aodh/ceilometer/api/controllers/v2.py
ChenZheng 7f542e3ac8 Add i18n warpping for all LOG messages
Add i18n wrapping for all LOG messages

Change-Id: I7f9c71a3aa76364b291f7e21a1737b927cbdc300
Fixes: bug #1199678
2013-11-29 21:09:53 +08:00

1515 lines
54 KiB
Python

# -*- encoding: utf-8 -*-
#
# Copyright © 2012 New Dream Network, LLC (DreamHost)
# Copyright 2013 IBM Corp.
# Copyright © 2013 eNovance <licensing@enovance.com>
#
# Authors: Doug Hellmann <doug.hellmann@dreamhost.com>
# Angus Salkeld <asalkeld@redhat.com>
# Eoghan Glynn <eglynn@redhat.com>
# Julien Danjou <julien@danjou.info>
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
"""Version 2 of the API.
"""
import ast
import base64
import copy
import datetime
import inspect
import json
import uuid
from oslo.config import cfg
import pecan
from pecan import rest
import six
import wsme
from wsme import types as wtypes
import wsmeext.pecan as wsme_pecan
from ceilometer.api import acl
from ceilometer.openstack.common import context
from ceilometer.openstack.common.gettextutils import _ # noqa
from ceilometer.openstack.common import log
from ceilometer.openstack.common.notifier import api as notify
from ceilometer.openstack.common import strutils
from ceilometer.openstack.common import timeutils
from ceilometer import sample
from ceilometer import storage
from ceilometer import utils
LOG = log.getLogger(__name__)
ALARM_API_OPTS = [
cfg.BoolOpt('record_history',
default=True,
help='Record alarm change events'
),
]
cfg.CONF.register_opts(ALARM_API_OPTS, group='alarm')
state_kind = ["ok", "alarm", "insufficient data"]
state_kind_enum = wtypes.Enum(str, *state_kind)
operation_kind = wtypes.Enum(str, 'lt', 'le', 'eq', 'ne', 'ge', 'gt')
class ClientSideError(wsme.exc.ClientSideError):
def __init__(self, error, status_code=400):
pecan.response.translatable_error = error
super(ClientSideError, self).__init__(error, status_code)
class EntityNotFound(ClientSideError):
def __init__(self, entity, id):
super(EntityNotFound, self).__init__(
_("%(entity)s %(id)s Not Found") % {'entity': entity,
'id': id},
status_code=404)
class BoundedInt(wtypes.UserType):
basetype = int
def __init__(self, min=None, max=None):
self.min = min
self.max = max
@property
def name(self):
if self.min is not None and self.max is not None:
return 'int between %d and %d' % (self.min, self.max)
elif self.min is not None:
return 'int greater than %d' % self.min
else:
return 'int lower than %d' % self.max
@staticmethod
def frombasetype(value):
return int(value) if value is not None else None
def validate(self, value):
if self.min is not None and value < self.min:
error = _('Value %(value)s is invalid (should be greater or equal '
'to %(min)s)') % dict(value=value, min=self.min)
raise ClientSideError(error)
if self.max is not None and value > self.max:
error = _('Value %(value)s is invalid (should be lower or equal '
'to %(max)s)') % dict(value=value, max=self.max)
raise ClientSideError(error)
return value
class AdvEnum(wtypes.wsproperty):
"""Handle default and mandatory for wtypes.Enum
"""
def __init__(self, name, *args, **kwargs):
self._name = '_advenum_%s' % name
self._default = kwargs.pop('default', None)
mandatory = kwargs.pop('mandatory', False)
enum = wtypes.Enum(*args, **kwargs)
super(AdvEnum, self).__init__(datatype=enum, fget=self._get,
fset=self._set, mandatory=mandatory)
def _get(self, parent):
if hasattr(parent, self._name):
value = getattr(parent, self._name)
return value or self._default
return self._default
def _set(self, parent, value):
if self.datatype.validate(value):
setattr(parent, self._name, value)
class _Base(wtypes.Base):
@classmethod
def from_db_model(cls, m):
return cls(**(m.as_dict()))
@classmethod
def from_db_and_links(cls, m, links):
return cls(links=links, **(m.as_dict()))
def as_dict(self, db_model):
valid_keys = inspect.getargspec(db_model.__init__)[0]
if 'self' in valid_keys:
valid_keys.remove('self')
return self.as_dict_from_keys(valid_keys)
def as_dict_from_keys(self, keys):
return dict((k, getattr(self, k))
for k in keys
if hasattr(self, k) and
getattr(self, k) != wsme.Unset)
class Link(_Base):
"""A link representation
"""
href = wtypes.text
"The url of a link"
rel = wtypes.text
"The name of a link"
@classmethod
def sample(cls):
return cls(href=('http://localhost:8777/v2/meters/volume?'
'q.field=resource_id&'
'q.value=bd9431c1-8d69-4ad3-803a-8d4a6b89fd36'),
rel='volume'
)
class Query(_Base):
"""Sample query filter.
"""
_op = None # provide a default
def get_op(self):
return self._op or 'eq'
def set_op(self, value):
self._op = value
field = wtypes.text
"The name of the field to test"
#op = wsme.wsattr(operation_kind, default='eq')
# this ^ doesn't seem to work.
op = wsme.wsproperty(operation_kind, get_op, set_op)
"The comparison operator. Defaults to 'eq'."
value = wtypes.text
"The value to compare against the stored data"
type = wtypes.text
"The data type of value to compare against the stored data"
def __repr__(self):
# for logging calls
return '<Query %r %s %r %s>' % (self.field,
self.op,
self.value,
self.type)
@classmethod
def sample(cls):
return cls(field='resource_id',
op='eq',
value='bd9431c1-8d69-4ad3-803a-8d4a6b89fd36',
type='string'
)
def as_dict(self):
return self.as_dict_from_keys(['field', 'op', 'type', 'value'])
def _get_value_as_type(self, forced_type=None):
"""Convert metadata value to the specified data type.
