Приветствую форумчане.

Столкнулись с проблемой крайне низкой производительности на новом сервере, по сравнению со старым.

Старый сервер более слабый по характеристикам и крутиться на виртуальной машине Windows Server 2012 + Postgre14

Новый сервак раза в два мощнее, стоит Astralinux + Postgre15.

И по итогу на новом сервере крайне низкая скорость работы, документ распроводится вместо 2х минут, 6 часов.


Кто что может подсказать?)

Какая конфа? Какой документ? Размер базы? Количество пользователей?

Почему документ распроводится, а не проводится?

(0) Проблема только с одним конкретным документом?

Проблема с любым документом, конфа УПП, взят бэкап. Документ пробовали и проводить и распроводить. Привёл сравнение времени работы в распроведении документа на новом и старом сервере.

На новом сервере исключительно один пользователь, так как пока только тестирование проводиться, а вот на старом, пользователей около 300.

(4) Сказки...

(6) Вот и сижу голову ломаю, просчитался, но где?

(6) есть подозрение на внутренние механизмы самой ОС

(8) это вряд-ли, тесты на производительность cpu и дисковой подсистемы, размер занятой ОЗУ 1С и БП снимут эти подозрения. Скорее уж настройки БД.

(0) УПП у вас под чем работается в толстом клиенте?

+ Под эмулятором виндовс?

(0) версии платформы какие?

Физический сервак, на ОС Астра линукс, тест гилева показал, в 2 раза выше попугаев чем на старом сервере.

(12) 8.3.22.1923

(13) В два раза - было 8, стало аж все 16? Или 45 и 90?

На SQl 10 на старом
На новом 38.

В файловом варианте, так вообще 98 баллов

для линукса нужен грамотный админ, пока его не будет - хоть сколько попугаев пусть показывает

Настройки PgSQL?

(17) Постгре чей? И какая версия?
Внутри 1С версия совместимости какая стоит?

Работал я когда-то на УПП под постресом...

Первая фича конечно что УПП на автоматических блокировках, а значит пострес блокирует всю табличку в которую пишет. Вот распроводите Вы документ, он пишет какие-нибудь остатки по партиям... Получается практически однопользовательская система :-)

Вторая фича это конечно то, что постгрес менее умно строит планы запросов, поэтому если запрос написан не слишком хорошо , то скуль ещё скушает, а пострес подавится. Лучше сложные запросы переделывать на временные таблички с индексами.

PG PRO 15 версия.

Ну так на прошлом сервере на postgre нормально работал)

(22) это прекрасно что работала
настройки pg надеюсь не дефолтные?

(22) конфигуратор-отладка-замер производительности

(0) у нас ЕРП на 1500 пользователей. Работает стабильно на Postgres + Astra.

1. зачем переехали?
2. Тестировали ли перед переездом? Функциональное тестирование/нагрузочное тестирование?

(22) На прошлом была 14я версия PGSQL, на новом 15я.
Поставить 14ю для сравнения никто еще не предлагал?

Слабо верится, но так бывает, особенно если в код запросов УПП кто-то залез и что-то поправил. Конфигурация тут не важна.

+(24) Именно так, а потом сравнить результаты.

(16) Маловато как-то, но вполне достойно.

Надеюсь ничего не твикали в системе перед тем как это обнаружили?
Некоторые рекомендованные твики уже устарели и оказывают негативное влияние на производительность (но не на тест Гилева), а некоторые настолько вредны, что их использовать вообще нельзя.

(0) "документ распроводится вместо 2х минут, 6 часов"
Классическая ситуация с nested loop join. Очевидно, новый сервер ошибается при выборе метода соединения таблиц, выбирая самый неподходящий для больших соединений.. Как его научить выбирать более адекватно - четкого алгоритма нет. Самый простой способ - это вообще исключит nested loop из предпочтений в настройках, но это замедлит работу в целом.

(22) Кто мешает поставить прошлый?

(24) Под замером 6 часов превратятся в сутки.

(25) УПП и ЕРП разные конфигурации работающие под разную платформу

Тут только один алгоритм.

1. Ищете узкое место
2. Решаете проблему

99% что какой-то запрос. Хотя 6 часов...

ПГ то хоть по методичке Дорошкевича настраивали?

(30)postgres то один. и алгоритмы решения проблем одинаковые.

