%PDF- %PDF-
Direktori : /opt/cloudlinux/venv/lib/python3.11/site-packages/lvestats/lib/ |
Current File : //opt/cloudlinux/venv/lib/python3.11/site-packages/lvestats/lib/cloudlinux_statistics.py |
# coding=utf-8 # # Copyright © Cloud Linux GmbH & Cloud Linux Software, Inc 2010-2019 All Rights Reserved # # Licensed under CLOUD LINUX LICENSE AGREEMENT # http://cloudlinux.com/docs/LICENSE.TXT import json import logging import math import os from collections import defaultdict from datetime import datetime # NOQA from typing import Dict, List, Optional, Tuple, Union # NOQA import sqlalchemy # NOQA import lvestats.lib.commons.decorators from clcommon import cpapi from clcommon.cpapi.pluginlib import getuser from clcommon.lib import GovernorStatus, MySQLGovernor, MySQLGovException from lvestats.lib import config from lvestats.lib.commons import dateutil, func from lvestats.lib.commons.func import deserialize_lve_id, get_users_for_reseller from lvestats.lib.commons.logsetup import setup_logging from lvestats.lib.commons.users_manager import g_users_manager from lvestats.lib.dbengine import make_db_engine from lvestats.lib.info.lveinfomain import main_ from lvestats.lib.lve_list import LVEList from lvestats.lib.lveinfolib import HistoryShowUnion from lvestats.lib.lveinfolib_gov import HistoryShowDBGov from lvestats.lib.parsers.cloudlinux_statistics_parser import setup_statistics_parser from lvestats.lib.uidconverter import ( username_to_uid_local, reseller_to_uid, uid_to_username, ) NOT_AVAILABLE = 'N/A' class CloudLinuxStatistics(object): def __init__( self, engine, # type: sqlalchemy.engine.base.Engine show_columns, # type: List[str] order_by='', # type: str server_id=None, # type: Optional[str] by_usage=None, # type: Optional[str] by_usage_percentage=0.9, # type: float by_fault=None, # type: Optional[str] threshold=1, # type: int limit=0, # type: int uid=None, # type: Optional[Union[int, List[int], Tuple[int]]] time_unit=60 # type: int ): """ Initializes statistics object used by SPA UI. Note, that if we set order_by to mysql_*, and set by_usage or by_fault, end result will will not be ordered by mysql_ column, using uid for ordering instead """ self.order_by_lve = None self.order_by_gov = None self.server_id = server_id self.engine = engine self.limit_gov = 0 self.limit_lve = 0 self.uid = uid self.time_unit = time_unit self.is_many_uids = self.uid is None or isinstance(self.uid, (list, tuple)) self._lve_results = [] self._gov_results = [] self._resellers_results = [] if 'all' in show_columns: show_columns = ['cpu', 'ep', 'vmem', 'pmem', 'nproc', 'io', 'iops', 'nproc', 'mysql'] self.show_columns = [s.lower() for s in show_columns] self.by_fault = self.normalize_by_fault(by_fault) self.order_by = order_by self.init_order_by(order_by, self.by_fault) self._show_columns_lve = [] self._show_columns_gov = [] self._gov_key_index = None self._lve_key_index = None self.init_show_columns() if self.order_by_gov: self.limit_gov = limit else: self.limit_lve = limit self.limit = limit self.by_usage_lve = None self.by_usage_gov = None self.init_by_usage(by_usage) self.by_usage_percentage = by_usage_percentage self.threshold = threshold self.mysql_governor = MySQLGovernor() self.log = logging.getLogger(self.__class__.__name__) self.show_mysql = self._show_mysql() self._admins_uids = self._get_admin_uids() if getuser() == 'root' else set() @staticmethod def _get_admin_uids(): try: return set(username_to_uid_local(admin) for admin in cpapi.admins()) - {None} except (cpapi.NotSupported, AttributeError): return set() def _show_mysql(self): # type: () -> bool return self.mysql_governor.is_governor_present() and 'mysql' in self.show_columns def _index_of(self, list_, key): # type: (List, Union[str,int]) -> Optional[int] if key in list_: return list_.index(key) def init_show_columns(self): show_columns_lve = [] show_columns_gov = [] show_columns_gov_singleuser = ['from', 'to'] show_columns_lve_singleuser = ['id', 'from', 'to'] columns = [] for column in self.show_columns: if column != 