This method is called during metadata query to help convert the
querying metadata to the data type specified by user. If there is no
data type given, the metadata will be parsed by ast.literal_eval to
try to do a smart converting.
NOTE (flwang) Using "_" as prefix to avoid an InvocationError raised
from wsmeext/sphinxext.py. It's OK to call it outside the Query class.
Because the "public" side of that class is actually the outside of the
API, and the "private" side is the API implementation. The method is
only used in the API implementation, so it's OK.
:returns: metadata value converted with the specified data type.
"""
type = forced_type or self.type
try:
converted_value = self.value
if not type:
try:
converted_value = ast.literal_eval(self.value)
except (ValueError, SyntaxError):
msg = _('Failed to convert the metadata value %s'
' automatically') % (self.value)
LOG.debug(msg)
else:
if type == 'integer':
converted_value = int(self.value)
elif type == 'float':
converted_value = float(self.value)
elif type == 'boolean':
converted_value = strutils.bool_from_string(self.value)
elif type == 'string':
converted_value = self.value
else:
# For now, this method only support integer, float,
# boolean and and string as the metadata type. A TypeError
# will be raised for any other type.
raise TypeError()
except ValueError:
msg = _('Failed to convert the metadata value %(value)s'
' to the expected data type %(type)s.') % \
{'value': self.value, 'type': type}
raise ClientSideError(msg)
except TypeError:
msg = _('The data type %s is not supported. The supported'
' data type list is: integer, float, boolean and'
' string.') % (type)
raise ClientSideError(msg)
except Exception:
msg = _('Unexpected exception converting %(value)s to'
' the expected data type %(type)s.') % \
{'value': self.value, 'type': type}
raise ClientSideError(msg)
return converted_value
class ProjectNotAuthorized(ClientSideError):
def __init__(self, id):
super(ProjectNotAuthorized, self).__init__(
_("Not Authorized to access project %s") % id,
status_code=401)
def _get_auth_project(on_behalf_of=None):
# when an alarm is created by an admin on behalf of another tenant
# we must ensure for:
# - threshold alarm, that an implicit query constraint on project_id is
# added so that admin-level visibility on statistics is not leaked
# - combination alarm, that alarm ids verification is scoped to
# alarms owned by the alarm project.
# hence for null auth_project (indicating admin-ness) we check if
# the creating tenant differs from the tenant on whose behalf the
# alarm is being created
auth_project = acl.get_limited_to_project(pecan.request.headers)
created_by = pecan.request.headers.get('X-Project-Id')
is_admin = auth_project is None
if is_admin and on_behalf_of != created_by:
auth_project = on_behalf_of
return auth_project
def _sanitize_query(query, db_func, on_behalf_of=None):
'''Check the query to see if:
1) the request is coming from admin - then allow full visibility
2) non-admin - make sure that the query includes the requester's
project.
'''
q = copy.copy(query)
auth_project = _get_auth_project(on_behalf_of)
if auth_project:
_verify_query_segregation(q, auth_project)
proj_q = [i for i in q if i.field == 'project_id']
valid_keys = inspect.getargspec(db_func)[0]
if not proj_q and 'on_behalf_of' not in valid_keys:
# The user is restricted, but they didn't specify a project
# so add it for them.
q.append(Query(field='project_id',
op='eq',
value=auth_project))
return q
def _verify_query_segregation(query, auth_project=None):
'''Ensure non-admin queries are not constrained to another project.'''
auth_project = (auth_project or
acl.get_limited_to_project(pecan.request.headers))
if auth_project:
for q in query:
if q.field == 'project_id' and (auth_project != q.value or
q.op != 'eq'):
raise ProjectNotAuthorized(q.value)
def _validate_query(query, db_func, internal_keys=[]):
_verify_query_segregation(query)
valid_keys = inspect.getargspec(db_func)[0]
internal_keys.append('self')
valid_keys = set(valid_keys) - set(internal_keys)
translation = {'user_id': 'user',
'project_id': 'project',
'resource_id': 'resource'}
has_timestamp = False
for i in query:
if i.field == 'timestamp':
has_timestamp = True
if i.op not in ('lt', 'le', 'gt', 'ge'):
raise wsme.exc.InvalidInput('op', i.op,
'unimplemented operator for %s' %
i.field)
else:
if i.op == 'eq':
if i.field == 'search_offset':
has_timestamp = True
elif i.field == 'enabled':
i._get_value_as_type('boolean')
elif i.field.startswith('metadata.'):
i._get_value_as_type()
elif i.field.startswith('resource_metadata.'):
i._get_value_as_type()
else:
key = translation.get(i.field, i.field)
if key not in valid_keys:
msg = ("unrecognized field in query: %s, "
"valid keys: %s") % (query, valid_keys)
raise wsme.exc.UnknownArgument(key, msg)
else:
raise wsme.exc.InvalidInput('op', i.op,
'unimplemented operator for %s' %
i.field)
if has_timestamp and not ('start' in valid_keys or