Настройки самой PostgrePro испробованы разные, как те, что на рабочих серверах использовались, так и несколько раз менялись, итог один.

Любые тяжелые операции многократно больше времени занимают, при этом в мониторинге видно, что на процесс расходуется один и тот же размер памяти(вообще не изменяется, условно дошло до 177000 КБ и висит) и 100% нагрузка на одно ядро с периодической просадкой на 1-2 секунды

(32) Основные проблемы с ПГ решили в 18-19 годах, конфигурации могут быть написаны с учетом изменения архитектуры, а могут быть в режиме совместимости.

Что касается алгоритмов, смотря какие цели,

если цель сделать быстро, то стенд, специалитет.
если разобраться, то (31)

При этом очень мелкий операции, выполняются быстрее чем на основном сервере.

Результаты тестирования гилева дают 45 очков с использованием  SQL

PostgrePro ?
Зачем???

(36) а в чём проблема? только конкретику

(37) Для начала - она платная.
А если вы заплатили за PostgrePro - то и поддержка их у вас должна быть. Просто опишите им проблему в письме.

я склоняюсь к тому, что очень криво настроили сам PG

(39) например оперативки не выдали (всего, для 1 запроса и т.д.), раз на мелких операциях быстро а больших затык

(40) выдали, все настройки самой PG уже несколько раз изменены.

Склоняемся всё больше в сторону настройки самой ОС Astra linux

Так отладчик запускали? На каком месте затык? Полюбому же эти 6 часов в какую-то конкретную строку кода упираются.

И как уже выше предполагали - наверняка это будет запрос. А если запрос - то стопудово там будет использование каких-нить виртуальных таблиц. Срез последних или еще чонить.

(41) Какая Фс на на которой лежит постгрес ?

(36) вроде только она в Реестре российского ПО.

Иначе зачем тогда Астру ставить?

Мне нравится:
1.Новый комп мощнее старого. А по факту: проц имеет больше ядер, но меньшую тактовую + меньше оперативки.
2.Мы уже много раз настройки меняли. Что меняли, кому меняли?

Пока не расскажешь характеристики сервера и не выложишь настройки - это будет вохоже экстрасенсов по ТНТ, ну или соревнование телепатов.

(46) С тактовой еще затык
Сейчас же турбобуст хитрый и поддержка в ОСи нужна
Может оно работает на базовой в 2ГГц

Имхо Debian и обычный PgSQL от 1С
Если не хочется танцев с бубнами, да и то админ нужен
И железо изначально правильное

(47) там много ща нюансов
(0) если надоест, мыло в личке

(47) ну не в 100 же раз посадка!? Из-за буста. Ну в 10поверю.

Явно косяк какой-то в настройках/поведении ПГ.

(48) а как же реест российского ПО?

(51) там уже появился MS Office?

(47) высокую производительность надо включать везде в настройках электропитания, кроме ноутбуков.

(47) хотя на Линуксе хз как процы работают

А Линукс вообще понимает современные большие/малые ядра?

(55) Не очень хорошо понимает

(46) Он уже второй день рассказывает как ему плохо живется но до сих пор не показал настройки постгри.
+ pgpro - может они решили мимо кассы его поставить, а там для таких пользователей маленький "нюанс"

(57) там в принципе на астре лбной сплошной нюанс

online_analyze смотри.

(58) Не пользовался ни разу у меня душа больше к RedOS лежит. Я бы его тоже попробовал, тем более сервер пока не в продакшене

В продолжение, вот запрос из логов, на котором система долго висит.

