# -*- encoding: utf-8 -*- # # Copyright © 2012, 2013 Dell Inc. # # Author: Stas Maksimov # Author: Shengjie Min # # 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. """HBase storage backend """ import copy import datetime import hashlib import itertools import json import os import re import urlparse import happybase from ceilometer.openstack.common.gettextutils import _ # noqa from ceilometer.openstack.common import log from ceilometer.openstack.common import network_utils from ceilometer.openstack.common import timeutils from ceilometer.storage import base from ceilometer.storage import models LOG = log.getLogger(__name__) class HBaseStorage(base.StorageEngine): """Put the data into a HBase database Collections: - user - { _id: user id source: [ array of source ids reporting for the user ] } - project - { _id: project id source: [ array of source ids reporting for the project ] } - meter - the raw incoming data - resource - the metadata for resources - { _id: uuid of resource, metadata: metadata dictionaries timestamp: datetime of last update user_id: uuid project_id: uuid meter: [ array of {counter_name: string, counter_type: string} ] } """ @staticmethod def get_connection(conf): """Return a Connection instance based on the configuration settings. """ return Connection(conf) class Connection(base.Connection): """HBase connection. """ _memory_instance = None PROJECT_TABLE = "project" USER_TABLE = "user" RESOURCE_TABLE = "resource" METER_TABLE = "meter" def __init__(self, conf): """Hbase Connection Initialization.""" opts = self._parse_connection_url(conf.database.connection) if opts['host'] == '__test__': url = os.environ.get('CEILOMETER_TEST_HBASE_URL') if url: # Reparse URL, but from the env variable now opts = self._parse_connection_url(url) self.conn = self._get_connection(opts) else: # This is a in-memory usage for unit tests if Connection._memory_instance is None: LOG.debug(_('Creating a new in-memory HBase ' 'Connection object')) Connection._memory_instance = MConnection() self.conn = Connection._memory_instance else: self.conn = self._get_connection(opts) self.conn.open() def upgrade(self): self.conn.create_table(self.PROJECT_TABLE, {'f': dict()}) self.conn.create_table(self.USER_TABLE, {'f': dict()}) self.conn.create_table(self.RESOURCE_TABLE, {'f': dict()}) self.conn.create_table(self.METER_TABLE, {'f': dict()}) def clear(self): LOG.debug(_('Dropping HBase schema...')) for table in [self.PROJECT_TABLE, self.USER_TABLE, self.RESOURCE_TABLE, self.METER_TABLE]: try: self.conn.disable_table(table) except Exception: LOG.debug(_('Cannot disable table but ignoring error')) try: self.conn.delete_table(table) except Exception: LOG.debug(_('Cannot delete table but ignoring error')) @staticmethod def _get_connection(conf): """Return a connection to the database. .. note:: The tests use a subclass to override this and return an in-memory connection. """ LOG.debug(_('connecting to HBase on %(host)s:%(port)s') % ( {'host': conf['host'], 'port': conf['port']})) return happybase.Connection(host=conf['host'], port=conf['port'], table_prefix=conf['table_prefix']) @staticmethod def _parse_connection_url(url): """Parse connection parameters from a database url. .. note:: HBase Thrift does not support authentication and there is no database name, so we are not looking for these in the url. """ opts = {} result = network_utils.urlsplit(url) opts['table_prefix'] = urlparse.parse_qs( result.query).get('table_prefix', [None])[0] opts['dbtype'] = result.scheme if ':' in result.netloc: opts['host'], port = result.netloc.split(':') else: opts['host'] = result.netloc port = 9090 opts['port'] = port and int(port) or 9090 return opts def record_metering_data(self, data): """Write the data to the backend storage system. :param data: a dictionary such as returned by ceilometer.meter.meter_message_from_counter """ project_table = self.conn.table(self.PROJECT_TABLE) user_table = self.conn.table(self.USER_TABLE) resource_table = self.conn.table(self.RESOURCE_TABLE) meter_table = self.conn.table(self.METER_TABLE) # store metadata fields with prefix "r_" resource_metadata = {} if data['resource_metadata']: resource_metadata = dict(('f:r_%s' % k, v) for (k, v) in data['resource_metadata'].iteritems()) # Make sure we know about the user and project if data['user_id']: user = user_table.row(data['user_id']) sources = _load_hbase_list(user, 's') # Update if source is new if data['source'] not in sources: user['f:s_%s' % data['source']] = "1" user_table.put(data['user_id'], user) project = project_table.row(data['project_id']) sources = _load_hbase_list(project, 's') # Update if source is new if data['source'] not in sources: project['f:s_%s' % data['source']] = "1" project_table.put(data['project_id'], project) rts = reverse_timestamp(data['timestamp']) resource = resource_table.row(data['resource_id']) new_meter = _format_meter_reference( data['counter_name'], data['counter_type'], data['counter_unit']) new_resource = {'f:resource_id': data['resource_id'], 'f:project_id': data['project_id'], 'f:user_id': data['user_id'], 'f:source': data["source"], # store meters with prefix "m_" 'f:m_%s' % new_meter: "1" } new_resource.update(resource_metadata) # Update if resource has new information if new_resource != resource: meters = _load_hbase_list(resource, 'm') if new_meter not in meters: new_resource['f:m_%s' % new_meter] = "1" resource_table.put(data['resource_id'], new_resource) # Rowkey consists of reversed timestamp, meter and an md5 of # user+resource+project for purposes of uniqueness m = hashlib.md5() m.update("%s%s%s" % (data['user_id'], data['resource_id'], data['project_id'])) # We use reverse timestamps in rowkeys as they are sorted # alphabetically. row = "%s_%d_%s" % (data['counter_name'], rts, m.hexdigest()) # Convert timestamp to string as json.dumps won't ts = timeutils.strtime(data['timestamp']) record = {'f:timestamp': ts, 'f:counter_name': data['counter_name'], 'f:counter_type': data['counter_type'], 'f:counter_volume': str(data['counter_volume']), 'f:counter_unit': data['counter_unit'], # TODO(shengjie) consider using QualifierFilter # keep dimensions as column qualifier for quicker look up # TODO(shengjie) extra dimensions need to be added as CQ 'f:user_id': data['user_id'], 'f:project_id': data['project_id'], 'f:resource_id': data['resource_id'], 'f:source': data['source'], # add in reversed_ts here for time range scan 'f:rts': str(rts) } # Need to record resource_metadata for more robust filtering. record.update(resource_metadata) # Don't want to be changing the original data object. data = copy.copy(data) data['timestamp'] = ts # Save original meter. record['f:message'] = json.dumps(data) meter_table.put(row, record) def get_users(self, source=None): """Return an iterable of user id strings. :param source: Optional source filter. """ user_table = self.conn.table(self.USER_TABLE) LOG.debug(_("source: %s") % source) scan_args = {} if source: scan_args['columns'] = ['f:s_%s' % source] return sorted(key for key, ignored in user_table.scan(**scan_args)) def get_projects(self, source=None): """Return an iterable of project id strings. :param source: Optional source filter. """ project_table = self.conn.table(self.PROJECT_TABLE) LOG.debug(_("source: %s") % source) scan_args = {} if source: scan_args['columns'] = ['f:s_%s' % source] return (key for key, ignored in project_table.scan(**scan_args)) def get_resources(self, user=None, project=None, source=None, start_timestamp=None, start_timestamp_op=None, end_timestamp=None, end_timestamp_op=None, metaquery={}, resource=None, pagination=None): """Return an iterable of models.Resource instances :param user: Optional ID for user that owns the resource. :param project: Optional ID for project that owns the resource. :param source: Optional source filter. :param start_timestamp: Optional modified timestamp start range. :param start_timestamp_op: Optional start time operator, like ge, gt. :param end_timestamp: Optional modified timestamp end range. :param end_timestamp_op: Optional end time operator, like lt, le. :param metaquery: Optional dict with metadata to match on. :param resource: Optional resource filter. :param pagination: Optional pagination query. """ if pagination: raise NotImplementedError(_('Pagination not implemented')) def make_resource(data, first_ts, last_ts): """Transform HBase fields to Resource model.""" # convert HBase metadata e.g. f:r_display_name to display_name data['f:metadata'] = _metadata_from_document(data) return models.Resource( resource_id=data['f:resource_id'], first_sample_timestamp=first_ts, last_sample_timestamp=last_ts, project_id=data['f:project_id'], source=data['f:source'], user_id=data['f:user_id'], metadata=data['f:metadata'], ) meter_table = self.conn.table(self.METER_TABLE) q, start_row, stop_row = make_query(user=user, project=project, source=source, resource=resource, start=start_timestamp, start_op=start_timestamp_op, end=end_timestamp, end_op=end_timestamp_op, require_meter=False, query_only=False) LOG.debug(_("Query Meter table: %s") % q) meters = meter_table.scan(filter=q, row_start=start_row, row_stop=stop_row) # We have to sort