'start_timestamp' in valid_keys):
raise wsme.exc.UnknownArgument('timestamp',
"not valid for this resource")
def _query_to_kwargs(query, db_func, internal_keys=[]):
_validate_query(query, db_func, internal_keys=internal_keys)
query = _sanitize_query(query, db_func)
internal_keys.append('self')
valid_keys = set(inspect.getargspec(db_func)[0]) - set(internal_keys)
translation = {'user_id': 'user',
'project_id': 'project',
'resource_id': 'resource'}
stamp = {}
metaquery = {}
kwargs = {}
for i in query:
if i.field == 'timestamp':
if i.op in ('lt', 'le'):
stamp['end_timestamp'] = i.value
stamp['end_timestamp_op'] = i.op
elif i.op in ('gt', 'ge'):
stamp['start_timestamp'] = i.value
stamp['start_timestamp_op'] = i.op
else:
if i.op == 'eq':
if i.field == 'search_offset':
stamp['search_offset'] = i.value
elif i.field == 'enabled':
kwargs[i.field] = i._get_value_as_type('boolean')
elif i.field.startswith('metadata.'):
metaquery[i.field] = i._get_value_as_type()
elif i.field.startswith('resource_metadata.'):
metaquery[i.field[9:]] = i._get_value_as_type()
else:
key = translation.get(i.field, i.field)
kwargs[key] = i.value
if metaquery and 'metaquery' in valid_keys:
kwargs['metaquery'] = metaquery
if stamp:
q_ts = _get_query_timestamps(stamp)
if 'start' in valid_keys:
kwargs['start'] = q_ts['query_start']
kwargs['end'] = q_ts['query_end']
elif 'start_timestamp' in valid_keys:
kwargs['start_timestamp'] = q_ts['query_start']
kwargs['end_timestamp'] = q_ts['query_end']
if 'start_timestamp_op' in stamp:
kwargs['start_timestamp_op'] = stamp['start_timestamp_op']
if 'end_timestamp_op' in stamp:
kwargs['end_timestamp_op'] = stamp['end_timestamp_op']
return kwargs
def _validate_groupby_fields(groupby_fields):
"""Checks that the list of groupby fields from request is valid and
if all fields are valid, returns fields with duplicates removed
"""
# NOTE(terriyu): Currently, metadata fields are not supported in our
# group by statistics implementation
valid_fields = set(['user_id', 'resource_id', 'project_id', 'source'])
invalid_fields = set(groupby_fields) - valid_fields
if invalid_fields:
raise wsme.exc.UnknownArgument(invalid_fields,
"Invalid groupby fields")
# Remove duplicate fields
# NOTE(terriyu): This assumes that we don't care about the order of the
# group by fields.
return list(set(groupby_fields))
def _get_query_timestamps(args={}):
"""Return any optional timestamp information in the request.
Determine the desired range, if any, from the GET arguments. Set
up the query range using the specified offset.
[query_start ... start_timestamp ... end_timestamp ... query_end]
Returns a dictionary containing:
query_start: First timestamp to use for query
start_timestamp: start_timestamp parameter from request
query_end: Final timestamp to use for query
end_timestamp: end_timestamp parameter from request
search_offset: search_offset parameter from request
"""
search_offset = int(args.get('search_offset', 0))
start_timestamp = args.get('start_timestamp')
if start_timestamp:
start_timestamp = timeutils.parse_isotime(start_timestamp)
start_timestamp = start_timestamp.replace(tzinfo=None)
query_start = (start_timestamp -
datetime.timedelta(minutes=search_offset))
else:
query_start = None
end_timestamp = args.get('end_timestamp')
if end_timestamp:
end_timestamp = timeutils.parse_isotime(end_timestamp)
end_timestamp = end_timestamp.replace(tzinfo=None)
query_end = end_timestamp + datetime.timedelta(minutes=search_offset)
else:
query_end = None
return {'query_start': query_start,
'query_end': query_end,
'start_timestamp': start_timestamp,
'end_timestamp': end_timestamp,
'search_offset': search_offset,
}
def _flatten_metadata(metadata):
"""Return flattened resource metadata without nested structures
and with all values converted to unicode strings.
"""
if metadata:
return dict((k, unicode(v))
for k, v in utils.recursive_keypairs(metadata,
separator='.')
if type(v) not in set([list, set]))
return {}
def _make_link(rel_name, url, type, type_arg, query=None):
query_str = ''
if query:
query_str = '?q.field=%s&q.value=%s' % (query['field'],
query['value'])
return Link(href=('%s/v2/%s/%s%s') % (url, type, type_arg, query_str),
rel=rel_name)
def _send_notification(event, payload):
notification = event.replace(" ", "_")
notification = "alarm.%s" % notification
notify.notify(None, notify.publisher_id("ceilometer.api"),
notification, notify.INFO, payload)
class Sample(_Base):
"""A single measurement for a given meter and resource.
"""
source = wtypes.text
"The ID of the source that identifies where the sample comes from"
counter_name = wsme.wsattr(wtypes.text, mandatory=True)
"The name of the meter"
# FIXME(dhellmann): Make this meter_name?
counter_type = wsme.wsattr(wtypes.text, mandatory=True)
"The type of the meter (see :ref:`measurements`)"
# FIXME(dhellmann): Make this meter_type?
counter_unit = wsme.wsattr(wtypes.text, mandatory=True)
"The unit of measure for the value in counter_volume"