логи

2024-02-15 12:07:15.631 +05 [2713] LOG:  statement: UPDATE _AccRgAT11724 SET _Fld1703 = _AccRgAT11724._Fld1703 + T2._Fld1703, _TurnoverDt1715 = _AccRgAT11724._TurnoverDt1715 + T2._TurnoverDt1715, _TurnoverCt1716 = _AccRgAT11724._TurnoverCt1716 + T2._TurnoverCt1716, _Turnover1717 = _AccRgAT11724._Turnover1717 + T2._Turnover1717, _Fld1704 = _AccRgAT11724._Fld1704 + T2._Fld1704, _TurnoverDt1718 = _AccRgAT11724._TurnoverDt1718 + T2._TurnoverDt1718, _TurnoverCt1719 = _AccRgAT11724._TurnoverCt1719 + T2._TurnoverCt1719, _Turnover1720 = _AccRgAT11724._Turnover1720 + T2._Turnover1720, _Fld1705 = _AccRgAT11724._Fld1705 + T2._Fld1705, _TurnoverDt1721 = _AccRgAT11724._TurnoverDt1721 + T2._TurnoverDt1721, _TurnoverCt1722 = _AccRgAT11724._TurnoverCt1722 + T2._TurnoverCt1722, _Turnover1723 = _AccRgAT11724._Turnover1723 + T2._Turnover1723
    FROM pg_temp.tt8 T2
    WHERE (T2._Period = _AccRgAT11724._Period AND T2._AccountRRef = _AccRgAT11724._AccountRRef AND T2._Fld1701RRef = _AccRgAT11724._Fld1701RRef AND (T2._Fld1702RRef == _AccRgAT11724._Fld1702RRef) AND (T2._Value1_TYPE == _AccRgAT11724._Value1_TYPE AND T2._Value1_RTRef == _AccRgAT11724._Value1_RTRef AND T2._Value1_RRRef == _AccRgAT11724._Value1_RRRef) AND _AccRgAT11724._Splitter = CAST(0 AS NUMERIC)) AND (T2._EDCount = CAST(1 AS NUMERIC))
2024-02-15 12:07:17.218 +05 [2713] LOG:  duration: 1586.834 ms
2024-02-15 12:07:17.218 +05 [2713] LOG:  statement: INSERT INTO _AccRgAT11724 (_Period, _AccountRRef, _Fld1701RRef, _Fld1702RRef, _Value1_TYPE, _Value1_RTRef, _Value1_RRRef, _Fld1703, _TurnoverDt1715, _TurnoverCt1716, _Turnover1717, _Fld1704, _TurnoverDt1718, _TurnoverCt1719, _Turnover1720, _Fld1705, _TurnoverDt1721, _TurnoverCt1722, _Turnover1723, _Splitter) SELECT
    T1._Period,
    T1._AccountRRef,
    T1._Fld1701RRef,
    T1._Fld1702RRef,
    T1._Value1_TYPE,
    T1._Value1_RTRef,
    T1._Value1_RRRef,
    T1._Fld1703,
    T1._TurnoverDt1715,
    T1._TurnoverCt1716,
    T1._Turnover1717,
    T1._Fld1704,
    T1._TurnoverDt1718,
    T1._TurnoverCt1719,
    T1._Turnover1720,
    T1._Fld1705,
    T1._TurnoverDt1721,
    T1._TurnoverCt1722,
    T1._Turnover1723,
    CAST(CAST(0 AS NUMERIC) AS NUMERIC(10, 0))
    FROM pg_temp.tt8 T1
    LEFT OUTER JOIN _AccRgAT11724 T2
    ON T1._Period = T2._Period AND T1._AccountRRef = T2._AccountRRef AND T1._Fld1701RRef = T2._Fld1701RRef AND (T1._Fld1702RRef == T2._Fld1702RRef) AND (T1._Value1_TYPE == T2._Value1_TYPE AND T1._Value1_RTRef == T2._Value1_RTRef AND T1._Value1_RRRef == T2._Value1_RRRef) AND T2._Splitter = CAST(0 AS NUMERIC)
    WHERE T2._Period IS NULL AND T1._EDCount = CAST(1 AS NUMERIC)
2024-02-15 12:07:17.359 +05 [2713] LOG:  duration: 140.895 ms
2024-02-15 12:07:17.360 +05 [2713] LOG:  statement: UPDATE _AccRgAT21725 SET _Fld1703 = _AccRgAT21725._Fld1703 + T2._Fld1703, _TurnoverDt1715 = _AccRgAT21725._TurnoverDt1715 + T2._TurnoverDt1715, _TurnoverCt1716 = _AccRgAT21725._TurnoverCt1716 + T2._TurnoverCt1716, _Turnover1717 = _AccRgAT21725._Turnover1717 + T2._Turnover1717, _Fld1704 = _AccRgAT21725._Fld1704 + T2._Fld1704, _TurnoverDt1718 = _AccRgAT21725._TurnoverDt1718 + T2._TurnoverDt1718, _TurnoverCt1719 = _AccRgAT21725._TurnoverCt1719 + T2._TurnoverCt1719, _Turnover1720 = _AccRgAT21725._Turnover1720 + T2._Turnover1720, _Fld1705 = _AccRgAT21725._Fld1705 + T2._Fld1705, _TurnoverDt1721 = _AccRgAT21725._TurnoverDt1721 + T2._TurnoverDt1721, _TurnoverCt1722 = _AccRgAT21725._TurnoverCt1722 + T2._TurnoverCt1722, _Turnover1723 = _AccRgAT21725._Turnover1723 + T2._Turnover1723
    FROM pg_temp.tt8 T2
