def get_instance_nw_info(self, context, instance, **kwargs):
"""Returns all network info related to an instance."""
if isinstance(instance, dict):
instance_uuid = instance['uuid']
else:
instance_uuid = instance.uuid
with lockutils.lock('refresh_cache-%s' % instance_uuid):
# NOTE(danms): Several places in the code look up instances without
# pulling system_metadata for performance, and call this function.
# If we get an instance without it, re-fetch so that the call
# to network_api (which requires it for instance_type) will
# succeed.
attrs = ['system_metadata']
use_slave = kwargs.get('use_slave', False)
# NOTE(Rui Chen): Refresh instance in order to avoid race
# condition between booting/attaching_interface and
# nova/neutron event reporting mechanism. See details:
# https://bugs.launchpad.net/nova/+bug/1407664
instance = objects.Instance.get_by_uuid(context,
instance_uuid,
expected_attrs=attrs,
use_slave=use_slave)
result = self._get_instance_nw_info(context, instance, **kwargs)
# NOTE(comstud): Don't update API cell with new info_cache every
# time we pull network info for an instance. The periodic healing
# of info_cache causes too many cells messages. Healing the API
# will happen separately.
update_instance_cache_with_nw_info(self,
context,
instance,
nw_info=result,
update_cells=False)
return result
def _spawn(self, node, context, instance, image_meta, injected_files,
admin_password, network_info=None, block_device_info=None):
try:
self._plug_vifs(instance, network_info, context=context)
self._attach_block_devices(instance, block_device_info)
self._start_firewall(instance, network_info)
# Caching images is both CPU and I/O expensive. When running many
# machines from a single nova-compute server, deploys of multiple
# machines can easily thrash the nova-compute server - unlike a
# virt hypervisor which is limited by CPU for VMs, baremetal only
# uses CPU and I/O when deploying. By only downloading one image
# at a time we serialise rather than thrashing, which leads to a
# lower average time-to-complete during overload situations, and
# a (relatively) insignificant delay for compute servers which
# have sufficient IOPS to handle multiple concurrent image
# conversions.
with lockutils.lock('nova-baremetal-cache-images', external=True):
self.driver.cache_images(
context, node, instance,
admin_password=admin_password,
image_meta=image_meta,
injected_files=injected_files,
network_info=network_info,
)
self.driver.activate_bootloader(context, node, instance,
network_info=network_info)
# NOTE(deva): ensure node is really off before we turn it on
# fixes bug https://code.launchpad.net/bugs/1178919
self.power_off(instance, node)
self.power_on(context, instance, network_info, block_device_info,
node)
_update_state(context, node, instance, baremetal_states.PREPARED)
self.driver.activate_node(context, node, instance)
_update_state(context, node, instance, baremetal_states.ACTIVE)
except Exception:
with excutils.save_and_reraise_exception():
LOG.error(_("Error deploying instance %(instance)s "
"on baremetal node %(node)s.") %
{'instance': instance['uuid'],
'node': node['uuid']})
# Do not set instance=None yet. This prevents another
# spawn() while we are cleaning up.
_update_state(context, node, instance, baremetal_states.ERROR)
self.driver.deactivate_node(context, node, instance)
self.power_off(instance, node)
self.driver.deactivate_bootloader(context, node, instance)
self.driver.destroy_images(context, node, instance)
self._detach_block_devices(instance, block_device_info)
self._stop_firewall(instance, network_info)
self._unplug_vifs(instance, network_info)
_update_state(context, node, None, baremetal_states.DELETED)
else:
# We no longer need the image since we successfully deployed.
self.driver.destroy_images(context, node, instance)
def _spawn(self, node, context, instance, image_meta, injected_files,
admin_password, network_info=None, block_device_info=None):
try:
self._plug_vifs(instance, network_info, context=context)
self._attach_block_devices(instance, block_device_info)
self._start_firewall(instance, network_info)