# FIXME(dhellmann): Make this meter_unit?
counter_volume = wsme.wsattr(float, mandatory=True)
"The actual measured value"
user_id = wtypes.text
"The ID of the user who last triggered an update to the resource"
project_id = wtypes.text
"The ID of the project or tenant that owns the resource"
resource_id = wsme.wsattr(wtypes.text, mandatory=True)
"The ID of the :class:`Resource` for which the measurements are taken"
timestamp = datetime.datetime
"UTC date and time when the measurement was made"
resource_metadata = {wtypes.text: wtypes.text}
"Arbitrary metadata associated with the resource"
message_id = wtypes.text
"A unique identifier for the sample"
def __init__(self, counter_volume=None, resource_metadata={},
timestamp=None, **kwds):
if counter_volume is not None:
counter_volume = float(counter_volume)
resource_metadata = _flatten_metadata(resource_metadata)
# this is to make it easier for clients to pass a timestamp in
if timestamp and isinstance(timestamp, basestring):
timestamp = timeutils.parse_isotime(timestamp)
super(Sample, self).__init__(counter_volume=counter_volume,
resource_metadata=resource_metadata,
timestamp=timestamp, **kwds)
if self.resource_metadata in (wtypes.Unset, None):
self.resource_metadata = {}
@classmethod
def sample(cls):
return cls(source='openstack',
counter_name='instance',
counter_type='gauge',
counter_unit='instance',
counter_volume=1,
resource_id='bd9431c1-8d69-4ad3-803a-8d4a6b89fd36',
project_id='35b17138-b364-4e6a-a131-8f3099c5be68',
user_id='efd87807-12d2-4b38-9c70-5f5c2ac427ff',
timestamp=datetime.datetime.utcnow(),
resource_metadata={'name1': 'value1',
'name2': 'value2'},
message_id='5460acce-4fd6-480d-ab18-9735ec7b1996',
)
class Statistics(_Base):
"""Computed statistics for a query.
"""
groupby = {wtypes.text: wtypes.text}
"Dictionary of field names for group, if groupby statistics are requested"
unit = wtypes.text
"The unit type of the data set"
min = float
"The minimum volume seen in the data"
max = float
"The maximum volume seen in the data"
avg = float
"The average of all of the volume values seen in the data"
sum = float
"The total of all of the volume values seen in the data"
count = int
"The number of samples seen"
duration = float
"The difference, in seconds, between the oldest and newest timestamp"
duration_start = datetime.datetime
"UTC date and time of the earliest timestamp, or the query start time"
duration_end = datetime.datetime
"UTC date and time of the oldest timestamp, or the query end time"
period = int
"The difference, in seconds, between the period start and end"
period_start = datetime.datetime
"UTC date and time of the period start"
period_end = datetime.datetime
"UTC date and time of the period end"
def __init__(self, start_timestamp=None, end_timestamp=None, **kwds):
super(Statistics, self).__init__(**kwds)
self._update_duration(start_timestamp, end_timestamp)
def _update_duration(self, start_timestamp, end_timestamp):
# "Clamp" the timestamps we return to the original time
# range, excluding the offset.
if (start_timestamp and
self.duration_start and
self.duration_start < start_timestamp):
self.duration_start = start_timestamp
LOG.debug(_('clamping min timestamp to range'))
if (end_timestamp and
self.duration_end and
self.duration_end > end_timestamp):
self.duration_end = end_timestamp
LOG.debug(_('clamping max timestamp to range'))
# If we got valid timestamps back, compute a duration in seconds.
#
# If the min > max after clamping then we know the
# timestamps on the samples fell outside of the time
# range we care about for the query, so treat them as
# "invalid."
#
# If the timestamps are invalid, return None as a
# sentinal indicating that there is something "funny"
# about the range.
if (self.duration_start and
self.duration_end and
self.duration_start <= self.duration_end):
self.duration = timeutils.delta_seconds(self.duration_start,
self.duration_end)
else:
self.duration_start = self.duration_end = self.duration = None
@classmethod
def sample(cls):
return cls(unit='GiB',
min=1,
max=9,
avg=4.5,
sum=45,
count=10,
duration_start=datetime.datetime(2013, 1, 4, 16, 42),
duration_end=datetime.datetime(2013, 1, 4, 16, 47),
period=7200,
period_start=datetime.datetime(2013, 1, 4, 16, 00),
period_end=datetime.datetime(2013, 1, 4, 18, 00),
)
class MeterController(rest.RestController):
"""Manages operations on a single meter.
"""
_custom_actions = {
'statistics': ['GET'],
}
def __init__(self, meter_id):
pecan.request.context['meter_id'] = meter_id
self._id = meter_id
@wsme_pecan.wsexpose([Sample], [Query], int)
def get_all(self, q=[], limit=None):
"""Return samples for the meter.
:param q: Filter rules for the data to be returned.
:param limit: Maximum number of samples to return.
"""
if limit and limit < 0:
raise ValueError("Limit must be positive")
kwargs = _query_to_kwargs(q, storage.SampleFilter.__init__)
kwargs['meter'] = self._id
f = storage.SampleFilter(**kwargs)
return [Sample.from_db_model(e)
for e in pecan.request.storage_conn.get_samples(f, limit=limit)
]
@wsme_pecan.wsexpose([Sample], body=[Sample])
def post(self, samples):
"""Post a list of new Samples to Ceilometer.
:param samples: a list of samples within the request body.
"""
now = timeutils.utcnow()
auth_project = acl.get_limited_to_project(pecan.request.headers)
def_source = pecan.request.cfg.sample_source
def_project_id = pecan.request.headers.get('X-Project-Id')
def_user_id = pecan.request.headers.get('X-User-Id')
published_samples = []
for s in samples:
if self._id != s.counter_name:
raise wsme.exc.InvalidInput('counter_name', s.counter_name,
'should be %s' % self._id)
if s.message_id:
raise wsme.exc.InvalidInput('message_id', s.message_id,
'The message_id must not be set')
if s.counter_type not in sample.TYPES:
raise wsme.exc.InvalidInput('counter_type', s.counter_type,
'The counter type must be: ' +
', '.join(sample.TYPES))
s.user_id = (s.user_id or def_user_id)
s.project_id = (s.project_id or def_project_id)
s.source = '%s:%s' % (s.project_id, (s.source or def_source))
s.timestamp = (s.timestamp or now)
if auth_project and auth_project != s.project_id:
# non admin user trying to cross post to another project_id
auth_msg = 'can not post samples to other projects'
raise wsme.exc.InvalidInput('project_id', s.project_id,
auth_msg)
published_sample = sample.Sample(
name=s.counter_name,
type=s.counter_type,
unit=s.counter_unit,
volume=s.counter_volume,
user_id=s.user_id,
project_id=s.project_id,
resource_id=s.resource_id,
timestamp=s.timestamp.isoformat(),
resource_metadata=s.resource_metadata,
source=s.source)
published_samples.append(published_sample)
s.message_id = published_sample.id
with pecan.request.pipeline_manager.publisher(
context.get_admin_context()) as publisher:
publisher(published_samples)
return samples
@wsme_pecan.wsexpose([Statistics], [Query], [unicode], int)
def statistics(self, q=[], groupby=[], period=None):
"""Computes the statistics of the samples in the time range given.
