john-griffith [Mon, 10 Feb 2014 21:42:43 +0000 (14:42 -0700)]
Check type argument on create from source and snap
Currently the create from snap and create from source
options will set the volume-type when specified in the
create call.
This is actually wrong, because clone and create from snap
MUST be performed on the same backend as the parent volume. The
result is that if the specified type doesn't match the specs for
it's parent, and actually needs to be on a different backend
the new volume will "appear" to have the correct type, but it's
actually invalid.
The --volume-type argument should be rejected when --source-volid
or --snapshot-id are used, and the parent type-id just picked up
and used. Then the user can retype if desired.
Auto failing this strictly causes backward compatability issues though
so we'll just check for a mismatch and raise an exception in that case.
This way if the user knows what they're doing and they're explicit
in the command we still allow it and things are fine.
ling-yun [Wed, 12 Feb 2014 02:02:54 +0000 (10:02 +0800)]
Removed unused context in _extend_snapshot method
Input parameter 'context' of _extend_snapshot method in
cinder/api/extended_snapshot_attributes.py is not being used, so remove
this 'context' parameter.
Avishay Traeger [Tue, 11 Feb 2014 20:34:24 +0000 (22:34 +0200)]
Storwize/SVC: Check wwpn not None
If the storage doesn't have WWPN information for a host but gets it in
the connector, initialize_connection can fail because it does
wwpn.lower(), and wwpn can be None. Make sure it isn't before calling
lower().
Glenn M. Gobeli [Tue, 11 Feb 2014 18:03:13 +0000 (13:03 -0500)]
Changes to cast variable as integer as XML API fails.
Code change to ensure that the type of the size variable is correct
when cinder is invoked through XML API. The extend volume code path
differs from volume create where the flow portion verifies that the
size variable is an integer; for extend the earliest place to cast
the size variable is in volume_actions.py.
Ann Kamyshnikova [Tue, 27 Aug 2013 12:17:25 +0000 (16:17 +0400)]
Move create_, remove_ and ensure_export from drivers
create_export, remove_export, ensure_export are almost the same in
LVM and BlockDevice drivers. So they should be moved to
a common cinder.volume.iscsi module.
Jim Branen [Tue, 7 Jan 2014 00:14:35 +0000 (16:14 -0800)]
New HP LeftHand array iSCSI driver
This driver is intended to replace the current OpenStack Block
Storage HP LeftHand (LH) StoreVirtual iSCSI Driver,
(cinder.volume.drivers.san.HpSanISCSIDriver), by moving the
existing SSH interface into the new driver to maintain backwards
compatibility, and add the new LH REST interface for new driver
features.
We have the driver broken into 3 files:
hp_lefthand_iscis.py (common interface)
hp_lefthand_cliq_proxy.py (old SSH interface)
hp_lefthand_rest_proxy.py (new REST interface)
The reason we are doing this is because the SSH interface on LH
array has connections and performance limitations. These problems
will be resolved by moving to the new LH OS REST interface. Also,
new LeftHand array capabilities will only be supported in the
REST(hplefthandclient) interface.
To support new driver capabilities (create cloned volume), the
python REST client (hplefthandclient) is required and can be
downloaded from the pypi repository:
http://pypi.python.org/pypi/hplefthandclient.
This REST client requires LeftHand firmware version 11.5 or
greater.
Florent Flament [Fri, 17 Jan 2014 10:54:27 +0000 (10:54 +0000)]
Allow operators to customize max header size
HTTP messages max header line size has been increased from 8K to 16K
by default to allow using tokens including up to 14-15 catalog
entries. The same issue
(https://bugs.launchpad.net/swift/+bug/1190149) may occur again in
the future, if keystone's catalog grows further.
Allowing operators to customize the max header size, will allow them
to have Cinder working whatever the size of the catalog (if the option
is properly set).
It appears that there are certain cases where a detach
and a delete call can overlap causing a race in the
remove_iscsi_target method. The result is that when
the slower method gets around to the final removal of
the persistence file it's no longer there, so we
hit an unhandled exception.
This patch does two things:
1. It adds the volume lock to the detach_volume
method in the manager (already in place for delete).
