403Webshell
Server IP : 66.29.132.122  /  Your IP : 3.141.19.165
Web Server : LiteSpeed
System : Linux business142.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64
User : admazpex ( 531)
PHP Version : 7.2.34
Disable Function : NONE
MySQL : OFF  |  cURL : ON  |  WGET : ON  |  Perl : ON  |  Python : ON  |  Sudo : OFF  |  Pkexec : OFF
Directory :  /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/proc/self/root/proc/thread-self/root/proc/self/root/opt/cloudlinux/venv/lib64/python3.11/site-packages/lvestats/lib/

Upload File :
current_dir [ Writeable ] document_root [ Writeable ]

 

Command :


[ Back ]     

Current File : /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/proc/self/root/proc/thread-self/root/proc/self/root/opt/cloudlinux/venv/lib64/python3.11/site-packages/lvestats/lib/lveinfolib.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

# pylint: disable=too-many-lines

import copy
import datetime
import logging
import pwd
import time
import warnings

from sqlalchemy import Float, Integer, text
from sqlalchemy import exc as sa_exc
from sqlalchemy.sql import and_, asc, desc, label, not_, or_, select
from sqlalchemy.sql.expression import ColumnElement, alias, case, cast, func

from lvestats.core.plugin import LveStatsPluginTerminated
from lvestats.lib.commons.dateutil import (
    gm_datetime_to_unixtimestamp,
    gm_to_local,
    local_to_gm,
    round_1m,
    str_to_timedelta,
    unixtimestamp_to_gm_datetime,
)
from lvestats.lib.commons.func import (
    get_current_max_lve_id,
    skip_user_by_maxuid,
)
from lvestats.lib.commons.sizeutil import convert_bytes, convert_powers_of_1000, mempages_to_bytes
from lvestats.lib.config import (
    HIDE_MAX_UID_LVE_PARAMETER,
    read_config,
    str_to_bool,
)
from lvestats.orm import LVE_STATS_2_TABLENAME_PREFIX, history, history_x60, servers

__all__ = ('HistoryShow', 'HistoryShowUnion', 'OutputFormatter', 'get_lve_version')


def get_lve_version(dbengine, server_id):
    sql_query_lve_version = select([servers.lve_version]).where(servers.server_id == server_id)
    with dbengine.begin() as transaction_:
        cursor_ = transaction_.execute(sql_query_lve_version)
        if cursor_.returns_rows:
            res = cursor_.fetchone()
            if res is None:
                return 6
            else:
                return int(res['lve_version'])
        else:
            return 6


def servers_info(dbengine):
    sql_query_lve_version = select([servers.server_id, servers.lve_version])
    return dbengine.execute(sql_query_lve_version)


def convert_key_to_label(key_):
    return str.lower(key_)


def add_labes_to_column(func_dict):
    """
    Add label (on SQL add ".. as ..") to aggregate_func_dict
    key of dict convert to lowercase and add as label
    :param dict func_dict:
    :return dict:
    """
    func_dict_labeled = {}
    for key_, sqlalchemy_func in list(func_dict.items()):
        if issubclass(sqlalchemy_func.__class__, ColumnElement):
            func_labeled = label(convert_key_to_label(key_), sqlalchemy_func)
        else:
            func_labeled = sqlalchemy_func
        func_dict_labeled[key_] = func_labeled
    return func_dict_labeled


def convert_to_list(arg):
    if isinstance(arg, (str, int)):
        return [arg]
    return arg


class OutputFormatter(object):
    def __init__(self, fields, rows=None, orders=None):
        self.rows = rows or []
        self.fields = list(fields)
        self._fields_lowered = [_.lower() for _ in fields]  # self.fields use for output
        self._orders = []
        self._hidden_fields = set()  # self._fields indexes list fields  to hide

        if orders:
            for field, order in orders:
                self.add_order(field, order)

    def get_fields(self):
        if self._hidden_fields:
            return [f_ for f_ in self.fields if f_ not in self._hidden_fields]
        else:
            return self.fields

    def set_rows(self, rows):
        """
        Use this method if you used hide_fields
        number items in row must bee same as in fields after hide
        """
        self.rows = rows

    def hide_fields(self, h_fields):
        """
        :param tuple|list h_fields:
        :return:
        """
        self._hidden_fields = self._hidden_fields.union(set(h_fields))

    def add_order(self, fields, order):
        """
        :param list|tuples fields:
        :param order: field to use to order result
        :return:
        """
        if isinstance(order, str):
            try:
                order = getattr(self, order)
            except AttributeError as e:
                raise ValueError(f'Non such {order} order') from e
            if not hasattr(order, '__call__'):
                raise ValueError(f'input object {order} must be callable')
        self._orders.append(({_.lower() for _ in fields}, order))

    def _convert_line(self, row_):
        """
        :param iterable row_:
        :return:
        """
        row_ = list(row_)
        row_out = []
        for field_, r_ in zip(self._fields_lowered, row_):
            if field_ in self._hidden_fields:  # continue if field must be hide
                continue
            # use many orders to one cell
            for order_fields, order in self._orders:
                if field_ in order_fields:
                    try:
                        r_ = order(r_)
                    except (TypeError, ValueError, KeyError, IndexError):
                        pass
            row_out.append(r_)
        return row_out

    def __iter__(self):
        for row_ in self.rows:
            yield self._convert_line(row_)

    def __getitem__(self, index):
        if isinstance(index, slice):
            return list(map(self._convert_line, self.rows[index.start: index.stop]))
        return self._convert_line(self.rows[index])

    def __len__(self):
        return len(self.rows)

    def get_corrected_list(self):
        return list(self)

