Mitsuhiro Tanino [Sat, 13 Jun 2015 04:23:40 +0000 (00:23 -0400)]
Wait until service thread is done on service stop
In order to be able to perform graceful shutdown of services, we need
to wait for all the current service threads to finish before exiting
from the service process.
Oslo.service provides this facility through it's wait() method, so we
need to call it in Cinder Service stop() method which gets called when
graceful shutdown is requested by sending the process SIGINT or SIGTERM.
This patch enables the reporting of the
multiattach capability in get_volume_stats.
The scheduler currently filters on the capability
for drivers that support/don't support that ability
to do more than one attachment for a volume.
By reporting multiattach = True, it lets the scheduler
know this driver allows attaching a volume more than once.
Enables native support for HGST Solutions software Spaces as
Cinder volumes or snapshots.
Each Cinder Volume or Snapshot is mapped to a single HGST Space.
This space may be named differently from the actual volume/snap ID
and so we store the Space Name<->ID mapping in the volume provider_id.
Snapshots are not supported with the current HGST Solutions
software, so they are implemented as heavyweight copies in the
driver.
All Spaces are made visible on the Cinder host for speed of access,
and only the spaces requested by Nova instances are actually made
visible on other members of the cluster.
Not all nodes need SSD storage to take advantage of these volumes,
cinder.conf entries specify which nodes share their local SSDs.
Prerequisites:
HGST Solutions is a software-SAN-like package which allows local
SSDs in a cluster to be combined into a single storage pool.
The driver has a series of configuration options which must be set
in the cinder.conf, prefixed with hgst_*.
Nodes should have the HGST software installed and connected to the
HGST domain prior to rolling out Nova nodes using this storage.
Additional patches required for full functionality (being tracked
under the same blueprint):
OS-brick patch @ https://review.openstack.org/#/c/186588/
Nova patch @ https://review.openstack.org/#/c/186594/
Nova patch required until os-brick<->nova connection finalized
in https://review.openstack.org/#/c/175569/
Daniel Wilson [Wed, 10 Jun 2015 21:32:49 +0000 (14:32 -0700)]
Volume manager should set filter_function and goodness_function
If a user defines a filter_function or goodness_function in their cinder.conf
file the scheduler should use that function. This change appends those
functions to volume_stats if the driver has not specified them.
Victor Stinner [Wed, 10 Jun 2015 12:55:52 +0000 (14:55 +0200)]
Replace it.next() with next(it) for py3 compat
The Python 2 next() method of iterators was renamed to __next__() on
Python 3. Use the builtin next() function instead which works on Python
2 and Python 3.
This patch was generated by the next operation of the sixer tool:
https://pypi.python.org/pypi/sixer
Manual changes:
* cinder/tests/unit/test_cmd.py: use "tpgs = iter([tpg])" instead of
using a magic mock for tpgs
This patch was generated by the six_moves operation of the sixer tool
version 0.4:
https://pypi.python.org/pypi/sixer
Manual changes:
* Disable pylint warnings on the HTTPSConnection classes because pylint
doesn't support importing from six.moves yet, see:
https://bitbucket.org/logilab/pylint/issue/550/
Alex Meade [Fri, 1 May 2015 16:56:48 +0000 (12:56 -0400)]
NetApp E-Series: Add Fibre Channel Support
The NetApp driver for E-series product lines currently support iSCSI. This
patch adds Fibre Channel support to the E-Series driver. This driver reuses
the same functionality as the E-Series iSCSI driver except for the
initialize_connection and terminate_connection driver methods which make
full use of Cinder's FibreChannel zone manager.
Alex Meade [Tue, 28 Apr 2015 01:19:36 +0000 (21:19 -0400)]
NetApp E-Series: Refactor class structure for FC
This patch moves classes to mimic the NetApp ONTAP drivers class
structure by creating a library module, making the iSCSI driver
a thin layer. This way the E-Series FC driver can be added in a
similar manner.
Alex Meade [Sun, 15 Feb 2015 22:01:54 +0000 (17:01 -0500)]
NetApp E-Series driver: Remove caching logic
This patch removes unnecessary API caching logic from the NetApp
E-Series driver. The cache, represented as the _objects dictionary,
is refreshed nearly every API call and therefore provides no value.