This should mitigate the race and allow us to avoid it.
2. In the iscsi_remove_target method check if the
persist file has already been removed when we get
to the last check. The fact is that if it was somehow
removed before we got here we don't really care, we're
just going to delete it anyway.
Jeremy Stanley [Mon, 10 Feb 2014 02:57:24 +0000 (02:57 +0000)]
Remove tox locale overrides
* tox.ini: The LANG, LANGUAGE and LC_ALL environment overrides were
introduced originally during the testr migration in an attempt to be
conservative about the possibility that locale settings in the
calling environment could cause consistency problems for test runs.
In actuality, this should be unnecessary and any place where it does
cause issues ought to be considered an actual bug. Also, having
these in the configuration actively causes older pip to have
problems with non-ASCII content in some package metadata files under
Python 3, so drop it now.
Avishay Traeger [Sat, 8 Feb 2014 20:59:58 +0000 (22:59 +0200)]
Storwize/SVC: Fix races in host-related functions
In the Storwize/SVC driver, initialize_connection checks if a host
object exists on the storage, and if not, creates one. Similarly,
terminate_connection checks if there are any more mappings for this
host, and deletes the host object if there are none. Consequently, there
exist races between two initialize_connection calls, two
terminate_connection calls, or one and one.
The easy solution here is to use locks. Because locks are files on the
local machine, this implies that all cinder-volume processes managing a
given Storwize/SVC contoller run on the same machine.
john-griffith [Sat, 8 Feb 2014 04:55:22 +0000 (21:55 -0700)]
Dont set error_extending status on API extend call
Recent change (Ifd523ac5e9039861cf87711dc5c4842b5cb524c2)
moved the quota check on extend_volume to the API layer
in order to provide feedback to caller if the quota check
failed.
The problem is that change still sets the volume status to
error_extending if the quota check fails. There's no reason
to do this if we're now giving immediate feedback that the call
failed.
I'm also unclear on what the purpose of adding the reservations
to the rpc call are, but I'll double check on that before changing.
This patch leaves most of the previous change in place but
removes the setting of error_extending to the volumes status
on the quota fail.
Jeff Applewhite [Thu, 30 Jan 2014 19:11:08 +0000 (14:11 -0500)]
Fixes cinder failed to create/restore a backup with NFS driver
The action cinder backup-create fails when trying to backup a newly
created volume when using cinder.volume.drivers.nfs.NfsDriver. This
patch removes not implemented stubs for backup_volume and
restore_volume. The inherited methods in cinder/volume/driver.py
succeed in creating a backup and in restoring it. After manual
testing of this change backups/restores succeed without the not
implemented stubs. This change also adds the not implemented
stubs to the glusterfs driver until a fix is submitted.
Eric Harney [Wed, 29 Jan 2014 15:45:03 +0000 (10:45 -0500)]
Rename self.tgtadm to self.target_helper
Using "tgtadm" for this member variable is confusing because it
actually represents tgt, IET, and LIO. Rename to a more generic
term to make things more readable.
Also remove "tgtadm" from method names and references in tests
where it is not tgtadm-specific.
I'm choosing the name "target_helper" because "tgtadm" = tgtd,
and "iscsi_helper" doesn't fit when using RDMA in ISERDriver.
There was a race condition between the two Event() objects being
created within loopingcall and Retry in api module. Did away with
the Event object in Retry and used loopingcall the right way.
This commit brings the implementation of the test_rbd and
test_backup_ceph unit tests more inline and tries to ensure
that no mock is set without being unset so as not to tread
on the toes of any other tests.
John Griffith [Fri, 31 Jan 2014 03:53:37 +0000 (03:53 +0000)]
Remove create_export from volume create
Currently the LVM driver creates iscsi targets on
volume creation, even though it's only used at
attach time. This isn't necessary and in fact
causes issues if a volume is extended because the
target information isn't currently updated after the
extend. In addition a number of other drivers inclduing
the LIO iscsi driver require the target be created with
initiator info, so the tgt that's created initially
again isn't valid.
This change removes the create_export call from the
volume create process and makes it part of the
managers intialize_connection routine which is
more appropriate.