    # build-in orders
    @staticmethod
    def strftime(value, format_='%m-%d %H:%M'):
        return value.strftime(format_)

    @staticmethod
    def percentage(value):
        if value is None:
            return '-'
        try:
            float(value)
        except ValueError:
            return value
        return str(value * 100) + '%'

    @staticmethod
    def bytes(value):
        if value is None:
            return '-'
        return convert_bytes(value)

    @staticmethod
    def powers_of_1000(value):
        if value is None:
            return '-'
        return convert_powers_of_1000(value)

    @staticmethod
    def username(value):
        try:
            return pwd.getpwuid(int(value)).pw_name
        except KeyError:
            return value

    @staticmethod
    def datetime(value):
        """
        Convert unix timestamp to datetime (local timezone)
        """
        return datetime.datetime.fromtimestamp(value)


def enumerate_duplicate_columns(columns):
    """
    Enumerate if columns name or functions are duplicated
    Use for force add duplicate columns to select operator
    :param list|tuple columns:
    :return list|tuple: enumerated columns
    """
    output = []
    output_str = []  # for controlling columns count
    for c_ in columns:
        c_str = str(c_)
        dubl_count = output_str.count(c_str)
        if dubl_count >= 1:  # check whether there is a duplicate; "c_ in columns" for sqlalchemy classes not work
            # check if column is string or function and get column name
            if isinstance(c_, str):
                c_name = c_
            else:
                c_name = c_.name

            # numbering duplicate columns
            c_name += '_' + str(dubl_count + 1)

            c_ = label(c_name, c_)  # rename column
        output.append(c_)
        output_str.append(c_str)
    return output


usage_to_limit_dict = {
    'aCPU'.lower(): 'lCPU'.lower(),
    'mCPU'.lower(): 'lCPU'.lower(),
    'aVMem'.lower(): 'lVMem'.lower(),
    'mVMem'.lower(): 'lVMem'.lower(),
    'aEP'.lower(): 'lEP'.lower(),
    'mEP'.lower(): 'lEP'.lower(),
    'aPMem'.lower(): 'lPMem'.lower(),
    'mPMem'.lower(): 'lPMem'.lower(),
    'aNproc'.lower(): 'lNproc'.lower(),
    'mNproc'.lower(): 'lNproc'.lower(),
    'aIO'.lower(): 'lIO'.lower(),
    'mIO'.lower(): 'lIO'.lower(),
    'aIOPS'.lower(): 'lIOPS'.lower(),
    'mIOPS'.lower(): 'lIOPS'.lower(),
}

FIELD_TO_TABLE_COLUMN = {
    'ID': 'id',
    'aCPU': 'cpu',
    'aVMem': 'mem',
    'aEP': 'mep',
    'aPMem': 'memphy',
    'aIO': 'io',
    'aNproc': 'nproc',
    'aIOPS': 'iops',
    'lCPU': 'cpu_limit',
    'lEP': 'mep_limit',
    'lVMem': 'mem_limit',
    'lPMem': 'lmemphy',
    'lIO': 'io_limit',
    'lNproc': 'lnproc',
    'lIOPS': 'liops',
    'VMemF': 'mem_fault',
    'PMemF': 'memphy_fault',
    'EPf': 'mep_fault',
    'NprocF': 'nproc_fault',
    'CPUf': 'cpu_fault',
    'IOf': 'io_fault',
    'IOPSf': 'iops_fault',
    'uCPU': 'cpu',
    'uEP': 'mep',
    'uVMem': 'mem',
    'uPMem': 'memphy',
    'uIO': 'io',
    'uNproc': 'nproc',
    'uIOPS': 'iops',
    'mCPU': 'cpu',
    'mEP': 'mep',
    'mVMem': 'mem',
    'mPMem': 'memphy',
    'mNproc': 'nproc',
    'mIO': 'io',
    'mIOPS': 'iops',
}

FIELD_AVERAGE = ['aCPU', 'aVMem', 'aPMem', 'aEP', 'aNproc', 'aIO', 'aIOPS']
FIELD_LIMIT = ['lCPU', 'lVMem', 'lPMem', 'lEP', 'lNproc', 'lIO', 'lIOPS']
FIELD_FAULT = ['CPUf', 'VMemF', 'PMemF', 'EPf', 'NprocF', 'IOf', 'IOPSf']
FIELD_USAGE = ['uCPU', 'uVMem', 'uPMem', 'uEP', 'uNproc', 'uIO', 'uIOPS']
FIELD_MAX = ['mCPU', 'mVMem', 'mPMem', 'mEP', 'mNproc', 'mIO', 'mIOPS']