'mysql': columns += ['a' + column, 'l' + column, column + 'f'] if len(columns): show_columns_lve = ['id'] + columns show_columns_lve_singleuser += columns if 'mysql' in self.show_columns: show_columns_gov = ['id', 'cpu', 'lcpu', 'read', 'lread', 'write', 'lwrite'] show_columns_gov_singleuser = show_columns_gov + ['from', 'to'] self._show_columns_lve = show_columns_lve if self.is_many_uids else show_columns_lve_singleuser self._show_columns_gov = show_columns_gov if self.is_many_uids else show_columns_gov_singleuser if self.is_many_uids: self._gov_key_index = self._index_of(self._show_columns_gov, 'id') self._lve_key_index = self._index_of(self._show_columns_lve, 'id') else: self._gov_key_index = self._index_of(self._show_columns_gov, 'from') self._lve_key_index = self._index_of(self._show_columns_lve, 'from') def init_order_by(self, order_by, by_fault): # type: (str) -> None if order_by: order_by = order_by.lower() if order_by.endswith('faults'): self.order_by_lve = order_by.replace('_faults', '') + 'f' elif order_by.startswith('mysql'): self.order_by_gov = order_by[len('mysql_'):] else: self.order_by_lve = 'a' + order_by elif by_fault and by_fault[0] == 'any': self.order_by_lve = 'cpuf' else: self.order_by_lve = by_fault and by_fault[0] def get_username(self, uid): # type: (int) -> Tuple[str] username = uid_to_username(uid, self.server_id, self.server_id, self.engine) return username or NOT_AVAILABLE def _get_default_user_usage(self): """ Return default info about user usage (mysql) :rtype: dict """ usage_field = defaultdict(dict) for column in self.show_columns: if column != 'mysql': usage_field[column]['lve'] = 0.0 if self.show_mysql: usage_field['cpu']['mysql'] = 0.0 usage_field['io']['mysql'] = 0.0 return usage_field def _get_default_user_limits(self, username): """ Return default info about user limits (mysql) :rtype: dict """ limit_field = defaultdict(dict) for column in self.show_columns: if column != 'mysql': limit_field[column]['lve'] = 0.0 if self.show_mysql: cpu_limit, io_limit = self.get_user_limits_safe(username) limit_field['cpu']['mysql'] = cpu_limit limit_field['io']['mysql'] = io_limit return limit_field def _get_default_user_faults(self): fault_field = {} for column in self.show_columns: if column != 'mysql': fault_field[column] = {'lve': 0} return fault_field def get_item(self, ikey): if self.is_many_uids: _, is_reseller = deserialize_lve_id(ikey) username = self.get_username(ikey) else: _, is_reseller = deserialize_lve_id(self.uid) username = self.get_username(self.uid) item = { 'usage': self._get_default_user_usage(), 'limits': self._get_default_user_limits(username), } if self.show_columns != ['mysql']: item['faults'] = self._get_default_user_faults() if self.is_many_uids: if is_reseller: item.update({'id': ikey, 'name': username}) else: item.update({ 'id': ikey, 'username': username, 'domain': g_users_manager.get_domain(username, raise_exc=False) or NOT_AVAILABLE, 'reseller': g_users_manager.get_reseller(username, raise_exc=False) or NOT_AVAILABLE, }) else: item.update({ 'from': int(ikey), 'to': int(ikey) + self.time_unit}) return item def fill_lve_item(self, item, row): if row is None: return item count = 1 if self.is_many_uids else 1 + 2 for column in self.show_columns: if column == 'mysql': continue item['usage'][column]['lve'] = row[count] item['limits'][column]['lve'] = row[count + 1] item['faults'][column]['lve'] = row[count + 2] count += 3 return item @staticmethod def __get_io_bytes(read, write): """ Get io value by read and write values. :param int|float|None read: read speed in MEGAbytes :param int|float|None write: write speed in MEGAbytes :return int: io, BYTES """ # convert None to zero read = read or 0 write = write or 0 # calculate io in bytes io = (read + write) * 1024 * 1024 return int(io) def fill_gov_item(self, item, row): if row is None: return item item['usage']['cpu']['mysql'] = round(float(row[1] or 0), 1) item['limits']['cpu']['mysql'] = round(float(row[2] or 0), 1) item['usage']['io']['mysql'] = self.