:param q: Filter rules for the data to be returned.
:param groupby: Fields for group by aggregation
:param period: Returned result will be an array of statistics for a
period long of that number of seconds.
"""
if period and period < 0:
raise ClientSideError(_("Period must be positive."))
kwargs = _query_to_kwargs(q, storage.SampleFilter.__init__)
kwargs['meter'] = self._id
f = storage.SampleFilter(**kwargs)
g = _validate_groupby_fields(groupby)
computed = pecan.request.storage_conn.get_meter_statistics(f,
period,
g)
LOG.debug(_('computed value coming from %r'),
pecan.request.storage_conn)
# Find the original timestamp in the query to use for clamping
# the duration returned in the statistics.
start = end = None
for i in q:
if i.field == 'timestamp' and i.op in ('lt', 'le'):
end = timeutils.parse_isotime(i.value).replace(tzinfo=None)
elif i.field == 'timestamp' and i.op in ('gt', 'ge'):
start = timeutils.parse_isotime(i.value).replace(tzinfo=None)
return [Statistics(start_timestamp=start,
end_timestamp=end,
**c.as_dict())
for c in computed]
class Meter(_Base):
"""One category of measurements.
"""
name = wtypes.text
"The unique name for the meter"
type = wtypes.Enum(str, *sample.TYPES)
"The meter type (see :ref:`measurements`)"
unit = wtypes.text
"The unit of measure"
resource_id = wtypes.text
"The ID of the :class:`Resource` for which the measurements are taken"
project_id = wtypes.text
"The ID of the project or tenant that owns the resource"
user_id = wtypes.text
"The ID of the user who last triggered an update to the resource"
source = wtypes.text
"The ID of the source that identifies where the meter comes from"
meter_id = wtypes.text
"The unique identifier for the meter"
def __init__(self, **kwargs):
meter_id = base64.encodestring('%s+%s' % (kwargs['resource_id'],
kwargs['name']))
kwargs['meter_id'] = meter_id
super(Meter, self).__init__(**kwargs)
@classmethod
def sample(cls):
return cls(name='instance',
type='gauge',
unit='instance',
resource_id='bd9431c1-8d69-4ad3-803a-8d4a6b89fd36',
project_id='35b17138-b364-4e6a-a131-8f3099c5be68',
user_id='efd87807-12d2-4b38-9c70-5f5c2ac427ff',
source='openstack',
)
class MetersController(rest.RestController):
"""Works on meters."""
@pecan.expose()
def _lookup(self, meter_id, *remainder):
# NOTE(gordc): drop last path if empty (Bug #1202739)
if remainder and not remainder[-1]:
remainder = remainder[:-1]
return MeterController(meter_id), remainder
@wsme_pecan.wsexpose([Meter], [Query])
def get_all(self, q=[]):
"""Return all known meters, based on the data recorded so far.
:param q: Filter rules for the meters to be returned.
"""
kwargs = _query_to_kwargs(q, pecan.request.storage_conn.get_meters)
return [Meter.from_db_model(m)
for m in pecan.request.storage_conn.get_meters(**kwargs)]
class Resource(_Base):
"""An externally defined object for which samples have been received.
"""
resource_id = wtypes.text
"The unique identifier for the resource"
project_id = wtypes.text
"The ID of the owning project or tenant"
user_id = wtypes.text
"The ID of the user who created the resource or updated it last"
timestamp = datetime.datetime
"UTC date and time of the last update to any meter for the resource"
metadata = {wtypes.text: wtypes.text}
"Arbitrary metadata associated with the resource"
links = [Link]
"A list containing a self link and associated meter links"
source = wtypes.text
"The source where the resource come from"
def __init__(self, metadata={}, **kwds):
metadata = _flatten_metadata(metadata)
super(Resource, self).__init__(metadata=metadata, **kwds)
@classmethod
def sample(cls):
return cls(resource_id='bd9431c1-8d69-4ad3-803a-8d4a6b89fd36',
project_id='35b17138-b364-4e6a-a131-8f3099c5be68',
user_id='efd87807-12d2-4b38-9c70-5f5c2ac427ff',
timestamp=datetime.datetime.utcnow(),
source="openstack",
metadata={'name1': 'value1',
'name2': 'value2'},
links=[Link(href=('http://localhost:8777/v2/resources/'
'bd9431c1-8d69-4ad3-803a-8d4a6b89fd36'),
rel='self'),
Link(href=('http://localhost:8777/v2/meters/volume?'
'q.field=resource_id&'
'q.value=bd9431c1-8d69-4ad3-803a-'
'8d4a6b89fd36'),
rel='volume')],
)
class ResourcesController(rest.RestController):
"""Works on resources."""
def _resource_links(self, resource_id):
links = [_make_link('self', pecan.request.host_url, 'resources',
resource_id)]
for meter in pecan.request.storage_conn.get_meters(resource=
resource_id):
query = {'field': 'resource_id', 'value': resource_id}
links.append(_make_link(meter.name, pecan.request.host_url,
'meters', meter.name, query=query))
return links
@wsme_pecan.wsexpose(Resource, unicode)
def get_one(self, resource_id):
"""Retrieve details about one resource.
:param resource_id: The UUID of the resource.