KEYS_NORMALIZATION_LOOKUP_TABLE = {'FROM': 'From', 'TO': 'To', 'ANYF': 'anyF'}
for key in list(FIELD_TO_TABLE_COLUMN.keys()):
    KEYS_NORMALIZATION_LOOKUP_TABLE[key.upper()] = key


def normalize_optional_column_names(names):
    if names is None:
        return None
    return normalize_column_names(names)


def normalize_column_names(names):
    result = []
    for name in names:
        result.append(normalize_column_name(name))
    return result


def normalize_column_name(name):
    if name:
        return KEYS_NORMALIZATION_LOOKUP_TABLE[name.upper()]
    return None


time_unit_orders = [
    ('10m', lambda dt: dt - datetime.timedelta(minutes=dt.minute % 10 + 10)),
    ('1h', lambda dt: dt.replace(minute=0) if dt.minute else dt - datetime.timedelta(hours=1)),  # round to hour
    ('1d', lambda dt: dt.replace(hour=0, minute=0)),  # round to day
]


def dyn_time_unit_groups(period_from, period_to):
    period_groups = [round_1m(period_from), round_1m(period_to + datetime.timedelta(minutes=1))]
    time_unit_list = ['1m']
    _from_order_fun = round_1m
    for time_unit, order_fun in time_unit_orders:
        from_to_point = order_fun(period_groups[1])
        if from_to_point - period_groups[0] >= str_to_timedelta(time_unit):
            _from_order_fun = order_fun
            period_groups.insert(1, from_to_point)
            time_unit_list.append(time_unit)
    period_groups[0] = _from_order_fun(period_groups[0])
    # prepare to output as list of tuples
    # (<From|datetime>, <To|datetime>, <time-unit|int>)
    from_to_groups = []
    for index, time_unit in enumerate(time_unit_list):
        from_to_groups.append(
            (period_groups[-index - 2], period_groups[-index - 1], int(str_to_timedelta(time_unit).total_seconds()))
        )
    return from_to_groups


class HistoryShow(object):
    def __init__(
        self,
        dbengine,
        period_from,
        period_to,
        uid=None,
        show_columns=None,
        server_id='localhost',
        time_unit=None,
        order_by=None,
        by_usage=None,
        by_usage_percentage=0.9,
        by_fault=None,
        threshold=1,
        limit=0,
        table=None,
        log=None,
        time_count=None,
        show_idle=False,
    ):
        """
        Show different statistics from history table
        :param sqlalchemy.engine.base.Engine dbengine: database engine to use
        :param datetime.datetime|float|int period_from:  start time retrieve data
        :param datetime.datetime|float|int period_to:    end time retrieve data
        :param int|None|list|tuple uid:                  filter the output information to the user uid
        :param tuple|list show_columns:        display columns in the order specified. If not, show all supported
                valid column names: 'aCPU', 'lPMem', 'uIO', 'uEP', 'lEP', 'aVMem', 'PMemF', 'lVMem', 'NprocF', 'anyF',
                'aNproc', 'VMemF', 'ID', 'lCPU', 'aIOPS', 'aEP', 'aPMem', 'uPMem', 'lIO', 'lIOPS', 'uCPU',
                'lNproc', 'aIO', 'uIOPS', 'EPf', 'uVMem', 'uNproc'

        :param str server_id:                  filtering the output for "server id"
        :param int time_unit:                  grouping output over an interval of time (in seconds)
        :param str|None order_by:              sorting output by column name (supported by columns)
        :param str|tuple|list by_usage:        filtering are grouped data for the percentage of the use of resources
        :param float by_usage_percentage:      percent for the parameter setting 'by_usage'
        :param tuple|list|None by_fault:       filtering data are grouped for quantity faults
                                               (None if it is not filtered)
            valid names: 'aCPU', 'lPMem', 'uIO', 'uEP', 'lEP', 'aVMem', 'PMemF', 'lVMem', 'NprocF', 'anyF', 'aNproc',
            'VMemF', 'ID', 'lCPU', 'aIOPS', 'aEP', 'aPMem', 'uPMem', 'lIO', 'lIOPS', 'uCPU', 'lNproc', 'aIO', 'uIOPS',
            'EPf', 'uVMem', 'uNproc'
        :param threshold:                      number faults for filtering the data are grouped
                                               (used together with by_fault)
        :param int|None limit:                 limit on the number of output data
                                               (if 0 or None, then the limit is not set)
        :return generator:                     returns a list/generator of data with the order set out
                                               in the 'show_columns'
        """
        self.dbengine = dbengine
        self.uid = uid
        self._is_multi_uids = not isinstance(uid, int)
        if show_columns is None:
            show_columns = get_supported_columns(lve_version=get_lve_version(dbengine=dbengine, server_id=server_id))
            show_columns.insert(0, 'ID')
        self.show_columns = normalize_column_names(show_columns)
        self.server_id = server_id
        self.time_unit = time_unit
        self.by_fault = normalize_optional_column_names(by_fault)
        if order_by:
            self.order_by = normalize_column_name(order_by)
        elif self.by_fault and normalize_column_name('anyF') in self.by_fault:
            self.order_by = normalize_column_name('CPUf')
        else:
            self.order_by = self.by_fault and self.by_fault[0]
        self.by_usage = normalize_optional_column_names(by_usage)
        self.by_usage_percentage = by_usage_percentage
        self.threshold = threshold
        self.limit = limit
        self.log = log or logging.getLogger('SQL')

