMBFS: Fix retrieving total allocated size
Currently, the Windows SMBFS driver parses all the VHD/X images
present on the configured share in order to retrieve the allocated
size. Not only that this can be slow when having many volumes, but
it can also fail in certain environments when the images are in use.
The Linux SMBFS driver uses 'du', which gives incorrect values in
case of VHD/X images.
In order to avoid this, the driver can keep track of the allocated
size according to each share, the most basic solution being a JSON
stored in a file.
As the Windows SMBFS driver inherits the Linux SMBFS driver, only
overwriting the os specific methods, the fix is applied to the base
SMBFS driver.
Michał Dulko [Thu, 22 Oct 2015 12:54:35 +0000 (14:54 +0200)]
Make relationships in objects consistent
In Volume object 1:n relationship on volume_attachment was modeled
by ListOfObjectsField. Moreover _from_db_object and obj_load_attr
actually were setting VolumeAttachmentList as value of that field,
which is wrong behavior.
In CGSnapshot similar relationship on snapshots was done by ObjectField
with SnapshotList inside.
This commit unifies the approach to use the latter. Also unit test is
added to prevent mismatch of field type and value set on it.
Vincent Hou [Wed, 14 Oct 2015 02:35:13 +0000 (19:35 -0700)]
Remove the jointly loaded model in finish_volume_migration
After a successful volume migration, the source volume and the
destination volume need to swap the data in the models in
finish_volume_migration. It is sufficient to load the volume
model only and there is no need to load other models, like
volume type, metadata, consistency group, etc. If we load
these additional models, it will lead to NULL key error, when
either source or destination model has a NULL key pointer.
Sean McGinnis [Wed, 21 Oct 2015 21:23:56 +0000 (16:23 -0500)]
Update docs to generate Guru Meditation Report
Guru Meditation Reports (GMR) code in oslo.reports was recently updated
to expect SIGUSR2 as opposed to SIGUSR1, because it is reserved by
Apache 'mod_wsgi' for its own use.
Signal was changed and merged in change:
I9d3b6079ba2cca41fe4723723a6f80b2c3c0b9c0
Code Review / openstack-build / cinder-build.git / log