"""
authorized_project = acl.get_limited_to_project(pecan.request.headers)
resources = list(pecan.request.storage_conn.get_resources(
resource=resource_id, project=authorized_project))
if not resources:
raise EntityNotFound(_('Resource'), resource_id)
return Resource.from_db_and_links(resources[0],
self._resource_links(resource_id))
@wsme_pecan.wsexpose([Resource], [Query])
def get_all(self, q=[]):
"""Retrieve definitions of all of the resources.
:param q: Filter rules for the resources to be returned.
"""
kwargs = _query_to_kwargs(q, pecan.request.storage_conn.get_resources)
resources = [
Resource.from_db_and_links(r,
self._resource_links(r.resource_id))
for r in pecan.request.storage_conn.get_resources(**kwargs)]
return resources
class AlarmThresholdRule(_Base):
meter_name = wsme.wsattr(wtypes.text, mandatory=True)
"The name of the meter"
#FIXME(sileht): default doesn't work
#workaround: default is set in validate method
query = wsme.wsattr([Query], default=[])
"""The query to find the data for computing statistics.
Ownership settings are automatically included based on the Alarm owner.
"""
period = wsme.wsattr(BoundedInt(min=1), default=60)
"The time range in seconds over which query"
comparison_operator = AdvEnum('comparison_operator', str,
'lt', 'le', 'eq', 'ne', 'ge', 'gt',
default='eq')
"The comparison against the alarm threshold"
threshold = wsme.wsattr(float, mandatory=True)
"The threshold of the alarm"
statistic = AdvEnum('statistic', str, 'max', 'min', 'avg', 'sum',
'count', default='avg')
"The statistic to compare to the threshold"
evaluation_periods = wsme.wsattr(BoundedInt(min=1), default=1)
"The number of historical periods to evaluate the threshold"
def __init__(self, query=None, **kwargs):
if query:
query = [Query(**q) for q in query]
super(AlarmThresholdRule, self).__init__(query=query, **kwargs)
@staticmethod
def validate(threshold_rule):
#note(sileht): wsme default doesn't work in some case
#workaround for https://bugs.launchpad.net/wsme/+bug/1227039
if not threshold_rule.query:
threshold_rule.query = []
timestamp_keys = ['timestamp', 'start', 'start_timestamp' 'end',
'end_timestamp']
_validate_query(threshold_rule.query, storage.SampleFilter.__init__,
internal_keys=timestamp_keys)
return threshold_rule
@property
def default_description(self):
return _(
'Alarm when %(meter_name)s is %(comparison_operator)s a '
'%(statistic)s of %(threshold)s over %(period)s seconds') % \
dict(comparison_operator=self.comparison_operator,
statistic=self.statistic,
threshold=self.threshold,
meter_name=self.meter_name,
period=self.period)
def as_dict(self):
rule = self.as_dict_from_keys(['period', 'comparison_operator',
'threshold', 'statistic',
'evaluation_periods', 'meter_name'])
rule['query'] = [q.as_dict() for q in self.query]
return rule
@classmethod
def sample(cls):
return cls(meter_name='cpu_util',
period=60,
evaluation_periods=1,
threshold=300.0,
statistic='avg',
comparison_operator='gt',
query=[{'field': 'resource_id',
'value': '2a4d689b-f0b8-49c1-9eef-87cae58d80db',
'op': 'eq',
'type': 'string'}])
class AlarmCombinationRule(_Base):
operator = AdvEnum('operator', str, 'or', 'and', default='and')
"How to combine the sub-alarms"
alarm_ids = wsme.wsattr([wtypes.text], mandatory=True)
"List of alarm identifiers to combine"
@property
def default_description(self):
return _('Combined state of alarms %s') % self.operator.join(
self.alarm_ids)
def as_dict(self):
return self.as_dict_from_keys(['operator', 'alarm_ids'])
@classmethod
def sample(cls):
return cls(operator='or',
alarm_ids=['739e99cb-c2ec-4718-b900-332502355f38',
'153462d0-a9b8-4b5b-8175-9e4b05e9b856'])
class Alarm(_Base):
"""Representation of an alarm.
.. note::
combination_rule and threshold_rule are mutually exclusive.
"""
alarm_id = wtypes.text
"The UUID of the alarm"
name = wsme.wsattr(wtypes.text, mandatory=True)
"The name for the alarm"
_description = None # provide a default
def get_description(self):
rule = getattr(self, '%s_rule' % self.type, None)
if not self._description and rule:
return six.text_type(rule.default_description)
return self._description
def set_description(self, value):
self._description = value
description = wsme.wsproperty(wtypes.text, get_description,
set_description)
"The description of the alarm"
enabled = wsme.wsattr(bool, default=True)
"This alarm is enabled?"
ok_actions = wsme.wsattr([wtypes.text], default=[])
"The actions to do when alarm state change to ok"
alarm_actions = wsme.wsattr([wtypes.text], default=[])
"The actions to do when alarm state change to alarm"
insufficient_data_actions = wsme.wsattr([wtypes.text], default=[])
"The actions to do when alarm state change to insufficient data"
repeat_actions = wsme.wsattr(bool, default=False)
"The actions should be re-triggered on each evaluation cycle"
type = AdvEnum('type', str, 'threshold', 'combination', mandatory=True)
"Explicit type specifier to select which rule to follow below."