        self.table = table if table is not None else history.__table__  # "or" not supported in this
        self._table_alive = alias(self.table, 'alive')  # alias of main table to using in self join
        self.period_from = (
            period_from if isinstance(period_from, (int, float)) else gm_datetime_to_unixtimestamp(period_from)
        )
        self.period_to = period_to if isinstance(period_to, (int, float)) else gm_datetime_to_unixtimestamp(period_to)
        self.time_count = time_count or self.get_time_count()
        # correct cpu/100
        # we still have to round it, as it seems <[27.333/100] = [0.27332999999999996] - who knows why :(
        self.result_corrector = OutputFormatter(
            fields=self.show_columns,
            orders=[
                [FIELD_MAX + FIELD_AVERAGE + FIELD_LIMIT + FIELD_USAGE, lambda x: round(x, 3)],
                [['aCPU', 'lCPU', 'mCPU'], lambda item: round(float(item) / 100.0, 5)],
            ],
        )
        self.hide_maxuid_lve = str_to_bool(read_config().get(HIDE_MAX_UID_LVE_PARAMETER, 'true'))

    def set_normalised_output(self):
        # correct data obtained from database
        # round EP IOPS Nproc output
        self.result_corrector.add_order(
            fields=['aEP', 'mEP', 'lEP', 'aNproc', 'mNproc', 'lNproc', 'aIOPS', 'mIOPS', 'lIOPS'],
            order=lambda x: int(round(x)),
        )
        if self.dbengine.url.drivername != "sqlite":
            self.result_corrector.add_order(
                fields=[
                    'aVMem',
                    'mVMem',
                    'lVMem',
                    'aPMem',
                    'mPMem',
                    'lPMem',
                    'aIO',
                    'mIO',
                    'lIO',
                    'uCPU',
                    'uEP',
                    'uVMem',
                    'uPMem',
                    'uIO',
                    'uNproc',
                    'uIOPS',
                ],
                order=float,
            )
            self.result_corrector.add_order(
                fields=['EPf', 'VMemF', 'CPUf', 'PMemF', 'NprocF', 'IOf', 'IOPSf'], order=int
            )

        # convert Mem to bytes
        self.result_corrector.add_order(
            fields=['aVMem', 'mVMem', 'lVMem', 'aPMem', 'mPMem', 'lPMem'], order=mempages_to_bytes
        )

    def _where_time_period(self, table=None):
        """
        Generate WHERE created BETWEEN xxxxxxxxx AND yyyyyyyy
        :return:
        """
        # filtering condition by time period
        if table is None:
            table = self.table
        return table.c.created.between(self.period_from, self.period_to)

    def _where_server_id(self):
        """
        Generate WHERE server_id = 'server_name'
        :return:
        """
        return self.table.c.server_id == self.server_id

    def _where_uid(self, uid=-1, table=None):
        """Generate WHERE id = 'user_uid'"""
        if table is None:
            table = self.table
        if uid == -1:
            uid = self.uid

        if uid is None:
            if self.hide_maxuid_lve:
                # skip ids in range(MAX_UID, MAX_LVE_ID), because ids > MAX_LVE_ID may contain info
                # about reseller`s limits
                return and_(
                    table.c.id > 0, or_(not_(skip_user_by_maxuid(table.c.id)), table.c.id > get_current_max_lve_id())
                )
            return table.c.id > 0

        elif isinstance(uid, (list, tuple)):
            if self.dbengine.url.drivername == 'sqlite':
                # little workaround for sqlite's limit of 999 variables
                # let's compile query manually
                return text(table.c.id.in_(list(uid)).expression.compile(compile_kwargs={"literal_binds": True}).string)
            else:
                # mysql or postgresql do not have such limit
                return table.c.id.in_(list(uid))
        else:
            return table.c.id == uid

    def get_time_count(self):
        """
        SELECT count(*) FROM lve_stats2_history WHERE id = 0 AND created BETWEN xxxx AND yyyy server_id = 'localhost'
        """
        where = and_(
            history.created.between(self.period_from, self.period_to),
            history.id == 0,
            history.server_id == self.server_id,
        )
        query = select([text('count(*)')]).where(where)
        time_start = time.time()
        q = str(query.compile(compile_kwargs={"literal_binds": True})).replace('\n', ' ')
        self.log.debug(q)
        data = self.dbengine.execute(query)
        self.log.debug('query time: %s', time.time() - time_start)
        return data.fetchall()[0][0]

    def _fun_avg(self, item):
        """
        Generate aggregate function for calculate average
        for example sum(lve_stats2_history.cpu) / 60
        :param item:
        :return:
        """
        return cast(func.sum(item) / cast(self._fun_time_count(), Float), Float)

    def _fun_limit(self, item):
        """
        Generate aggregate function for calculate limit
        if column_limit is zero its mean no limit
        for example CASE WHEN (min(lve_stats2_history.cpu_limit) > 0) THEN max(lve_stats2_history.cpu_limit) ELSE 0 END
        :param item:
        :return:
        """
        return case([(func.min(item) > 0, func.max(item))], else_=0)

    def _fun_fault(self, item):
        """
        Generate aggregate function for calculate fault
        for example sum(lve_stats2_history.cpu_fault)
        :param item:
        :return:
        """
        return func.sum(item)