This refactor is also preparation for refactoring the class structure
so we can easily add an E-Series Fibre Channel driver.
John Griffith [Mon, 1 Jun 2015 20:04:15 +0000 (20:04 +0000)]
Add secondary account capability to SolidFire
The SolidFire Cluster has a limit of 2K volumes per SolidFire
account. Given that the SolidFire account maps directly to the
OpenStack tenant this can be a limitation for some large scale
deployments that may have tenants that wish to create more than
2K volumes in their OpenStack cloud.
This patch introduces the concept of a secondary account for
an OpenStack tenant on the SolidFire backend.
If an OpenStack tenant has more than 2K volumes, and wishes to
create additional volumes, we can create a secondary/supplemental
SolidFire account that is owned by the same OpenStack tenant.
If they wish to go beyond 4K volumes... well, SolidFire V3
driver will deal with that, for now 4K is a line in the sand.
While working on unit tests, we again move to using mock
opportunistically. In addition, a few tests that either
weren't actually testing anything, or were just a fake-loop
that didn't actually work have been removed:
NOTE:
This patch is a temporary for backporting purposes, we are
still planning a refresh in Liberty of the driver.
Victor Stinner [Tue, 9 Jun 2015 13:37:02 +0000 (15:37 +0200)]
Replace urllib and urllib2 with six.moves.urllib
The Python 2 urllib and urllib2 modules changed a lot in Python 3. On
Python 3, the urllib functions are now grouped into submodules of
urllib: urllib.parse, urllib.error, etc. The urllib2 has been removed.
Replace urllib and urllib2 with six.moves.urllib to make the Cinder code
compatible with Python 2 and Python 3.
The new code now looks like Python 3 code (ex: "urllib.parse.urlsplit"),
but it also works on Python 2.
Replace also "import six.moves.urllib.parse as urlparse" with "from
six.moves import urllib" to uniformize the code:
"urlparse.urlsplit(href)" becomes "urllib.parse.urlsplit(href)".
This patch was generated by the urllib operation of the sixer tool version 0.4:
https://pypi.python.org/pypi/sixer
Manual changes:
* Add a try/except ImportError to get the parse_http_list function
in cinder/api/urlmap.py
* Import the socket module instead of using urllib2.socket
in cinder/volume/drivers/huawei/rest_common.py
* test_nimble.py: replace NIMBLE_URLLIB2 value with
'six.moves.urllib.request'
* Remove "from six.moves.urllib import parse as urlparse", it was
replaced with urllib.parse from six.moves.
* Reformat to respect the 80 columns constraint
Victor Stinner [Wed, 10 Jun 2015 12:39:37 +0000 (14:39 +0200)]
Replace unicode with six.text_type
The "unicode" type was renamed to "str" in Python 3. Use six.text_type
to make Cinder compatible with Python 3.
The initial patch was generated by the unicode operation of sixer tool:
https://pypi.python.org/pypi/sixer
Manual changes:
* cinder/volume/drivers/san/hp/hp_3par_iscsi.py:
replace "isinstance(value, str) or isinstance(value, unicode)"
with "isinstance(value, six.string_types)"
* cinder/volume/drivers/san/hp/hp_3par_iscsi.py:
bump the version to 2.0.17
* Revert changes on hacking: hacking explicitly searchs the "unicode"
pattern for Python 2. Hacking may require more work to be ported to
Python 3.
* Revert changes on docstrings and strings
Jay S. Bryant [Tue, 9 Jun 2015 22:10:18 +0000 (17:10 -0500)]
Dispose DB connections between backend proc starts
Currently we do not dispose the existing DB pools before
forking off additional storage backend processes. This can
result in timing issues and DB errors.
This change simply ensures that we do a dispose_engine() in
the loop where we are forking off processes for the storage
backends.
Tom Swanson [Tue, 2 Jun 2015 19:02:33 +0000 (14:02 -0500)]
Dell SC: Expanded comments and update var names
Additional comments have been added to help explain what is happening
in the driver. This revolves mostly around dell_storagecenter_api.py
which is not an obvious file for a reviewer to review. Hopefully
additional comments will aid this.
Updated single character variable names except for r which is the
REST calls return throughout the code by convention.
Reworked create_volume for clarity. No functional changes.
Changed two obvious error messages from LOG.debug to LOG.error.