threshold_rule = AlarmThresholdRule
"Describe when to trigger the alarm based on computed statistics"
combination_rule = AlarmCombinationRule
"""Describe when to trigger the alarm based on combining the state of
other alarms"""
# These settings are ignored in the PUT or POST operations, but are
# filled in for GET
project_id = wtypes.text
"The ID of the project or tenant that owns the alarm"
user_id = wtypes.text
"The ID of the user who created the alarm"
timestamp = datetime.datetime
"The date of the last alarm definition update"
state = AdvEnum('state', str, *state_kind,
default='insufficient data')
"The state offset the alarm"
state_timestamp = datetime.datetime
"The date of the last alarm state changed"
def __init__(self, rule=None, **kwargs):
super(Alarm, self).__init__(**kwargs)
if rule:
if self.type == 'threshold':
self.threshold_rule = AlarmThresholdRule(**rule)
elif self.type == 'combination':
self.combination_rule = AlarmCombinationRule(**rule)
@staticmethod
def validate(alarm):
if (alarm.threshold_rule == wtypes.Unset
and alarm.combination_rule == wtypes.Unset):
error = _("either threshold_rule or combination_rule "
"must be set")
raise ClientSideError(error)
if alarm.threshold_rule and alarm.combination_rule:
error = _("threshold_rule and combination_rule "
"cannot be set at the same time")
raise ClientSideError(error)
if alarm.threshold_rule:
# ensure an implicit constraint on project_id is added to
# the query if not already present
alarm.threshold_rule.query = _sanitize_query(
alarm.threshold_rule.query,
storage.SampleFilter.__init__,
on_behalf_of=alarm.project_id
)
elif alarm.combination_rule:
project = _get_auth_project(alarm.project_id
if alarm.project_id != wtypes.Unset
else None)
for id in alarm.combination_rule.alarm_ids:
alarms = list(pecan.request.storage_conn.get_alarms(
alarm_id=id, project=project))
if not alarms:
raise ClientSideError(_("Alarm %s doesn't exist") % id)
return alarm
@classmethod
def sample(cls):
return cls(alarm_id=None,
name="SwiftObjectAlarm",
description="An alarm",
type='threshold',
threshold_rule=None,
combination_rule=None,
user_id="c96c887c216949acbdfbd8b494863567",
project_id="c96c887c216949acbdfbd8b494863567",
enabled=True,
timestamp=datetime.datetime.utcnow(),
state="ok",
state_timestamp=datetime.datetime.utcnow(),
ok_actions=["http://site:8000/ok"],
alarm_actions=["http://site:8000/alarm"],
insufficient_data_actions=["http://site:8000/nodata"],
repeat_actions=False,
)
def as_dict(self, db_model):
d = super(Alarm, self).as_dict(db_model)
for k in d:
if k.endswith('_rule'):
del d[k]
d['rule'] = getattr(self, "%s_rule" % self.type).as_dict()
return d
class AlarmChange(_Base):
"""Representation of an event in an alarm's history
"""
event_id = wtypes.text
"The UUID of the change event"
alarm_id = wtypes.text
"The UUID of the alarm"
type = wtypes.Enum(str,
'creation',
'rule change',
'state transition',
'deletion')
"The type of change"
detail = wtypes.text
"JSON fragment describing change"
project_id = wtypes.text
"The project ID of the initiating identity"
user_id = wtypes.text
"The user ID of the initiating identity"
on_behalf_of = wtypes.text
"The tenant on behalf of which the change is being made"
timestamp = datetime.datetime
"The time/date of the alarm change"
@classmethod
def sample(cls):
return cls(alarm_id='e8ff32f772a44a478182c3fe1f7cad6a',
type='rule change',
detail='{"threshold": 42.0, "evaluation_periods": 4}',
user_id="3e5d11fda79448ac99ccefb20be187ca",
project_id="b6f16144010811e387e4de429e99ee8c",
on_behalf_of="92159030020611e3b26dde429e99ee8c",
timestamp=datetime.datetime.utcnow(),
)
class AlarmController(rest.RestController):
"""Manages operations on a single alarm.
"""
_custom_actions = {
'history': ['GET'],
'state': ['PUT', 'GET'],
}
def __init__(self, alarm_id):
pecan.request.context['alarm_id'] = alarm_id
self._id = alarm_id
def _alarm(self):
self.conn = pecan.request.storage_conn
auth_project = acl.get_limited_to_project(pecan.request.headers)
alarms = list(self.conn.get_alarms(alarm_id=self._id,
project=auth_project))
if len(alarms) < 1:
raise EntityNotFound(_('Alarm'), self._id)
return alarms[0]
def _record_change(self, data, now, on_behalf_of=None, type=None):
if not cfg.CONF.alarm.record_history:
return
type = type or storage.models.AlarmChange.RULE_CHANGE
scrubbed_data = utils.stringify_timestamps(data)
detail = json.dumps(scrubbed_data)
user_id = pecan.request.headers.get('X-User-Id')
project_id = pecan.request.headers.get('X-Project-Id')
on_behalf_of = on_behalf_of or project_id
payload = dict(event_id=str(uuid.uuid4()),
alarm_id=self._id,
type=type,
detail=detail,
user_id=user_id,
project_id=project_id,
on_behalf_of=on_behalf_of,
timestamp=now)
try:
self.conn.record_alarm_change(payload)
except NotImplementedError:
pass
# Revert to the pre-json'ed details ...
payload['detail'] = scrubbed_data
_send_notification(type, payload)
@wsme_pecan.wsexpose(Alarm)
def get(self):
"""Return this alarm.
"""
return Alarm.from_db_model(self._alarm())
@wsme_pecan.wsexpose(Alarm, body=Alarm)
def put(self, data):
"""Modify this alarm.
:param data: a alarm within the request body.