    def _fun_usage(self, item, item_limit):
        """
        Generate aggregate function for calculate resource usage equivalent average/limit
        for example
            CASE
                WHEN (
                    CASE
                        WHEN min(lve_stats2_history.cpu_limit) > 0 THEN max(lve_stats2_history.cpu_limit)
                        ELSE 0
                    END IS NULL
                ) THEN NULL
                WHEN (
                    CASE
                        WHEN min(lve_stats2_history.cpu_limit) > 0 THEN max(lve_stats2_history.cpu_limit)
                        ELSE 0
                    END > 0
                ) THEN (
                    sum(lve_stats2_history.cpu) / 1422
                ) / CASE
                        WHEN min(lve_stats2_history.cpu_limit) > 0 THEN max(lve_stats2_history.cpu_limit)
                        ELSE 0
                    END
            END
        :param item:
        :return:
        """
        # noinspection PyComparisonWithNone
        return case(
            [
                # Don't use "is None" here.
                (self._fun_limit(item_limit) == None, None),  # NOQA
                (self._fun_limit(item_limit) > 0, self._fun_avg(item) / self._fun_limit(item_limit)),
            ],
            else_=None,
        )

    def _fun_max(self, item, item_limit, item_fault):
        """
        Generate aggregate function for calculate maximum resource usage; for backward capability with lve-stats 0.x
        for example:
            CASE
                WHEN
                    (sum(lve_stats2_history.cpu_fault) > 0)
                THEN
                    max(lve_stats2_history.cpu_limit)
                ELSE
                    max(lve_stats2_history.cpu) END
        :param item:
        :param item_limit:
        :param item_fault:
        :return:
        """
        return case([(func.sum(item_fault) > 0, func.max(item_limit))], else_=func.max(item))

    def _fun_time_count(self):
        if self._check_need_join():
            return text('count(*)')
        else:
            return self.time_count

    def _fun_time_from(self):
        if self._is_multi_uids:
            return self.period_from
        else:
            if self._check_need_join():
                _table = self._table_alive
            else:
                _table = self.table
            if self.dbengine.url.drivername == 'sqlite':
                # cast(..., Integer) using for compatibility with lve-stats-2.1-8 database; 'created' saved as float
                return (
                    cast((_table.c.created - self.period_from) / self.time_unit, Integer) * self.time_unit
                    + self.period_from
                )
            else:
                return (
                    func.floor((_table.c.created - self.period_from) / self.time_unit) * self.time_unit
                    + self.period_from
                )

    def _fun_time_to(self):
        # in case of changes here don't forget to check _group_by_query
        if self._is_multi_uids:
            return self.period_to
        else:
            return self._fun_time_from() + self.time_unit

    def _fun_user_id(self):
        if self._is_multi_uids:
            return label('ID', self.table.c.id)
        else:
            return label('ID', text(str(self.uid)))

    def _aggregate_fun_case(self, item):
        """
        Function for obtain aggregate function (or column name) by column name
        :param item: 'aCPU', 'aVMem', 'aPMem', 'aEP', 'aNproc',  'aIO', 'aIOPS', 'lCPU', 'lVMem', 'lPMem', 'lEP',
                     'lNproc',  'lIO', 'lIOPS', 'CPUf', 'VMemF', 'PMemF', 'EPf', 'NprocF',  'IOf', 'IOPSf', 'uCPU',
                     'uVMem', 'uPMem', 'uEP', 'uNproc',  'uIO', 'uIOPS', 'mCPU', 'mVMem', 'mPMem', 'mEP',  'mNproc',
                     'mIO', 'mIOPS', 'anyF', 'ID', 'From', 'To'
        :type item: str
        :return:
        """
        if item == 'anyF':
            fun_ = func.sum(
                self.table.c.mem_fault
                + self.table.c.memphy_fault
                + self.table.c.mep_fault
                + self.table.c.nproc_fault
                + self.table.c.cpu_fault
                + self.table.c.io_fault
                + self.table.c.iops_fault
            )
        elif item == 'ID':
            fun_ = self._fun_user_id()
        elif item == 'From':
            fun_ = self._fun_time_from()
        elif item == 'To':
            fun_ = self._fun_time_to()
        else:
            column_name = FIELD_TO_TABLE_COLUMN[item]
            table_column = getattr(self.table.c, column_name)
            if item in FIELD_AVERAGE:
                fun_ = self._fun_avg(table_column)
            elif item in FIELD_LIMIT:
                fun_ = self._fun_limit(table_column)
            elif item in FIELD_FAULT:
                fun_ = self._fun_fault(table_column)
            elif item in FIELD_USAGE:
                column_name_limit = FIELD_TO_TABLE_COLUMN[FIELD_LIMIT[FIELD_USAGE.index(item)]]
                table_column_limit = getattr(self.table.c, column_name_limit)
                fun_ = self._fun_usage(table_column, table_column_limit)
            elif item in FIELD_MAX:
                column_name_limit = FIELD_TO_TABLE_COLUMN[FIELD_LIMIT[FIELD_MAX.index(item)]]
                column_name_fault = FIELD_TO_TABLE_COLUMN[FIELD_FAULT[FIELD_MAX.index(item)]]
                table_column_limit = getattr(self.table.c, column_name_limit)
                table_column_fault = getattr(self.table.c, column_name_fault)
                fun_ = self._fun_max(table_column, table_column_limit, table_column_fault)
        return fun_