    WHERE (T2._Period = _AccRgAT21725._Period AND T2._AccountRRef = _AccRgAT21725._AccountRRef AND T2._Fld1701RRef = _AccRgAT21725._Fld1701RRef AND (T2._Fld1702RRef == _AccRgAT21725._Fld1702RRef) AND (T2._Value1_TYPE == _AccRgAT21725._Value1_TYPE AND T2._Value1_RTRef == _AccRgAT21725._Value1_RTRef AND T2._Value1_RRRef == _AccRgAT21725._Value1_RRRef) AND (T2._Value2_TYPE == _AccRgAT21725._Value2_TYPE AND T2._Value2_RTRef == _AccRgAT21725._Value2_RTRef AND T2._Value2_RRRef == _AccRgAT21725._Value2_RRRef) AND _AccRgAT21725._Splitter = CAST(0 AS NUMERIC)) AND (T2._EDCount = CAST(2 AS NUMERIC))
2024-02-15 12:07:30.265 +05 [2713] LOG:  duration: 12905.457 ms
2024-02-15 12:07:30.266 +05 [2713] LOG:  statement: INSERT INTO _AccRgAT21725 (_Period, _AccountRRef, _Fld1701RRef, _Fld1702RRef, _Value1_TYPE, _Value1_RTRef, _Value1_RRRef, _Value2_TYPE, _Value2_RTRef, _Value2_RRRef, _Fld1703, _TurnoverDt1715, _TurnoverCt1716, _Turnover1717, _Fld1704, _TurnoverDt1718, _TurnoverCt1719, _Turnover1720, _Fld1705, _TurnoverDt1721, _TurnoverCt1722, _Turnover1723, _Splitter) SELECT
    T1._Period,
    T1._AccountRRef,
    T1._Fld1701RRef,
    T1._Fld1702RRef,
    T1._Value1_TYPE,
    T1._Value1_RTRef,
    T1._Value1_RRRef,
    T1._Value2_TYPE,
    T1._Value2_RTRef,
    T1._Value2_RRRef,
    T1._Fld1703,
    T1._TurnoverDt1715,
    T1._TurnoverCt1716,
    T1._Turnover1717,
    T1._Fld1704,
    T1._TurnoverDt1718,
    T1._TurnoverCt1719,
    T1._Turnover1720,
    T1._Fld1705,
    T1._TurnoverDt1721,
    T1._TurnoverCt1722,
    T1._Turnover1723,
    CAST(CAST(0 AS NUMERIC) AS NUMERIC(10, 0))
    FROM pg_temp.tt8 T1
    LEFT OUTER JOIN _AccRgAT21725 T2
    ON T1._Period = T2._Period AND T1._AccountRRef = T2._AccountRRef AND T1._Fld1701RRef = T2._Fld1701RRef AND (T1._Fld1702RRef == T2._Fld1702RRef) AND (T1._Value1_TYPE == T2._Value1_TYPE AND T1._Value1_RTRef == T2._Value1_RTRef AND T1._Value1_RRRef == T2._Value1_RRRef) AND (T1._Value2_TYPE == T2._Value2_TYPE AND T1._Value2_RTRef == T2._Value2_RTRef AND T1._Value2_RRRef == T2._Value2_RRRef) AND T2._Splitter = CAST(0 AS NUMERIC)
    WHERE T2._Period IS NULL AND T1._EDCount = CAST(2 AS NUMERIC)
2024-02-15 12:07:32.908 +05 [2713] LOG:  duration: 2642.143 ms
2024-02-15 12:07:32.909 +05 [2713] LOG:  statement: UPDATE _AccRgAT31726 SET _Fld1703 = _AccRgAT31726._Fld1703 + T2._Fld1703, _TurnoverDt1715 = _AccRgAT31726._TurnoverDt1715 + T2._TurnoverDt1715, _TurnoverCt1716 = _AccRgAT31726._TurnoverCt1716 + T2._TurnoverCt1716, _Turnover1717 = _AccRgAT31726._Turnover1717 + T2._Turnover1717, _Fld1704 = _AccRgAT31726._Fld1704 + T2._Fld1704, _TurnoverDt1718 = _AccRgAT31726._TurnoverDt1718 + T2._TurnoverDt1718, _TurnoverCt1719 = _AccRgAT31726._TurnoverCt1719 + T2._TurnoverCt1719, _Turnover1720 = _AccRgAT31726._Turnover1720 + T2._Turnover1720, _Fld1705 = _AccRgAT31726._Fld1705 + T2._Fld1705, _TurnoverDt1721 = _AccRgAT31726._TurnoverDt1721 + T2._TurnoverDt1721, _TurnoverCt1722 = _AccRgAT31726._TurnoverCt1722 + T2._TurnoverCt1722, _Turnover1723 = _AccRgAT31726._Turnover1723 + T2._Turnover1723
    FROM pg_temp.tt8 T2
    WHERE (T2._Period = _AccRgAT31726._Period AND T2._AccountRRef = _AccRgAT31726._AccountRRef AND T2._Fld1701RRef = _AccRgAT31726._Fld1701RRef AND (T2._Fld1702RRef == _AccRgAT31726._Fld1702RRef) AND (T2._Value1_TYPE == _AccRgAT31726._Value1_TYPE AND T2._Value1_RTRef == _AccRgAT31726._Value1_RTRef AND T2._Value1_RRRef == _AccRgAT31726._Value1_RRRef) AND (T2._Value2_TYPE == _AccRgAT31726._Value2_TYPE AND T2._Value2_RTRef == _AccRgAT31726._Value2_RTRef AND T2._Value2_RRRef == _AccRgAT31726._Value2_RRRef) AND (T2._Value3_TYPE == _AccRgAT31726._Value3_TYPE AND T2._Value3_RTRef == _AccRgAT31726._Value3_RTRef AND T2._Value3_RRRef == _AccRgAT31726._Value3_RRRef) AND _AccRgAT31726._Splitter = CAST(0 AS NUMERIC)) AND (T2._EDCount = CAST(3 AS NUMERIC))
2024-02-15 12:17:42.386 +05 [2394] СООБЩЕНИЕ:  начата контрольная точка: time