__get_io_bytes(row[3], row[5]) item['limits']['io']['mysql'] = self.__get_io_bytes(row[4], row[6]) return item def process_data(self, lve_results, gov_results, period_from, period_to): from_gov_to_lve_indexes, from_lve_to_gov_indexes,\ gov_keys, gov_results,\ lve_keys, lve_results,\ not_in_gov_indexes, not_in_lve_indexes = self.process_indexes(gov_results, lve_results) if not self.show_mysql: lve_results = list(lve_results) elif self.order_by_gov: lve_results = self.sort_by_another_list(from_gov_to_lve_indexes, lve_results, not_in_gov_indexes) elif self.order_by_lve: gov_results = self.sort_by_another_list(from_lve_to_gov_indexes, gov_results, not_in_lve_indexes) else: gov_results, lve_results = self.sort_by_key(gov_keys, gov_results, lve_keys, lve_results) lendiff = (len(lve_results) - len(gov_results)) if lendiff != 0: from_formatted = period_from.strftime('%Y-%m-%d %H:%M') to_formatted = period_to.strftime('%Y-%m-%d %H:%M') self.log.warning( 'len(lve_results) != len(gov_results), it might cause duplicated timestamps ' 'or mix of lve and governor stats from different periods in aggregated data.\n' 'Utility was called with the following parameters:\n' 'cloudlinux-statistics --json --time-unit %ss --id %s ' '--from %s --to %s ' '--show %s --server_id %s\n' 'engine=%s\norder_by=%s\nby_fault=%s\n', self.time_unit, self.uid, from_formatted, to_formatted, " ".join(self.show_columns), self.server_id, self.engine, self.order_by, self.by_fault ) self._lve_results = lve_results self._gov_results = gov_results @staticmethod def sort_by_key(gov_keys, gov_results, lve_keys, lve_results): # type: (List[int], List, List[int], List) -> Tuple[List, List] lve_pos = 0 gov_pos = 0 lve_len = len(lve_results) gov_len = len(gov_results) _lve_results = [] _gov_results = [] # maximum # of iterations will be in case when # lve_keys like [1,2,7] and gov_keys like [4,5,6] for _ in range(lve_len + gov_len): if lve_pos == lve_len: # became out of range _lve_results.extend([None] * (gov_len - gov_pos)) # if also gov_pos == gov_len slice returns [] without error _gov_results.extend(gov_results[gov_pos:]) break if gov_pos == gov_len: _gov_results.extend([None] * (lve_len - lve_pos)) _lve_results.extend(lve_results[lve_pos:]) break if lve_keys[lve_pos] == gov_keys[gov_pos]: _lve_results.append(lve_results[lve_pos]) _gov_results.append(gov_results[gov_pos]) lve_pos += 1 gov_pos += 1 elif lve_keys[lve_pos] > gov_keys[gov_pos]: _lve_results.append(None) _gov_results.append(gov_results[gov_pos]) gov_pos += 1 else: _gov_results.append(None) _lve_results.append(lve_results[lve_pos]) lve_pos += 1 return _gov_results, _lve_results @staticmethod def sort_by_another_list(indexes, results, not_in_indexes): # type: (List[int], List, List[int]) -> List _results = [] for index in indexes: if index is None: _results.append(None) else: _results.append(results[index]) for index in not_in_indexes: _results.append(results[index]) results = _results return results def process_indexes(self, gov_results, lve_results): # type: (List, List) -> Tuple[List[int],List[int],List,List,List,List,List,List] if self.by_usage_gov: gov_uids = [_[0] for _ in gov_results] lve_results = [row for row in lve_results if row[0] in gov_uids] if self.by_usage_lve or self.by_fault: lve_uids = [_[0] for _ in lve_results] gov_results = [row for row in gov_results if row[0] in lve_uids] gov_keys = [_[self._gov_key_index] for _ in gov_results] lve_keys = [_[self._lve_key_index] for _ in lve_results] from_gov_to_lve_indexes = self.get_indexes(gov_keys, lve_keys) not_in_lve_indexes = [i for i, _ in enumerate(from_gov_to_lve_indexes) if _ is None] from_lve_to_gov_indexes = self.get_indexes(lve_keys, gov_keys) not_in_gov_indexes = [i for i, _ in enumerate(from_lve_to_gov_indexes) if _ is None] return from_gov_to_lve_indexes, from_lve_to_gov_indexes,\ gov_keys, gov_results,\ lve_keys, lve_results,\ not_in_gov_indexes, not_in_lve_indexes def get_indexes(self, first_uids, second_uids): # type: (List[int], List[int]) -> List[int] return [self._index_of(second_uids, uid) for