    def _group_by_query(self, select_query):
        by_usage_list = convert_to_list(self.by_usage)
        by_fault_list = convert_to_list(self.by_fault)
        if self._is_multi_uids:
            query = select_query.group_by(self.table.c.id)
        else:
            # it is correct because in query result all records having the same _fun_time_from()
            # also have same _fun_time_to()
            # so .group_by(_fun_time_to()) doesn't create new groups after .group_by(_fun_time_from())
            # but we need both this group conditions because mysql throws an error otherwise
            query = select_query.group_by(self._fun_time_from()).group_by(self._fun_time_to())
        if self.order_by:
            order_by_aggregate_func = self._aggregate_fun_case(self.order_by)
            query.append_order_by(desc(order_by_aggregate_func))
        else:
            if self._is_multi_uids:
                query.append_order_by(asc(self.table.c.id))
            else:
                query.append_order_by(self._fun_time_from())

        # add filter for having grouped data
        if by_usage_list:  # add filtering by user id
            filter_fun_list = []
            for item in by_usage_list:
                if item in FIELD_AVERAGE:
                    index = FIELD_AVERAGE.index(item)
                    filter_fun = self.by_usage_percentage <= self._aggregate_fun_case(FIELD_USAGE[index])
                else:
                    index = FIELD_MAX.index(item)
                    filter_fun = self.by_usage_percentage * self._aggregate_fun_case(
                        FIELD_LIMIT[index]
                    ) <= self._aggregate_fun_case(item)
                filter_fun_list.append(filter_fun)
            query.append_having(or_(*filter_fun_list))

        if by_fault_list:
            by_fault_filter = [self.threshold <= self._aggregate_fun_case(funk_key_) for funk_key_ in by_fault_list]
            query.append_having(or_(*by_fault_filter))
        if self.limit != 0 and self.limit is not None:
            query = query.limit(self.limit)

        return query

    def _columns_query(self):
        """
        Generate output columns for SELECT <_columns_query(self)> FROM ...
        :return:
        """
        columns_agregate_func = []
        for column_key in self.show_columns:
            column_fun = self._aggregate_fun_case(column_key)
            if isinstance(column_fun, list):
                columns_agregate_func.extend(column_fun)
            else:
                if column_key not in ('From', 'To'):  # digest not support label
                    column_fun = label(column_key, column_fun)
                columns_agregate_func.append(column_fun)  # add label
        return columns_agregate_func

    def _check_need_time_count(self):
        columns = {
            'aCPU',
            'uCPU',
            'aEP',
            'uEP',
            'aVMem',
            'uVMem',
            'aPMem',
            'uPMem',
            'aNproc',
            'uNproc',
            'aIO',
            'uIO',
            'aIOPS',
            'uIOPS',
        }
        return bool(columns & (set(self.show_columns) | {self.order_by} | set(self.by_usage or set())))

    def _check_need_join(self):
        return self._check_need_time_count() and not self._is_multi_uids

    def select_query(self, columns_=None, need_join=False):
        """
        :type need_join: bool
        """
        if columns_ is None:
            columns_ = self._columns_query()
        if need_join:
            where_query = and_(
                self._where_time_period(table=self._table_alive), self._where_uid(uid=0, table=self._table_alive)
            )
        else:
            where_query = and_(self._where_time_period(), self._where_uid())
        if self.server_id:  # add filtering by server id
            where_query = and_(where_query, self._where_server_id())
        query = select(columns_).where(where_query)
        if need_join:
            _table_joined = self._table_alive.outerjoin(
                self.table, and_(self._table_alive.c.created == self.table.c.created, self._where_uid(uid=self.uid))
            )
            query = query.select_from(_table_joined)
        return query

    def main_query(self):
        columns_ = self._columns_query()
        query = self.select_query(columns_=columns_, need_join=self._check_need_join())
        query = self._group_by_query(query)
        return query

    def _min_max_created(self):
        """
        SELECT
            MIN(created) AS MinCreated,
            MAX(created) AS MaxCreated
        FROM
            lve_stats2_history
        WHERE
            id = <ID> AND
            created BETWEEN 'xxxx' AND 'yyyy' AND
            server_id = 'localhost';
        """
        where_query = and_(self._where_time_period(), self._where_uid(), self._where_server_id())
        query = select([func.min(self.table.c.created), func.max(self.table.c.created)]).where(where_query)
        time_start = time.time()
        q = str(query.compile(compile_kwargs={"literal_binds": True})).replace('\n', ' ')
        self.log.debug(q)
        data = self.dbengine.execute(query)
        self.log.debug('query time: %s', time.time() - time_start)
        return data.fetchall()[0]

    def proceed_dyn_time_unit(self):
        min_created, max_created = self._min_max_created()
        if max_created is None:  # no data
            return self.result_corrector
        # we need manipulate with datetime data in local timezone
        period_from = gm_to_local(unixtimestamp_to_gm_datetime(min_created))
        period_to = gm_to_local(unixtimestamp_to_gm_datetime(max_created))
        time_unit_groups = dyn_time_unit_groups(period_from, period_to)