(61) покажи настройки СУБД

Сейчас они сброшены до стандартных, за исключение включенного логирования.

(63) Стандарты у всех свои.

# -----------------------------
# PostgreSQL configuration file
# -----------------------------
#
# This file consists of lines of the form:
#
#   name = value
#
# (The "=" is optional.)  Whitespace may be used.  Comments are introduced with
# "#" anywhere on a line.  The complete list of parameter names and allowed
# values can be found in the PostgreSQL documentation.
#
# The commented-out settings shown in this file represent the default values.
# Re-commenting a setting is NOT sufficient to revert it to the default value;
# you need to reload the server.
#
# This file is read on server startup and when the server receives a SIGHUP
# signal.  If you edit the file on a running system, you have to SIGHUP the
# server for the changes to take effect, run "pg_ctl reload", or execute
# "SELECT pg_reload_conf()".  Some parameters, which are marked below,
# require a server shutdown and restart to take effect.
#
# Any parameter can also be given as a command-line option to the server, e.g.,
# "postgres -c log_connections=on".  Some parameters can be changed at run time
# with the "SET" SQL command.
#
# Memory units:  B  = bytes            Time units:  us  = microseconds
#                kB = kilobytes                     ms  = milliseconds
#                MB = megabytes                     s   = seconds
#                GB = gigabytes                     min = minutes
#                TB = terabytes                     h   = hours
#                                                   d   = days


#------------------------------------------------------------------------------
# FILE LOCATIONS
#------------------------------------------------------------------------------

# The default values of these variables are driven from the -D command-line
# option or PGDATA environment variable, represented here as ConfigDir.