This change also removes the target on detach since
it's not used any longer and again as the volume may
be modified or extended.
Time unit is missing in the description of vmdk driver option
"vmware_task_poll_interval". Due to this, the vmdk driver config
reference document doesn't have the timeunit in its description
column. This fix adds the missing time unit (seconds) in the
driver option.
Avishay Traeger [Sun, 26 Jan 2014 10:39:27 +0000 (12:39 +0200)]
Allow spaces in host names in the storwize driver
Storwize/SVC naming rules allow host objects to have spaces in their
names. The SSH injection filter will currently reject such names, and so
we surround the names with quotes. Please note that this doesn't allow
to have an arbitrary character in there except a space (anything else
will be cleaned up by the driver or rejected by the SSH injection filter
up the call chain).
Bertrand Lallau [Sat, 1 Feb 2014 13:33:33 +0000 (14:33 +0100)]
Remove a catching exception during delete_volume
GlanceMetadataNotFound Exception was catching during
a call to volume_glance_metadata_delete_by_volume().
This exception is never thrown by the implementation.
This catching has been removed.
This patch migrates all of the communication
to the 3PAR array into the client library.
Some of the calls to the array happen over ssh
and others happen over REST. Now the drivers
don't care.
This allows us to change the external client
library to replace SSH calls to REST calls,
without the need of driver changes.
Ryan McNair [Mon, 27 Jan 2014 21:48:51 +0000 (21:48 +0000)]
Add support for special char in volume metadata
Using special characters such as "&" in volume metadata caused SAX
parser to split text into multiple child nodes, as stated in the
specs for SAX parsers. In this case extract_text would return
'None' because text was return if there was exactly
one child node. Concatenating the values of child nodes accounts
for SAX parsers splitting contigous characters into
multiple chunks.
Brianna Poulos [Wed, 29 Jan 2014 20:49:20 +0000 (15:49 -0500)]
Stop volume_type_encryption creation when in use
This bug fix addresses bug #1274252. It requires a volume
type to have no volumes with the type before allowing it
to be made encrypted. This prevents a situation where
non-encrypted volumes could have an encrypted volume type.
This fix also adds a unit test to confirm functionality.
john-griffith [Wed, 29 Jan 2014 04:30:59 +0000 (21:30 -0700)]
Revert initialize_connection changes
A change was merged that reusulted in interittent
gate failures. It seems that in certain cases the
intitialize is doing an update targets and is
deleting the existing targets and not recreating them.
The idea of this patch was to fix up the iscis targets
for cases where volumes are extended.
A better solution is in progress that will remove the
target creation from the create_volume process altogether.
Currently we seem to have create_export, ensure_export and
now initialize that all are designed to do very similar
things.
A bug has been filed to address this and will attempt
to collapse these functions to be done at attach time.
Florian Haas [Mon, 20 Jan 2014 21:42:11 +0000 (22:42 +0100)]
VolumeManager: initialize even if a volume can't be found
If a previously volume cannot be exported (for example, because
an iSCSI target's backing LV is no longer present), VolumeManager
would previously bail out, leaving the volume service uninitialized.
This left volumes dangling in limbo: their state would be reported
as available (clearly not a reflection of reality), and they could
not be deleted (because the volume service that would be responsible
for deletion was unavailable).
Instead, catch an exception raised by ensure_export() separately,
log it, and set the volume to the error state.
For the tgt backend, this will also remove the volume_path file.
Previously this would cause an ISCSITargetRemoveFailed exception
when the volume would be removed. Instead, simply log a warning
and return, the way other backends (example: RBD) already do.
Also, add an info message that reflects the actual path and
contents of the volume_path file.
Finally, fix up the tgtadm unit test so that it tests for
identical IQNs,
Eric Harney [Mon, 27 Jan 2014 23:46:26 +0000 (18:46 -0500)]
Add create_iscsi_target stub to TargetAdmin
This should be defined in TargetAdmin, as it is assumed to be
present in all subclasses by callers.
IetAdm and LioAdm may need additional work to behave as desired
in all circumstances (such as volume extend), but this should
at least stabilize things in the meantime.
Code Review / openstack-build / cinder-build.git / log