        rows = []
        for _from, _to, _time_unit in reversed(time_unit_groups):
            # create instance copy for modify some attributes
            self_copy = copy.copy(self)
            self_copy.period_from = gm_datetime_to_unixtimestamp(local_to_gm(_from))
            self_copy.period_to = gm_datetime_to_unixtimestamp(local_to_gm(_to)) - 1
            self_copy.time_unit = _time_unit
            self_copy.limit = 0
            rows.extend(self_copy.proceed())
        return rows

    def proceed(self):
        # check and return some data without run sql query
        if self.uid == tuple() or self.uid == []:
            return []
        if self.uid is not None and not isinstance(self.uid, (list, tuple)) and self.uid <= 0:
            return []

        query = self.main_query()
        time_start = time.time()
        q = str(query.compile(compile_kwargs={"literal_binds": True}))
        self.log.debug(q.replace('\n', ' '))
        conn = self.dbengine.connect()
        try:
            cursor = conn.execute(query)
            self.log.debug('query time: %s', time.time() - time_start)
            self.result_corrector.rows = cursor.fetchall()
        except LveStatsPluginTerminated as e:
            conn.close()
            raise LveStatsPluginTerminated() from e
        else:
            conn.close()
            return self.result_corrector

    def proceed_dict(self):
        return [dict(zip(self.show_columns, items_val)) for items_val in self.proceed()]


class _HistoryShowX1(HistoryShow):
    def __init__(self, *args, **kwargs):
        HistoryShow.__init__(self, *args, **kwargs)
        if 'ID' not in self.show_columns:
            self.show_columns = ['ID'] + self.show_columns
        self._labels = []  # variable for control duplicated labels

    def _aggregate_fun_case(self, item):
        """
        :type item: str
        """
        if item == 'anyF':
            fun_ = [
                self.table.c.mem_fault,
                self.table.c.memphy_fault,
                self.table.c.mep_fault,
                self.table.c.nproc_fault,
                self.table.c.cpu_fault,
                self.table.c.io_fault,
                self.table.c.iops_fault,
            ]
        elif item == 'ID':
            fun_ = label('id', self.table.c.id)
        elif item == 'From':
            fun_ = self.period_from
        elif item == 'To':
            fun_ = self.period_to
        else:
            column_name = FIELD_TO_TABLE_COLUMN[item]
            table_column = getattr(self.table.c, column_name)
            if item in (FIELD_AVERAGE + FIELD_LIMIT + FIELD_FAULT):
                fun_ = label(column_name, table_column)
            elif item in FIELD_USAGE:
                column_name_limit = FIELD_TO_TABLE_COLUMN[FIELD_LIMIT[FIELD_USAGE.index(item)]]
                table_column_limit = label(column_name_limit, getattr(self.table.c, column_name_limit))
                fun_ = [table_column, table_column_limit]
            elif item in FIELD_MAX:
                column_name_limit = FIELD_TO_TABLE_COLUMN[FIELD_LIMIT[FIELD_MAX.index(item)]]
                column_name_fault = FIELD_TO_TABLE_COLUMN[FIELD_FAULT[FIELD_MAX.index(item)]]
                table_column_fault = label(column_name_fault, getattr(self.table.c, column_name_fault))
                table_column_limit = label(column_name_limit, getattr(self.table.c, column_name_limit))
                table_column_max = label(column_name + '_max', table_column)
                fun_ = [table_column, table_column_limit, table_column_fault, table_column_max]
        return fun_

    def _columns_query(self):
        columns_agregate_func = []

        show_columns = self.show_columns + (self.by_fault or [])
        if self.by_usage:
            for item in convert_to_list(self.by_usage):
                if item in FIELD_AVERAGE:
                    index = FIELD_AVERAGE.index(item)
                    show_columns.append(FIELD_USAGE[index])
                else:
                    index = FIELD_MAX.index(item)
                    show_columns.extend([FIELD_FAULT[index], item])

        if self.order_by:
            show_columns.append(self.order_by)

        for column_key in show_columns:
            column_fun = self._aggregate_fun_case(column_key)
            if isinstance(column_fun, list):
                for fun_ in column_fun:
                    if hasattr(fun_, 'name') and fun_.name not in self._labels:  # prevent alias duplication
                        columns_agregate_func.append(fun_)
                        self._labels.append(fun_.name)
            else:
                if hasattr(column_fun, 'name') and column_fun.name not in self._labels:  # prevent alias duplication
                    columns_agregate_func.append(column_fun)
                    self._labels.append(column_fun.name)
        return columns_agregate_func


class _HistoryShowX60(_HistoryShowX1):
    AGGREGATE_PERIOD = 60 * 60

    def __init__(self, *args, **kwargs):
        _HistoryShowX1.__init__(self, *args, table=history_x60.__table__, **kwargs)

        # correct and rewrite time count and period
        self.period_from, self.period_to = self.get_history_x60_from_to()
        self.time_count = kwargs.get('time_count') or self.get_time_count()

    def get_time_count(self):
        if (self.period_from, self.period_to) == (None, None):
            return 0
        return _HistoryShowX1.get_time_count(self)

    def get_history_x60_from_to(self):
        """
        calculate present in aggregate table from and to time
        """
        if self.period_to - self.period_from <= self.AGGREGATE_PERIOD:
            return None, None
        between_query = self.table.c.created.between(self.period_from + self.AGGREGATE_PERIOD, self.period_to)
        query = select([func.min(self.table.c.created), func.max(self.table.c.created)]).where(
            and_(between_query, self._where_server_id())
        )
        time_start = time.time()
        self.log.debug(str(query.compile(compile_kwargs={"literal_binds": True})).replace('\n', ' '))
        result = self.dbengine.execute(query).fetchall()[0]
        self.log.debug('query time: %s', time.time() - time_start)
        create_min, create_max = result
        if create_max is not None:
            return create_min - self.AGGREGATE_PERIOD + 1, create_max  # "+1" for exclude from timestamp
        else:
            return result