# Caching images is both CPU and I/O expensive. When running many
# machines from a single nova-compute server, deploys of multiple
# machines can easily thrash the nova-compute server - unlike a
# virt hypervisor which is limited by CPU for VMs, baremetal only
# uses CPU and I/O when deploying. By only downloading one image
# at a time we serialise rather than thrashing, which leads to a
# lower average time-to-complete during overload situations, and
# a (relatively) insignificant delay for compute servers which
# have sufficient IOPS to handle multiple concurrent image
# conversions.
with lockutils.lock('nova-baremetal-cache-images', external=True):
self.driver.cache_images(
context, node, instance,
admin_password=admin_password,
image_meta=image_meta,
injected_files=injected_files,
network_info=network_info,
)
self.driver.activate_bootloader(context, node, instance,
network_info=network_info)
# NOTE(deva): ensure node is really off before we turn it on
# fixes bug https://code.launchpad.net/bugs/1178919
self.power_off(instance, node)
self.power_on(context, instance, network_info, block_device_info,
node)
_update_state(context, node, instance, baremetal_states.PREPARED)
self.driver.activate_node(context, node, instance)
_update_state(context, node, instance, baremetal_states.ACTIVE)
except Exception:
with excutils.save_and_reraise_exception():
LOG.error(_("Error deploying instance %(instance)s "
"on baremetal node %(node)s.") %
{'instance': instance['uuid'],
'node': node['uuid']})
# Do not set instance=None yet. This prevents another
# spawn() while we are cleaning up.
_update_state(context, node, instance, baremetal_states.ERROR)
self.driver.deactivate_node(context, node, instance)
self.power_off(instance, node)
self.driver.deactivate_bootloader(context, node, instance)
self.driver.destroy_images(context, node, instance)
self._detach_block_devices(instance, block_device_info)
self._stop_firewall(instance, network_info)
self._unplug_vifs(instance, network_info)
_update_state(context, node, None, baremetal_states.DELETED)
else:
# We no longer need the image since we successfully deployed.
self.driver.destroy_images(context, node, instance)
def manage_main():
fname = ETC_LXC_USERNET
parser = argparse.ArgumentParser()
parser.add_argument("--type", "-t", help="nic type (default: 'veth')",
default="veth", dest="ntype")
parser.add_argument('operation', choices=("set", "inc", "dec", "del", "get"))
parser.add_argument('user', help="username")
parser.add_argument('bridge', help="bridge")
parser.add_argument('count', nargs="?", help="number to operate with.",
default=None, const=int)
args = parser.parse_args()
if args.operation == "del":
args.operation = "set"
args.count = 0
elif args.operation in ("set", "inc", "dec") and args.count is None:
args.count = 1
if args.operation == "get":
if args.bridge == "*":
args.bridge = None
matching = lfilter(fname, user=args.user, bridge=args.bridge,
count=args.count, ntype=args.ntype)
for l in matching:
print(str(l))
return 0
with lockutils.lock(str(fname)):
if not os.path.exists(fname):
fp.write("# USERNAME TYPE BRIDGE COUNT\n")
update_usernet(user=args.user, bridge=args.bridge, op=args.operation,
count=args.count, ntype=args.ntype, fname=fname)
def get_instance_nw_info(self,
context,
instance,
networks=None,
port_ids=None,
use_slave=False,
pci_list=None):
"""Return network information for specified instance
and update cache.
"""
# NOTE(): It would be nice if use_slave had us call
# special APIs that pummeled slaves instead of
# the master. For now we just ignore this arg.
with lockutils.lock('refresh_cache-%s' % instance['uuid']):
instance = objects.Instance.get_by_uuid(
context,
instance['uuid'],
expected_attrs=['system_metadata'],
use_slave=use_slave)
result = self._get_instance_nw_info(context, instance, networks,
port_ids, pci_list)
base_api.update_instance_cache_with_nw_info(self,
context,
instance,
nw_info=result,
update_cells=False)
return result
def _age_cached_images(self, context, datastore, dc_info, ds_path):
"""Ages cached images."""
age_seconds = CONF.remove_unused_original_minimum_age_seconds
unused_images = self.originals - self.used_images
ds_browser = self._get_ds_browser(datastore['ref'])
for image in unused_images:
path = self.timestamp_folder_get(ds_path, image)
# Lock to ensure that the spawn will not try and access a image
# that is currently being deleted on the datastore.
with lockutils.lock(path,
lock_file_prefix='nova-vmware-ts',
external=True):
ts = self._get_timestamp(ds_browser, path)
if not ts:
ts_path = '%s/%s' % (path, self._get_timestamp_filename())
try:
ds_util.mkdir(self._session, ts_path, dc_info.ref)
except error_util.FileAlreadyExistsException:
LOG.debug(_("Timestamp already exists."))