#data_directory = 'ConfigDir'        # use data in another directory
                    # (change requires restart)
#hba_file = 'ConfigDir/pg_hba.conf'    # host-based authentication file
                    # (change requires restart)
#ident_file = 'ConfigDir/pg_ident.conf'    # ident configuration file
                    # (change requires restart)

# If external_pid_file is not explicitly set, no extra PID file is written.
#external_pid_file = ''            # write an extra PID file
                    # (change requires restart)


#------------------------------------------------------------------------------
# CONNECTIONS AND AUTHENTICATION
#------------------------------------------------------------------------------

# - Connection Settings -

#listen_addresses = 'localhost'        # what IP address(es) to listen on;
                    # comma-separated list of addresses;
                    # defaults to 'localhost'; use '*' for all
                    # (change requires restart)
port = 5432 # (change requires restart)
max_connections = 100            # (change requires restart)
#reserved_connections = 0        # (change requires restart)
#superuser_reserved_connections = 3    # (change requires restart)
#unix_socket_directories = '/tmp'    # comma-separated list of directories
                    # (change requires restart)
#unix_socket_group = ''            # (change requires restart)
#unix_socket_permissions = 0777        # begin with 0 to use octal notation
                    # (change requires restart)
#bonjour = off                # advertise server via Bonjour
                    # (change requires restart)
#bonjour_name = ''            # defaults to the computer name
                    # (change requires restart)

# - TCP settings -
# see "man tcp" for details

#tcp_keepalives_idle = 0        # TCP_KEEPIDLE, in seconds;
                    # 0 selects the system default
#tcp_keepalives_interval = 0        # TCP_KEEPINTVL, in seconds;
                    # 0 selects the system default
#tcp_keepalives_count = 0        # TCP_KEEPCNT;
                    # 0 selects the system default
#tcp_user_timeout = 0            # TCP_USER_TIMEOUT, in milliseconds;
                    # 0 selects the system default

#client_connection_check_interval = 0    # time between checks for client
                    # disconnection while running queries;
                    # 0 for never

# - Authentication -

#authentication_timeout = 1min        # 1s-600s
password_encryption = md5        # scram-sha-256 or md5
#scram_iterations = 4096
#db_user_namespace = off

# GSSAPI using Kerberos
#krb_server_keyfile = 'FILE:${sysconfdir}/krb5.keytab'
#krb_caseins_users = off
#gss_accept_delegation = off

# - SSL -

#ssl = off
#ssl_ca_file = ''
#ssl_cert_file = 'server.crt'
#ssl_crl_file = ''
#ssl_crl_dir = ''
#ssl_key_file = 'server.key'
#ssl_ciphers = 'HIGH:MEDIUM:+3DES:!aNULL' # allowed SSL ciphers
#ssl_prefer_server_ciphers = on
#ssl_ecdh_curve = 'prime256v1'
#ssl_min_protocol_version = 'TLSv1.2'
#ssl_max_protocol_version = ''
#ssl_dh_params_file = ''
#ssl_passphrase_command = ''
#ssl_passphrase_command_supports_reload = off


#------------------------------------------------------------------------------
# RESOURCE USAGE (except WAL)
#------------------------------------------------------------------------------

# - Memory -

shared_buffers = 128MB            # min 128kB
                    # (change requires restart)
#huge_pages = try            # on, off, or try
                    # (change requires restart)
#huge_page_size = 0            # zero for system default
                    # (change requires restart)
#temp_buffers = 8MB            # min 800kB
#max_prepared_transactions = 0        # zero disables the feature
                    # (change requires restart)
# Caution: it is not advisable to set max_prepared_transactions nonzero unless
# you actively intend to use prepared transactions.
#work_mem = 4MB                # min 64kB
#hash_mem_multiplier = 2.0        # 1-1000.0 multiplier on hash table work_mem
#maintenance_work_mem = 64MB        # min 1MB
#autovacuum_work_mem = -1        # min 1MB, or -1 to use maintenance_work_mem
#logical_decoding_work_mem = 64MB    # min 64kB
#max_stack_depth = 2MB            # min 100kB
#shared_memory_type = mmap        # the default is the first option
                    # supported by the operating system:
                    #   mmap
                    #   sysv
                    #   windows
                    # (change requires restart)
dynamic_shared_memory_type = posix    # the default is usually the first option
                    # supported by the operating system:
                    #   posix
                    #   sysv
                    #   windows
                    #   mmap
                    # (change requires restart)
#min_dynamic_shared_memory = 0MB    # (change requires restart)
#vacuum_buffer_usage_limit = 256kB    # size of vacuum and analyze buffer access strategy ring;
                    # 0 to disable vacuum buffer access strategy;
                    # range 128kB to 16GB

# - Disk -

#temp_file_limit = -1            # limits per-process temp file space
                    # in kilobytes, or -1 for no limit

# - Kernel Resources -

#max_files_per_process = 1000        # min 64
                    # (change requires restart)