    # rewrite average function generating
    def _aggregate_fun_case(self, item):
        """
        :type item: str
        """
        if item in FIELD_AVERAGE:
            column_name = FIELD_TO_TABLE_COLUMN[item]
            table_column = getattr(self.table.c, column_name)
            return label(column_name, table_column * self.table.c.time)
        else:
            return _HistoryShowX1._aggregate_fun_case(self, item)


class HistoryShowUnion(HistoryShow):
    """
    Class for retrieve statistics data using two tables
    """

    def __init__(self, *args, **kwargs):
        HistoryShow.__init__(self, *args, **kwargs)
        self._alias = LVE_STATS_2_TABLENAME_PREFIX + 'union'
        kwargs.update({"time_count": self.time_count})
        self.x60 = _HistoryShowX60(*args, **kwargs)
        self.x1 = _HistoryShowX1(*args, **kwargs)
        self._need_union = self.x60.period_to is not None and self._is_multi_uids  # detect need union tables
        if self._need_union:
            self.table = self._select_union_query()

    # rewrite '_aggregate_fun_case' for correct calculate maximum
    def _aggregate_fun_case(self, item):
        """
        :type item: str
        """
        if self._need_union and item in FIELD_MAX:
            column_name = FIELD_TO_TABLE_COLUMN[item]
            column_name_limit = FIELD_TO_TABLE_COLUMN[FIELD_LIMIT[FIELD_MAX.index(item)]]
            column_name_fault = FIELD_TO_TABLE_COLUMN[FIELD_FAULT[FIELD_MAX.index(item)]]
            column_limit = getattr(self.table.c, column_name_limit)
            column_fault = getattr(self.table.c, column_name_fault)
            column_max = getattr(self.table.c, column_name + '_max')
            fun_ = self._fun_max(column_max, column_limit, column_fault)
            return fun_
        else:
            return HistoryShow._aggregate_fun_case(self, item)

    def _select_union_query(self):
        """
        union two tables
        """
        with warnings.catch_warnings():
            warnings.simplefilter("ignore", category=sa_exc.SAWarning)
            union_query = self.x1.select_query().where(
                not_(self.x1.table.c.created.between(self.x60.period_from, self.x60.period_to))
            )
            union_query = union_query.union_all(self.x60.select_query())
            union_query = alias(union_query, self._alias)
        return union_query

    def select_query(self, columns_=None, need_join=None):
        if self._need_union:
            return select(columns_)
        else:
            return HistoryShow.select_query(self, columns_=columns_, need_join=need_join)


def get_supported_columns(lve_version=None, mode=None):
    """
    preparation list columns depending of the lve version
    :type mode: Union[None, str]
    :type lve_version: Union[None, int]
    """
    columns = []
    if mode == 'v1':
        columns = [
            'aCPU',
            'mCPU',
            'lCPU',
            'aEP',
            'mEP',
            'lEP',
            'aVMem',
            'mVMem',
            'lVMem',
            'VMemF',
            'EPf',
            'aPMem',
            'mPMem',
            'lPMem',
            'aNproc',
            'mNproc',
            'lNproc',
            'PMemF',
            'NprocF',
            'aIO',
            'mIO',
            'lIO',
        ]
        if lve_version is None or lve_version > 6:
            columns.extend(['aIOPS', 'mIOPS', 'lIOPS'])
    elif mode == 'v2':
        columns = [
            'aCPU',
            'lCPU',
            'CPUf',
            'aEP',
            'lEP',
            'EPf',
            'aVMem',
            'lVMem',
            'VMemF',
            'aPMem',
            'lPMem',
            'PMemF',
            'aNproc',
            'lNproc',
            'NprocF',
            'aIO',
            'lIO',
            'IOf',
        ]
        if lve_version is None or lve_version > 6:
            columns.extend(['aIOPS', 'lIOPS', 'IOPSf'])
    elif mode is None:  # show all columns, v1 and v2
        columns = [
            'aCPU',
            'uCPU',
            'mCPU',
            'lCPU',
            'CPUf',
            'aEP',
            'uEP',
            'mEP',
            'lEP',
            'EPf',
            'aVMem',
            'uVMem',
            'mVMem',
            'lVMem',
            'VMemF',
            'aPMem',
            'uPMem',
            'mPMem',
            'lPMem',
            'PMemF',
            'aNproc',
            'uNproc',
            'mNproc',
            'lNproc',
            'NprocF',
            'aIO',
            'uIO',
            'mIO',
            'lIO',
            'IOf',
        ]
        if lve_version is None or lve_version > 6:
            columns.extend(['aIOPS', 'mIOPS', 'uIOPS', 'lIOPS', 'IOPSf'])
    return columns

Youez - 2016 - github.com/yon3zu
LinuXploit