on resource_id before we can group by it. According # to the itertools documentation a new group is generated when the # value of the key function changes (it breaks there). meters = sorted(meters, key=_resource_id_from_record_tuple) for resource_id, r_meters in itertools.groupby( meters, key=_resource_id_from_record_tuple): meter_rows = [data[1] for data in sorted( r_meters, key=_timestamp_from_record_tuple)] latest_data = meter_rows[-1] min_ts = timeutils.parse_strtime(meter_rows[0]['f:timestamp']) max_ts = timeutils.parse_strtime(latest_data['f:timestamp']) if metaquery: for k, v in metaquery.iteritems(): if latest_data['f:r_' + k.split('.', 1)[1]] == v: yield make_resource( latest_data, min_ts, max_ts ) else: yield make_resource( latest_data, min_ts, max_ts ) def get_meters(self, user=None, project=None, resource=None, source=None, metaquery={}, pagination=None): """Return an iterable of models.Meter instances :param user: Optional ID for user that owns the resource. :param project: Optional ID for project that owns the resource. :param resource: Optional resource filter. :param source: Optional source filter. :param metaquery: Optional dict with metadata to match on. :param pagination: Optional pagination query. """ if pagination: raise NotImplementedError(_('Pagination not implemented')) resource_table = self.conn.table(self.RESOURCE_TABLE) q = make_query(user=user, project=project, resource=resource, source=source, require_meter=False, query_only=True) LOG.debug(_("Query Resource table: %s") % q) # handle metaquery if len(metaquery) > 0: meta_q = [] for k, v in metaquery.iteritems(): meta_q.append( "SingleColumnValueFilter ('f', '%s', =, 'binary:%s')" % ('r_' + k.split('.', 1)[1], v)) meta_q = " AND ".join(meta_q) # join query and metaquery if q is not None: q += " AND " + meta_q else: q = meta_q # metaquery only gen = resource_table.scan(filter=q) for ignored, data in gen: # Meter columns are stored like this: # "m_{counter_name}|{counter_type}|{counter_unit}" => "1" # where 'm' is a prefix (m for meter), value is always set to 1 meter = None for m in data: if m.startswith('f:m_'): meter = m break if meter is None: continue name, type, unit = meter[4:].split("!") yield models.Meter( name=name, type=type, unit=unit, resource_id=data['f:resource_id'], project_id=data['f:project_id'], source=data['f:source'], user_id=data['f:user_id'], ) def get_samples(self, sample_filter, limit=None): """Return an iterable of models.Sample instances. :param sample_filter: Filter. :param limit: Maximum number of results to return. """ def make_sample(data): """Transform HBase fields to Sample model.""" data = json.loads(data['f:message']) data['timestamp'] = timeutils.parse_strtime(data['timestamp']) return models.Sample(**data) meter_table = self.conn.table(self.METER_TABLE) q, start, stop = make_query_from_filter(sample_filter, require_meter=False) LOG.debug(_("Query Meter Table: %s") % q) gen = meter_table.scan(filter=q, row_start=start, row_stop=stop) for ignored, meter in gen: # TODO(shengjie) put this implementation here because it's failing # the test. bp hbase-meter-table-enhancement will address this # properly. # handle metaquery metaquery = sample_filter.metaquery # TODO(jd) implements using HBase capabilities if limit == 0: break if len(metaquery) > 0: for k, v in metaquery.iteritems(): message = json.loads(meter['f:message']) metadata = message['resource_metadata'] keys = k.split('.') # Support the dictionary type of metadata for key in keys[1:]: if key in metadata: metadata = metadata[key] else: break # NOTE (flwang) For multiple level searching, the matadata # object will be drilled down to check if it's matched # with the searched value. if metadata != v: break else: if limit: limit -= 1 yield make_sample(meter) else: if limit: limit -= 1 yield make_sample(meter) @staticmethod def _update_meter_stats(stat, meter): """Do the stats calculation on a requested time bucket in stats dict :param stats: dict where aggregated stats are kept :param index: time bucket index in stats :param meter: meter record as returned from HBase :param start_time: query start time :param period: length of the time bucket """ vol = int(meter['f:counter_volume']) ts = timeutils.parse_strtime(meter['f:timestamp']) stat.unit = meter['f:counter_unit'] stat.min = min(vol, stat.min or vol) stat.max = max(vol, stat.max) stat.sum = vol + (stat.sum or 0) stat.count += 1 stat.avg = (stat.sum / float(stat.count)) stat.duration_start = min(ts, stat.duration_start or ts) stat.duration_end = max(ts, stat.duration_end or ts) stat.duration = \ timeutils.delta_seconds(stat.duration_start, stat.duration_end) def get_meter_statistics(self, sample_filter, period=None, groupby=None): """Return an iterable of models.Statistics instances containing meter statistics described by the query parameters. The filter must have a meter value set. .. note:: Due to HBase limitations the aggregations are implemented in the driver itself, therefore this method will be quite slow because of all the Thrift traffic it is going to create. """ if groupby: raise NotImplementedError("Group by not implemented.") meter_table = self.conn.table(self.METER_TABLE) q, start, stop = make_query_from_filter(sample_filter) meters = list(meter for (ignored, meter) in meter_table.scan(filter=q, row_start=start, row_stop=stop) ) if sample_filter.start: start_time = sample_filter.start elif meters: start_time = timeutils.parse_strtime(meters[-1]['f:timestamp']) else: start_time = None if sample_filter.end: end_time = sample_filter.end elif meters: end_time = timeutils.parse_strtime(meters[0]['f:timestamp']) else: end_time = None results = [] if not period: period = 0 period_start = start_time period_end = end_time # As our HBase meters are stored as newest-first, we need to iterate # in the reverse order for meter in meters[::-1]: ts = timeutils.parse_strtime(meter['f:timestamp']) if period: offset = int(timeutils.delta_seconds( start_time, ts) / period) * period period_start = start_time + datetime.timedelta(0, offset) if not len(results) or not results[-1].period_start == \ period_start: if period: period_end = period_start + datetime.timedelta( 0, period) results.append( models.Statistics(unit='', count=0, min=0, max=0, avg=0, sum=0, period=period, period_start=period_start, period_end=period_end, duration=None, duration_start=None, duration_end=None, groupby=None) ) self._update_meter_stats(results[-1], meter) return results ############### # This is a very crude version of "in-memory HBase", which implements just # enough functionality of HappyBase API to support testing of our driver. # class MTable(object): """HappyBase.Table mock """ def __init__(self, name, families): self.name = name self.families = families self._rows = {} def row(self, key): return self._rows.get(key, {}) def rows(self, keys): return ((k, self.row(k)) for k in keys) def put(self, key, data): self._rows[key] = data def scan(self, filter=None, columns=[], row_start=None, row_stop=None): sorted_keys = sorted(self._rows) # copy data between row_start and row_stop into a dict rows = {} for row in sorted_keys: if row_start and row < row_start: continue if row_stop and row > row_stop: break rows[row] = copy.copy(self._rows[row]) if columns: ret = {} for row in rows.keys(): data = rows[row] for key in data: if key in columns: ret[row] = data rows = ret elif filter: # TODO(jdanjou): we should really parse this properly, # but at the moment we are only going to support AND here filters = filter.split('AND') for f in filters: # Extract filter name and its arguments g = re.search("(.*)\((.*),?\)", f) fname = g.group(1).strip() fargs = [s.strip().replace('\'', '').replace('\"', '') for s in g.group(2).split(',')] m = getattr(self, fname) if callable(m): # overwrite rows for filtering to take effect # in case of multiple filters rows = m(fargs, rows) else: raise NotImplementedError("%s filter is not implemented, " "you may want to add it!") for k in sorted(rows): yield k, rows[k] @staticmethod def SingleColumnValueFilter(args, rows): """This method is called from scan() when 'SingleColumnValueFilter' is found in the 'filter' argument """ op = args[2] column = "%s:%s" % (args[0], args[1]) value = args[3] if value.startswith('binary:'): value = value[7:] r = {} for row in rows: data = rows[row] if op == '=': if column in data and data[column] == value: r[row] = data elif op == '<=': if column in data and data[column] <= value: r[row] = data elif op == '>=': if column in data and data[column] >= value: r[row] = data else: raise NotImplementedError("In-memory " "SingleColumnValueFilter " "doesn't support the %s operation " "yet" % op) return r class MConnection(object): """HappyBase.Connection mock """ def __init__(self): self.tables = {} def open(self): LOG.debug(_("Opening in-memory HBase connection")) def create_table(self, n, families={}): if n in self.tables: return self.tables[n] t = MTable(n, families) self.tables[n] = t return t def delete_table(self, name, use_prefix=True): del self.tables[name] def table(self, name): return self.create_table(name) ################################################# # Here be various