"""
# Ensure alarm exists
alarm_in = self._alarm()
now = timeutils.utcnow()
data.alarm_id = self._id
user, project = acl.get_limited_to(pecan.request.headers)
if user:
data.user_id = user
elif data.user_id == wtypes.Unset:
data.user_id = alarm_in.user_id
if project:
data.project_id = project
elif data.project_id == wtypes.Unset:
data.project_id = alarm_in.project_id
data.timestamp = now
if alarm_in.state != data.state:
data.state_timestamp = now
else:
data.state_timestamp = alarm_in.state_timestamp
old_alarm = Alarm.from_db_model(alarm_in).as_dict(storage.models.Alarm)
updated_alarm = data.as_dict(storage.models.Alarm)
try:
alarm_in = storage.models.Alarm(**updated_alarm)
except Exception:
LOG.exception(_("Error while putting alarm: %s") % updated_alarm)
raise ClientSideError(_("Alarm incorrect"))
alarm = self.conn.update_alarm(alarm_in)
change = dict((k, v) for k, v in updated_alarm.items()
if v != old_alarm[k] and k not in
['timestamp', 'state_timestamp'])
self._record_change(change, now, on_behalf_of=alarm.project_id)
return Alarm.from_db_model(alarm)
@wsme_pecan.wsexpose(None, status_code=204)
def delete(self):
"""Delete this alarm.
"""
# ensure alarm exists before deleting
alarm = self._alarm()
self.conn.delete_alarm(alarm.alarm_id)
change = Alarm.from_db_model(alarm).as_dict(storage.models.Alarm)
self._record_change(change,
timeutils.utcnow(),
type=storage.models.AlarmChange.DELETION)
# TODO(eglynn): add pagination marker to signature once overall
# API support for pagination is finalized
@wsme_pecan.wsexpose([AlarmChange], [Query])
def history(self, q=[]):
"""Assembles the alarm history requested.
:param q: Filter rules for the changes to be described.
"""
# allow history to be returned for deleted alarms, but scope changes
# returned to those carried out on behalf of the auth'd tenant, to
# avoid inappropriate cross-tenant visibility of alarm history
auth_project = acl.get_limited_to_project(pecan.request.headers)
conn = pecan.request.storage_conn
kwargs = _query_to_kwargs(q, conn.get_alarm_changes, ['on_behalf_of'])
return [AlarmChange.from_db_model(ac)
for ac in conn.get_alarm_changes(self._id, auth_project,
**kwargs)]
@wsme.validate(state_kind_enum)
@wsme_pecan.wsexpose(state_kind_enum, body=state_kind_enum)
def put_state(self, state):
"""Set the state of this alarm.
:param state: a alarm state within the request body.
"""
# note(sileht): body are not validated by wsme
# Workaround for https://bugs.launchpad.net/wsme/+bug/1227229
if state not in state_kind:
raise ClientSideError(_("state invalid"))
now = timeutils.utcnow()
alarm = self._alarm()
alarm.state = state
alarm.state_timestamp = now
alarm = self.conn.update_alarm(alarm)
change = {'state': alarm.state}
self._record_change(change, now, on_behalf_of=alarm.project_id,
type=storage.models.AlarmChange.STATE_TRANSITION)
return alarm.state
@wsme_pecan.wsexpose(state_kind_enum)
def get_state(self):
"""Get the state of this alarm.
"""
alarm = self._alarm()
return alarm.state
class AlarmsController(rest.RestController):
"""Manages operations on the alarms collection.
"""
@pecan.expose()
def _lookup(self, alarm_id, *remainder):
if remainder and not remainder[-1]:
remainder = remainder[:-1]
return AlarmController(alarm_id), remainder
def _record_creation(self, conn, data, alarm_id, now):
if not cfg.CONF.alarm.record_history:
return
type = storage.models.AlarmChange.CREATION
scrubbed_data = utils.stringify_timestamps(data)
detail = json.dumps(scrubbed_data)
user_id = pecan.request.headers.get('X-User-Id')
project_id = pecan.request.headers.get('X-Project-Id')
payload = dict(event_id=str(uuid.uuid4()),
alarm_id=alarm_id,
type=type,
detail=detail,
user_id=user_id,
project_id=project_id,
on_behalf_of=project_id,
timestamp=now)
try:
conn.record_alarm_change(payload)
except NotImplementedError:
pass
# Revert to the pre-json'ed details ...
payload['detail'] = scrubbed_data
_send_notification(type, payload)
@wsme_pecan.wsexpose(Alarm, body=Alarm, status_code=201)
def post(self, data):
"""Create a new alarm.
:param data: a alarm within the request body.
"""
conn = pecan.request.storage_conn
now = timeutils.utcnow()
data.alarm_id = str(uuid.uuid4())
user, project = acl.get_limited_to(pecan.request.headers)
if user:
data.user_id = user
elif data.user_id == wtypes.Unset:
data.user_id = pecan.request.headers.get('X-User-Id')
if project:
data.project_id = project
elif data.project_id == wtypes.Unset:
data.project_id = pecan.request.headers.get('X-Project-Id')
data.timestamp = now
data.state_timestamp = now
change = data.as_dict(storage.models.Alarm)
# make sure alarms are unique by name per project.
alarms = list(conn.get_alarms(name=data.name,
project=data.project_id))
if len(alarms) > 0:
raise ClientSideError(_("Alarm with that name exists"))
try:
alarm_in = storage.models.Alarm(**change)
except Exception:
LOG.exception(_("Error while posting alarm: %s") % change)
raise ClientSideError(_("Alarm incorrect"))
alarm = conn.create_alarm(alarm_in)
self._record_creation(conn, change, alarm.alarm_id, now)
return Alarm.from_db_model(alarm)
@wsme_pecan.wsexpose([Alarm], [Query])
def get_all(self, q=[]):
"""Return all alarms, based on the query provided.
:param q: Filter rules for the alarms to be returned.
"""
kwargs = _query_to_kwargs(q,
pecan.request.storage_conn.get_alarms)
return [Alarm.from_db_model(m)
for m in pecan.request.storage_conn.get_alarms(**kwargs)]
class V2Controller(object):
"""Version 2 API controller root."""
resources = ResourcesController()
meters = MetersController()
alarms = AlarmsController()