LOG.info(
_("Image %s is no longer used by this node. "
"Pending deletion!"), image)
else:
dt = self._get_datetime_from_filename(ts)
if timeutils.is_older_than(dt, age_seconds):
LOG.info(_("Image %s is no longer used. "
"Deleting!"), path)
# Image has aged - delete the image ID folder
self._folder_delete(path, dc_info.ref)
# If the image is used and the timestamp file exists then we delete
# the timestamp.
for image in self.used_images:
path = self.timestamp_folder_get(ds_path, image)
with lockutils.lock(path,
lock_file_prefix='nova-vmware-ts',
external=True):
self.timestamp_cleanup(dc_info.ref, ds_browser,
datastore['ref'], datastore['name'],
path)
def _age_cached_images(self, context, datastore, dc_info,
ds_path):
"""Ages cached images."""
age_seconds = CONF.remove_unused_original_minimum_age_seconds
unused_images = self.originals - self.used_images
ds_browser = self._get_ds_browser(datastore['ref'])
for image in unused_images:
path = self.timestamp_folder_get(ds_path, image)
# Lock to ensure that the spawn will not try and access a image
# that is currently being deleted on the datastore.
with lockutils.lock(path, lock_file_prefix='nova-vmware-ts',
external=True):
ts = self._get_timestamp(ds_browser, path)
if not ts:
ts_path = '%s/%s' % (path,
self._get_timestamp_filename())
try:
ds_util.mkdir(self._session, ts_path, dc_info.ref)
except error_util.FileAlreadyExistsException:
LOG.debug(_("Timestamp already exists."))
LOG.info(_("Image %s is no longer used by this node. "
"Pending deletion!"), image)
else:
dt = self._get_datetime_from_filename(ts)
if timeutils.is_older_than(dt, age_seconds):
LOG.info(_("Image %s is no longer used. "
"Deleting!"), path)
# Image has aged - delete the image ID folder
self._folder_delete(path, dc_info.ref)
# If the image is used and the timestamp file exists then we delete
# the timestamp.
for image in self.used_images:
path = self.timestamp_folder_get(ds_path, image)
with lockutils.lock(path, lock_file_prefix='nova-vmware-ts',
external=True):
self.timestamp_cleanup(dc_info.ref, ds_browser,
datastore['ref'], datastore['name'],
path)
def enlist_image(self, image_id, datastore, dc_ref):
ds_browser = self._get_ds_browser(datastore.ref)
cache_root_folder = datastore.build_path(self._base_folder)
# Check if the timestamp file exists - if so then delete it. This
# will ensure that the aging will not delete a cache image if it
# is going to be used now.
path = self.timestamp_folder_get(cache_root_folder, image_id)
# Lock to ensure that the spawn will not try and access a image
# that is currently being deleted on the datastore.
with lockutils.lock(str(path), lock_file_prefix='nova-vmware-ts',
external=True):
self.timestamp_cleanup(dc_ref, ds_browser, path)
def get_instance_nw_info(self, context, instance, networks=None, port_ids=None, use_slave=False, pci_list=None):
"""Return network information for specified instance
and update cache.
"""
# NOTE(): It would be nice if use_slave had us call
# special APIs that pummeled slaves instead of
# the master. For now we just ignore this arg.
with lockutils.lock("refresh_cache-%s" % instance["uuid"]):
instance = objects.Instance.get_by_uuid(
context, instance["uuid"], expected_attrs=["system_metadata"], use_slave=use_slave
)
result = self._get_instance_nw_info(context, instance, networks, port_ids, pci_list)
base_api.update_instance_cache_with_nw_info(self, context, instance, nw_info=result, update_cells=False)
return result
def enlist_image(self, image_id, datastore, dc_ref):
ds_browser = self._get_ds_browser(datastore.ref)
cache_root_folder = datastore.build_path(self._base_folder)
# Check if the timestamp file exists - if so then delete it. This
# will ensure that the aging will not delete a cache image if it
# is going to be used now.
path = self.timestamp_folder_get(cache_root_folder, image_id)