HBase helpers def reverse_timestamp(dt): """Reverse timestamp so that newer timestamps are represented by smaller numbers than older ones. Reverse timestamps is a technique used in HBase rowkey design. When period queries are required the HBase rowkeys must include timestamps, but as rowkeys in HBase are ordered lexicographically, the timestamps must be reversed. """ epoch = datetime.datetime(1970, 1, 1) td = dt - epoch ts = (td.microseconds + (td.seconds + td.days * 24 * 3600) * 100000) / 100000 return 0x7fffffffffffffff - ts def make_query(user=None, project=None, meter=None, resource=None, source=None, start=None, start_op=None, end=None, end_op=None, require_meter=True, query_only=False): """Return a filter query string based on the selected parameters. :param user: Optional user-id :param project: Optional project-id :param meter: Optional counter-name :param resource: Optional resource-id :param source: Optional source-id :param start: Optional start timestamp :param start_op: Optional start timestamp operator, like gt, ge :param end: Optional end timestamp :param end_op: Optional end timestamp operator, like lt, le :param require_meter: If true and the filter does not have a meter, raise an error. :param query_only: If true only returns the filter query, otherwise also returns start and stop rowkeys """ q = [] if user: q.append("SingleColumnValueFilter ('f', 'user_id', =, 'binary:%s')" % user) if project: q.append("SingleColumnValueFilter ('f', 'project_id', =, 'binary:%s')" % project) if resource: q.append("SingleColumnValueFilter ('f', 'resource_id', =, 'binary:%s')" % resource) if source: q.append("SingleColumnValueFilter " "('f', 'source', =, 'binary:%s')" % source) start_row, end_row = "", "" rts_start = str(reverse_timestamp(start) + 1) if start else "" rts_end = str(reverse_timestamp(end) + 1) if end else "" #By default, we are using ge for lower bound and lt for upper bound if start_op == 'gt': rts_start = str(long(rts_start) - 2) if end_op == 'le': rts_end = str(long(rts_end) - 1) # when start_time and end_time is provided, # if it's filtered by meter, # rowkey will be used in the query; # else it's non meter filter query(e.g. project_id, user_id etc), # SingleColumnValueFilter against rts will be appended to the query # query other tables should have no start and end passed in if meter: start_row, end_row = _make_rowkey_scan(meter, rts_start, rts_end) q.append("SingleColumnValueFilter " "('f', 'counter_name', =, 'binary:%s')" % meter) elif require_meter: raise RuntimeError('Missing required meter specifier') else: if rts_start: q.append("SingleColumnValueFilter ('f', 'rts', <=, 'binary:%s')" % rts_start) if rts_end: q.append("SingleColumnValueFilter ('f', 'rts', >=, 'binary:%s')" % rts_end) sample_filter = None if len(q): sample_filter = " AND ".join(q) if query_only: return sample_filter else: return sample_filter, start_row, end_row def make_query_from_filter(sample_filter, require_meter=True): """Return a query dictionary based on the settings in the filter. :param sample_filter: SampleFilter instance :param require_meter: If true and the filter does not have a meter, raise an error. """ return make_query(sample_filter.user, sample_filter.project, sample_filter.meter, sample_filter.resource, sample_filter.source, sample_filter.start, sample_filter.start_timestamp_op, sample_filter.end, sample_filter.end_timestamp_op, require_meter) def _make_rowkey_scan(meter, rts_start=None, rts_end=None): """If it's meter filter without start and end, start_row = meter while end_row = meter + MAX_BYTE """ if not rts_start: rts_start = chr(127) end_row = "%s_%s" % (meter, rts_start) start_row = "%s_%s" % (meter, rts_end) return start_row, end_row def _load_hbase_list(d, prefix): """Deserialise dict stored as HBase column family """ ret = [] prefix = 'f:%s_' % prefix for key in (k for k in d if k.startswith(prefix)): ret.append(key[len(prefix):]) return ret def _format_meter_reference(counter_name, counter_type, counter_unit): """Format reference to meter data. """ return "%s!%s!%s" % (counter_name, counter_type, counter_unit) def _metadata_from_document(doc): """Extract resource metadata from HBase document using prefix specific to HBase implementation. """ return dict( (k[4:], v) for k, v in doc.iteritems() if k.startswith('f:r_')) def _timestamp_from_record_tuple(record): """Extract timestamp from HBase tuple record """ return timeutils.parse_strtime(record[1]['f:timestamp']) def _resource_id_from_record_tuple(record): """Extract resource_id from HBase tuple record """ return record[1]['f:resource_id']