uid in first_uids] def get_value(self, key, resellers): # type: (int, bool) -> Dict[str, Union[int,float]] """ :type resellers: bool :type key: int :rtype: dict """ if resellers: item = self.get_item(self._resellers_results[key][self._lve_key_index]) return self.fill_lve_item(item, self._resellers_results[key]) elif not self.limit or key < self.limit: if not self.show_mysql: item = self.get_item(self._lve_results[key][self._lve_key_index]) return self.fill_lve_item(item, self._lve_results[key]) elif self.order_by_gov or self.limit_gov: row = self._gov_results[key] if row is not None: item = self.get_item(row[self._gov_key_index]) self.fill_gov_item(item, row) else: item = self.get_item(self._lve_results[key][self._lve_key_index]) return self.fill_lve_item(item, self._lve_results[key]) else: row = self._lve_results[key] if row is not None: item = self.get_item(row[self._lve_key_index]) self.fill_lve_item(item, row) else: item = self.get_item(self._gov_results[key][self._gov_key_index]) return self.fill_gov_item(item, self._gov_results[key]) else: raise IndexError def get_gov_row(self, key, indexes, result): if key > len(indexes) + 1 or indexes[key] is None: return None return result[indexes[key]] def get_user_limits_safe(self, username, ignore_watched=False): # type: (str, bool) -> Tuple[int,int] """ Get cpu and io limit for user :return tuple: cpu limit in %, io limit in bytes/s """ cpu_limit, io_limit = 0, 0 if self.mysql_governor.is_governor_present(): try: if ignore_watched or self.mysql_governor.get_governor_status_by_username(username) == 'watched': cpu_limit, io_limit = self.mysql_governor.get_limits_by_user(username) except MySQLGovException as e: self.log.info('Cannot get mysql limits for user %s. Error raised: %s', username, e.message % e.context) return cpu_limit, io_limit * 1024 @staticmethod def normalize_by_fault(by_fault): # type: (Optional[str]) -> Optional[Tuple[str]] if by_fault: if by_fault == 'mem': by_fault = 'vmem' return by_fault + 'f', # it is a tuple, so we need coma at the end def init_by_usage(self, by_usage): # type: (Optional[str]) -> None if by_usage: if by_usage.startswith("mysql"): self.by_usage_gov = by_usage[len('mysql_'):] else: self.by_usage_lve = 'a' + by_usage, # it is a tuple, so we need coma at the end def generate_statistics(self, period_from, period_to): # type: (datetime, datetime) -> None """Generate statistics.""" period_from_gm = dateutil.local_to_gm(period_from) period_to_gm = dateutil.local_to_gm(period_to) if self._show_columns_lve: history_show = HistoryShowUnion( dbengine=self.engine, period_from=period_from_gm, period_to=period_to_gm, uid=self.uid, time_unit=self.time_unit, show_columns=self._show_columns_lve, server_id=self.server_id, order_by=self.order_by_lve, by_usage=self.by_usage_lve, by_usage_percentage=self.by_usage_percentage, by_fault=self.by_fault, threshold=self.threshold, limit=self.limit_lve) history_show.set_normalised_output() lve_results = history_show.proceed() else: lve_results = [] if self.show_mysql: gov_results = HistoryShowDBGov( dbengine=self.engine, period_from=period_from_gm, period_to=period_to_gm, time_unit=self.time_unit, server_id=self.server_id, show_columns=self._show_columns_gov, order_by=self.order_by_gov, by_usage=self.by_usage_gov, by_usage_percentage=self.by_usage_percentage, limit=self.limit_gov, reverse=True, uid=self.uid, ).history_dbgov_show() else: gov_results = [] if self.is_many_uids: users = [] for item in lve_results: uid, is_reseller = deserialize_lve_id(item[self._lve_key_index]) if uid in self._admins_uids: continue if is_reseller: self._resellers_results.append(item) else: users.append(item) lve_results = users self.process_data(lve_results, gov_results, period_from, period_to) def _get_uid_for_select(for_reseller, uid): # type: (str, int) -> Union[List[int], int] """ Get uid that should be selected from history table; """ if getuser() != 'root': return os.getuid() g_users_manager.build_users_cache(for_reseller) if for_reseller: users_list = get_users_for_reseller(for_reseller or getuser()) id_ = list(filter(bool, list(map(username_to_uid_local, users_list)))) reseller_uid = reseller_to_uid(for_reseller or getuser()) # LVESTATS-97 # If uid not in users_list, we return empty list if uid: return uid if uid in id_ or uid == reseller_uid else [] # select also information about reseller's containers (if exists) if reseller_uid is not None: id_.append(reseller_uid) else: id_ = uid return id_ def execute(engine, options, log=None): # type: (sqlalchemy.engine.base.Engine, object, Optional[logging.Logger]) -> Tuple[str, int] """ Get command response and exitcode. :type log: None | logging.Logger :type engine: sqlalchemy.engine.base.Engine :rtype: (str, int) """ if not options.json: return "Only JSON mode supported for now", -1 id_ = _get_uid_for_select(options.for_reseller, options.id) statistics = get_statistics(engine, id_, log, options) json_data = dump_json(statistics) return json_data, 0 def dump_json(statistics): # type: (Dict) -> str json_data = json.dumps(statistics) return json_data def get_statistics( engine, # type: sqlalchemy.engine.base.Engine id_, # type: Optional[Union[int, List[int], Tuple[int]]] log, # type: Optional[logging.Logger] options, timestamp=dateutil.gm_datetime_to_unixtimestamp() # type: int ): # type: (...) -> Dict[str, Union[LVEList, str, int]] result = { 'result': 'success', 'timestamp': timestamp } # LVESTATS-97 # If both params (for_reseller and id) are used # and id_ list is empty, we return `Permission denied` message if options.for_reseller is not None and options.id is not None and id_ == []: error_msg = ( 'Permission denied. User with id ' f'{options.id} does not belong reseller `{options.for_reseller}`' ) result['result'] = error_msg return result try: stat = CloudLinuxStatistics( engine, show_columns=options.show, order_by=options.order_by, server_id=options.server_id, by_usage=options.by_usage, by_usage_percentage=options.percentage / 100., by_fault=options.by_fault, threshold=options.threshold, limit=options.limit, uid=id_, time_unit=options.time_unit) stat.generate_statistics(getattr(options, 'from'), options.to) lvelist = LVEList(stat) lvelist_resellers = LVEList(stat, resellers=True) if stat.is_many_uids: result['users'] = lvelist result['resellers'] = lvelist_resellers else: result['user'] = lvelist except Exception as e: result["result"] = str(e) if log: log.error(str(e)) # add mysql status info status, error = func.get_governor_status() result['mySqlGov'] = status if status == GovernorStatus.ERROR: result['warning'] = error.message result['context'] = error.context return result @lvestats.lib.commons.decorators.no_sigpipe def main(engine, argv_, server_id='localhost', log=None): # type: (sqlalchemy.engine.base.Engine, List[str], str, Optional[logging.Logger]) -> int """ Execute command and return exitcode """ if log is None: log = setup_logging( {}, caller_name="CloudLinuxStatistics", file_level=logging.WARNING, console_level=logging.FATAL, ) options = setup_statistics_parser(argv_, server_id) json_str, exit_code = execute(engine, options, log=log) print(json_str) return exit_code def get_users_and_resellers_with_faults(period='1d'): """ Return amount of users and resellers with faults """ cnf = config.read_config() dbengine = make_db_engine(cnf) server_id = cnf.get('server_id', 'localhost') opt_list = ['--by-fault', 'any', '--json', '--period=' + period] options = setup_statistics_parser(opt_list, server_id) json_str, _ = execute(dbengine, options, log=None) json_data = json.loads(json_str) resellers = json_data['resellers'] users = json_data['users'] return len(users), len(resellers) def get_max_memory_for_lve(lve_id, period_seconds): data = main_( config.read_config(), argv_=['--json', '--id', str(lve_id), '--show-columns', 'mPMem', '--period', str(int(math.ceil(period_seconds / 60))) + 'm'], ) if data: parsed_data = json.loads(data) try: max_pmem = max(x['mPMem'] for x in parsed_data['data']) except (ValueError, KeyError): max_pmem = None return max_pmem