# - Cost-Based Vacuum Delay -

#vacuum_cost_delay = 0            # 0-100 milliseconds (0 disables)
#vacuum_cost_page_hit = 1        # 0-10000 credits
#vacuum_cost_page_miss = 2        # 0-10000 credits
#vacuum_cost_page_dirty = 20        # 0-10000 credits
#vacuum_cost_limit = 200        # 1-10000 credits

# - Background Writer -

#bgwriter_delay = 200ms            # 10-10000ms between rounds
#bgwriter_lru_maxpages = 100        # max buffers written/round, 0 disables
#bgwriter_lru_multiplier = 2.0        # 0-10.0 multiplier on buffers scanned/round
#bgwriter_flush_after = 512kB        # measured in pages, 0 disables

# - Asynchronous Behavior -

#backend_flush_after = 0        # measured in pages, 0 disables
#effective_io_concurrency = 1        # 1-1000; 0 disables prefetching
#maintenance_io_concurrency = 10    # 1-1000; 0 disables prefetching
#max_worker_processes = 8        # (change requires restart)
#max_parallel_workers_per_gather = 2    # taken from max_parallel_workers
#max_parallel_maintenance_workers = 2    # taken from max_parallel_workers
#max_parallel_workers = 8        # maximum number of max_worker_processes that
                    # can be used in parallel operations
#parallel_leader_participation = on
#old_snapshot_threshold = -1        # 1min-60d; -1 disables; 0 is immediate
                    # (change requires restart)


#------------------------------------------------------------------------------
# WRITE-AHEAD LOG
#------------------------------------------------------------------------------

# - Settings -

#wal_level = replica            # minimal, replica, or logical
                    # (change requires restart)
#fsync = on                # flush data to disk for crash safety
                    # (turning this off can cause
                    # unrecoverable data corruption)
#synchronous_commit = on        # synchronization level;
                    # off, local, remote_write, remote_apply, or on
#wal_sync_method = fsync        # the default is the first option
                    # supported by the operating system:
                    #   open_datasync
                    #   fdatasync (default on Linux and FreeBSD)
                    #   fsync
                    #   fsync_writethrough
                    #   open_sync
#full_page_writes = on            # recover from partial page writes
#wal_log_hints = off            # also do full page writes of non-critical updates
                    # (change requires restart)
#wal_compression = off            # enables compression of full-page writes;
                    # off, pglz, lz4, zstd, or on
#wal_init_zero = on            # zero-fill new WAL files
#wal_recycle = on            # recycle WAL files
#wal_buffers = -1            # min 32kB, -1 sets based on shared_buffers
                    # (change requires restart)
#wal_writer_delay = 200ms        # 1-10000 milliseconds
#wal_writer_flush_after = 1MB        # measured in pages, 0 disables
#wal_skip_threshold = 2MB

#commit_delay = 0            # range 0-100000, in microseconds
#commit_siblings = 5            # range 1-1000

# - Checkpoints -

#checkpoint_timeout = 5min        # range 30s-1d
#checkpoint_completion_target = 0.9    # checkpoint target duration, 0.0 - 1.0
#checkpoint_flush_after = 256kB        # measured in pages, 0 disables
#checkpoint_warning = 30s        # 0 disables
max_wal_size = 1GB
min_wal_size = 80MB

# - Prefetching during recovery -

#recovery_prefetch = try        # prefetch pages referenced in the WAL?
#wal_decode_buffer_size = 512kB        # lookahead window used for prefetching
                    # (change requires restart)

# - Archiving -

#archive_mode = off        # enables archiving; off, on, or always
                # (change requires restart)
#archive_library = ''        # library to use to archive a WAL file
                # (empty string indicates archive_command should
                # be used)
#archive_command = ''        # command to use to archive a WAL file
                # placeholders: %p = path of file to archive
                #               %f = file name only
                # e.g. 'test ! -f /mnt/server/archivedir/%f && cp %p /mnt/server/archivedir/%f'
#archive_timeout = 0        # force a WAL file switch after this
                # number of seconds; 0 disables

# - Archive Recovery -

# These are only used in recovery mode.

#restore_command = ''        # command to use to restore an archived WAL file
                # placeholders: %p = path of file to restore
                #               %f = file name only
                # e.g. 'cp /mnt/server/archivedir/%f %p'
#archive_cleanup_command = ''    # command to execute at every restartpoint
#recovery_end_command = ''    # command to execute at completion of recovery