# Lock to ensure that the spawn will not try and access a image
# that is currently being deleted on the datastore.
with lockutils.lock(str(path), lock_file_prefix='nova-vmware-ts',
external=True):
self.timestamp_cleanup(dc_ref, ds_browser, path)
def wrapper(self, context, *args, **kwargs):
res = f(self, context, *args, **kwargs)
try:
# get the instance from arguments (or raise ValueError)
instance = kwargs.get('instance')
if not instance:
instance = args[argspec.args.index('instance') - 2]
except ValueError:
msg = _('instance is a required argument to use @refresh_cache')
raise Exception(msg)
with lockutils.lock('refresh_cache-%s' % instance['uuid']):
update_instance_cache_with_nw_info(self, context, instance,
nw_info=res)
# return the original function's return value
return res
def manage_main():
fname = ETC_LXC_USERNET
parser = argparse.ArgumentParser()
parser.add_argument("--type",
"-t",
help="nic type (default: 'veth')",
default="veth",
dest="ntype")
parser.add_argument('operation',
choices=("set", "inc", "dec", "del", "get"))
parser.add_argument('user', help="username")
parser.add_argument('bridge', help="bridge")
parser.add_argument('count',
nargs="?",
help="number to operate with.",
default=None,
const=int)
args = parser.parse_args()
if args.operation == "del":
args.operation = "set"
args.count = 0
elif args.operation in ("set", "inc", "dec") and args.count is None:
args.count = 1
if args.operation == "get":
if args.bridge == "*":
args.bridge = None
matching = lfilter(fname,
user=args.user,
bridge=args.bridge,
count=args.count,
ntype=args.ntype)
for l in matching:
print(str(l))
return 0
with lockutils.lock(str(fname)):
if not os.path.exists(fname):
fp.write("# USERNAME TYPE BRIDGE COUNT\n")
update_usernet(user=args.user,
bridge=args.bridge,
op=args.operation,
count=args.count,
ntype=args.ntype,
fname=fname)
def _open_cached_file(self, context, image_ref, image_meta, dst):
image_id = image_meta['id']
fpath = self._get_cached_file(image_id)
f = self._open(fpath)
if f:
return f
with lockutils.lock(image_id, external=True, lock_path=self.locks_dir):
f = self._open(fpath)
if f:
return f
tmp = tempfile.mktemp(dir=self.tmp_dir)
self._cache_image(context, image_ref, image_meta, tmp)
f = open(tmp)
os.rename(tmp, fpath)
return f
def _open_cached_file(self, context, image_ref, image_meta, dst):
image_id = image_meta['id']
fpath = self._get_cached_file(image_id)
f = self._open(fpath)
if f:
return f
with lockutils.lock(image_id, external=True, lock_path=self.locks_dir):
f = self._open(fpath)
if f:
return f
tmp = tempfile.mktemp(dir=self.tmp_dir)
self._cache_image(context, image_ref, image_meta, tmp)
f = open(tmp)
os.rename(tmp, fpath)
return f
def get_client(context, admin=False):
# NOTE(dprince): In the case where no auth_token is present
# we allow use of neutron admin tenant credentials if
# it is an admin context.
# This is to support some services (metadata API) where
# an admin context is used without an auth token.
if admin or (context.is_admin and not context.auth_token):
with lockutils.lock('neutron_admin_auth_token_lock'):
orig_token = AdminTokenStore.get().admin_auth_token
client = _get_client(orig_token, admin=True)
return ClientWrapper(client)
# We got a user token that we can use that as-is
if context.auth_token:
token = context.auth_token
return _get_client(token=token)
# We did not get a user token and we should not be using
# an admin token so log an error
raise exceptions.Unauthorized()
def get_client(context, admin=False):
# NOTE(dprince): In the case where no auth_token is present
# we allow use of neutron admin tenant credentials if
# it is an admin context.
# This is to support some services (metadata API) where
# an admin context is used without an auth token.
if admin or (context.is_admin and not context.auth_token):
with lockutils.lock('neutron_admin_auth_token_lock'):
orig_token = AdminTokenStore.get().admin_auth_token
client = _get_client(orig_token, admin=True)
return ClientWrapper(client)
# We got a user token that we can use that as-is
if context.auth_token:
token = context.auth_token
return _get_client(token=token)
# We did not get a user token and we should not be using
# an admin token so log an error
raise exceptions.Unauthorized()
def __init__(self, name, lock_file_prefix=None):
self.mgr = lockutils.lock(name, lock_file_prefix, True)
def _update_token(new_token):
with lockutils.lock('neutron_admin_auth_token_lock'):
token_store = AdminTokenStore.get()
token_store.admin_auth_token = new_token
def _update_token(new_token):
with lockutils.lock('neutron_admin_auth_token_lock'):
token_store = AdminTokenStore.get()
token_store.admin_auth_token = new_token