Jeremy Stanley [Tue, 9 Jun 2015 17:37:43 +0000 (17:37 +0000)]
Merge tag '2015.1.0'
This is a null-merge of the 2015.1.0 release tag back into the master
branch so that the 2015.1.0 tag will appear in the git commit history of
the master branch. It contains no actual changes to the master branch,
regardless of how our code review system's UI represents it. Please
ask in #openstack-infra if you have any questions, and otherwise try
to merge this as quickly as possible to avoid later conflicts on the
master branch.
Jeremy Stanley [Tue, 9 Jun 2015 16:31:41 +0000 (16:31 +0000)]
Merge tag '2014.2'
This is a null-merge of the 2014.2 release tag back into the
master branch. It contains no actual changes to the master branch,
regardless of how our code review system's UI represents it. Please
ask in #openstack-infra if you have any questions, and otherwise try
to merge this as quickly as possible to avoid later conflicts on the
master branch.
Declare the existing StorPool volume driver functionality by
deriving from the appropriate driver.*VD classes.
6. Implement volume retyping and migrations
initialize_connection(): store the StorPool client ID in the connection
data to be used by the "StorPool" Brick connector.
validate_connector(): actually check the connector data for a valid
hostname defined in the StorPool configuration.
_storpool_client_id(): new method to get the numeric ID of a host in
the StorPool cluster from the /etc/storpool.conf file.
_attach_volume(), _detach_volume(): if the "remote" flag is set, pass
the request on to the default Cinder driver implementation; it knows
how to make remote volumes accessible.
_attach_volume(), _detach_volume(): support the case when these methods
are invoked from a method that is not overridden in the StorPool Cinder
driver - create a storpool.spopenstack.AttachDB attachment structure
and destroy it upon detaching.
ensure_export(): add a stub method, the actual work is handled by
the Nova volume attachment driver.
retype(): new method: handle a StorPool-to-StorPool retyping where
at most the volume's StorPool template or replication is changed.
For anything else, return False to indicate an unsupported request so
that a full volume migration is triggered.
update_migrated_volume(): rename the StorPool volume to correspond to
the real Cinder volume's name.
Add the StorPoolConfigurationMissing and StorPoolConfigurationInvalid
exceptions.
Flesh out the validate_connector() and initialize_connection() unit tests.
Edwin Wang [Sat, 13 Dec 2014 20:42:09 +0000 (04:42 +0800)]
Add iSCSI protocol support for IBM FlashSystem
This change implements iSCSI protocol for Cinder driver.
It is inherited from driver.ISCSIDriver and flashsystem.py.
We will separate flashsystem.py into common and FC protocol.
iSCSI test result:
https://bugs.launchpad.net/cinder/+bug/1406442
Kurt Martin [Mon, 8 Jun 2015 17:26:20 +0000 (10:26 -0700)]
Fixes 3PAR snapshot failure with optional params
Create snapshot fails when HP3PAR hp3par_snapshot_retention and/or
hp3par_snapshot_expiration parameters were set in cinder.conf. The
params were being sent to the backend as strings instead on integers.
peter_wang [Fri, 29 May 2015 00:38:56 +0000 (20:38 -0400)]
Add 'source-id' and 'source-name' support in VNX driver
Currently, VNX driver only supports 'id' as the id-type in manage/unmanage volume.
With release of new version python-cinderclient, 'source-id' and 'source-name' are
preferred id-type when using cinder manage.
This fix is to add 'source-id' and 'source-name' support
and obsolete the self-defined 'id' support.
Tom Swanson [Wed, 13 May 2015 17:00:54 +0000 (12:00 -0500)]
Dell SC: Added support for alternate iscsi portals
Added support for returning alternate target portals,
luns and iqns in a non multipath setup. This was done
per https://review.openstack.org/#/c/140877/ Tests
were modified to support this change.
The iscsi initialize_connection and find_iscsi_properties
functions took the brunt of this. The full list of portals,
luns and iqns are returned along with the preferred portal,
lun and iqn rather than a single iqn, lun and portal.
Some minor log message changes for clarity. (Removed the
one character vars.)
Tom Swanson [Tue, 12 May 2015 17:37:43 +0000 (12:37 -0500)]
Dell: Added verify cert option for REST calls
If one goes through the trouble to create a cert and
install it on the Dell Data Collector and on the cinder
node then we can enable certificate verification on the
request calls to the Dell REST API.
This simply makes dell_sc_verify_cert a cinder.conf option
for a Dell backend and directly uses that as the verify
option on the REST requests. Default is False.
The update_migrated_volume function arrived via a rebase
and had a merge issue that was fixed. Specifically the
dell_sc_ssn value is no longer part of the api.find_sc
call.
Thang Pham [Thu, 12 Mar 2015 17:18:05 +0000 (13:18 -0400)]
Complete switch to snapshot objects
The following patch switches the remainder of cinder internals
to use snapshot object instead of direct db calls. Note that
db calls within cinder drivers were not switched over. This
is left to driver maintainers to do themselves and to
properly test out the changes on their hardware.
Also, note that there are three occurrences of
db.snapshot_update and one of db.snapshot_destroy left in
cinder/volume/manager.py. This is intentional because
driver.create_cgsnapshot and driver.delete_cgsnapshot returns
a list of snapshot dicts. Each driver needs to switched over
occurences of db.snapshot_get_all_for_cgsnapshot() to
SnapshotList.get_all_for_cgsnapshot() in create_cgsnapshot
and delete_cgsnapshot. Once each driver has done so, a follow
up patch can be created to remove db.snapshot_update and
db.snapshot_destroy in cinder/volume/manager.py. There are
bugs filed for these to be fixed -
https://bugs.launchpad.net/cinder/+bugs?field.tag=
cgsnapshot-objects.
Zhiteng Huang [Thu, 4 Jun 2015 18:23:58 +0000 (11:23 -0700)]
DriverFilter: don't check volume_backend_name
Currently DriverFilter checks if host_state.volume_backend_name
matches type.extra_specs.volume_backend_name. The same check has
already been done in CapabilitiesFilter and done more thoroughly and
gracefully, for example, volume_backend_name can be a set of names
not limited to one to one match. DriverFilter should just do one
thing, that is evaluating filter_function supplied by backend.
This change removes the volume_backend_name check from DriverFilter
without touching its core function, thus has no impact on existing
documentation because the checking of volume_backend_name is an
undocumented side-effect of this filter. This change also reduces
one warning log to debug log when no filter_function found in
host_state, for the reason it would generate a lot of warning log
for those backends don't supply filter_function.
UpgradeImpact:
DriverFilter had an undocumented side-effect before it evaluates
filter_function - it checked if volume_backend_name in host_state
reported by backend matches volume_backend_name in volume type
extra specs, and if they don't, the backend would fail to pass
the filter. Now that this side-effect has been removed, it
would impact users only under following circumstance: 1) user
enabled DriverFilter but not CapabilitiesFilter; 2) user relied
on DriverFilter to filter backends based on exact match of
'volume_backend_name' in type extra spec. If unfortunately this
is the case for user, enabling CapabilitiesFilter by adding
it to 'scheduler_default_filters' configure option and restart
cinder scheduler service then everything should work just fine.
CapabilitiesFilter actually does a much better job checking
volume_backend_name.
Daniel Wilson [Tue, 12 May 2015 17:47:35 +0000 (10:47 -0700)]
Add Pure Storage FibreChannel driver
This commit adds a FibreChannel driver for Pure Storage. The driver shares
the majority of functionality with base driver and only has the code
necessary to support connection to FibreChannel.
Tom Barron [Thu, 19 Mar 2015 23:10:54 +0000 (19:10 -0400)]
Fix remaining memory issues with nfs backup unit tests
Here we fix another place in the nfs backup unit tests where
the configured file size was larger than the chunk size in
the corresponding Swift backup tests. I missed this one in
https://review.openstack.org/#/c/165394
and even though code with that fix ran fine in a bare 1GB VM,
there are still memory issues when run in CI with a VM of that
size.
Michal Dulko [Thu, 4 Jun 2015 15:25:24 +0000 (17:25 +0200)]
Don't send heartbeats if Manager reports a problem
This commit adds is_working method to base Manager class that can be
used to indicate that service has some problems and isn't running
correctly. This is used to block refreshing Service heartbeats if
Manager will return False from is_working.
The method is implemented in c-vol Manager to return initialization
state of the underlying volume driver.
Code Review / openstack-build / cinder-build.git / log