# - Recovery Target -

# Set these only when performing a targeted recovery.

#recovery_target = ''        # 'immediate' to end recovery as soon as a
                                # consistent state is reached
                # (change requires restart)
#recovery_target_name = ''    # the named restore point to which recovery will proceed
                # (change requires restart)
#recovery_target_time = ''    # the time stamp up to which recovery will proceed
                # (change requires restart)
#recovery_target_xid = ''    # the transaction ID up to which recovery will proceed
                # (change requires restart)
#recovery_target_lsn = ''    # the WAL LSN up to which recovery will proceed
                # (change requires restart)
#recovery_target_inclusive = on # Specifies whether to stop:
                # just after the specified recovery target (on)
                # just before the recovery target (off)
                # (change requires restart)
#recovery_target_timeline = 'latest'    # 'current', 'latest', or timeline ID
                # (change requires restart)
#recovery_target_action = 'pause'    # 'pause', 'promote', 'shutdown'
                # (change requires restart)


#------------------------------------------------------------------------------
# REPLICATION
#------------------------------------------------------------------------------

# - Sending Servers -

# Set these on the primary and on any standby that will send replication data.

#max_wal_senders = 10        # max number of walsender processes
                # (change requires restart)
#max_replication_slots = 10    # max number of replication slots
                # (change requires restart)
#wal_keep_size = 0        # in megabytes; 0 disables
#max_slot_wal_keep_size = -1    # in megabytes; -1 disables
#wal_sender_timeout = 60s    # in milliseconds; 0 disables
#track_commit_timestamp = off    # collect timestamp of transaction commit
                # (change requires restart)

# - Primary Server -

# These settings are ignored on a standby server.

#synchronous_standby_names = ''    # standby servers that provide sync rep
                # method to choose sync standbys, number of sync standbys,
                # and comma-separated list of application_name
                # from standby(s); '*' = all

# - Standby Servers -

# These settings are ignored on a primary server.

#primary_conninfo = ''            # connection string to sending server
#primary_slot_name = ''            # replication slot on sending server
#hot_standby = on            # "off" disallows queries during recovery
                    # (change requires restart)
#max_standby_archive_delay = 30s    # max delay before canceling queries
                    # when reading WAL from archive;
                    # -1 allows indefinite delay
#max_standby_streaming_delay = 30s    # max delay before canceling queries
                    # when reading streaming WAL;
                    # -1 allows indefinite delay
#wal_receiver_create_temp_slot = off    # create temp slot if primary_slot_name
                    # is not set
#wal_receiver_status_interval = 10s    # send replies at least this often
                    # 0 disables
#hot_standby_feedback = off        # send info from standby to prevent
                    # query conflicts
#wal_receiver_timeout = 60s        # time that receiver waits for
                    # communication from primary
                    # in milliseconds; 0 disables
#wal_retrieve_retry_interval = 5s    # time to wait before retrying to
                    # retrieve WAL after a failed attempt
#recovery_min_apply_delay = 0        # minimum delay for applying changes during recovery

# - Subscribers -

# These settings are ignored on a publisher.

#max_logical_replication_workers = 4    # taken from max_worker_processes
                    # (change requires restart)
#max_sync_workers_per_subscription = 2    # taken from max_logical_replication_workers
#max_parallel_apply_workers_per_subscription = 2    # taken from max_logical_replication_workers


#------------------------------------------------------------------------------
# QUERY TUNING
#------------------------------------------------------------------------------

# - Planner Method Configuration -

#enable_async_append = on
#enable_bitmapscan = on
#enable_gathermerge = on
#enable_hashagg = on
#enable_hashjoin = on
#enable_incremental_sort = on
#enable_indexscan = on
#enable_indexonlyscan = on
#enable_material = on
#enable_memoize = on
#enable_mergejoin = on
#enable_nestloop = on
#enable_parallel_append = on
#enable_parallel_hash = on
#enable_partition_pruning = on
#enable_partitionwise_join = off
#enable_partitionwise_aggregate = off
#enable_presorted_aggregate = on
#enable_seqscan = on
#enable_sort = on
#enable_tidscan = on

# - Planner Cost Constants -

#seq_page_cost = 1.0            # measured on an arbitrary scale
#random_page_cost = 4.0            # same scale as above
#cpu_tuple_cost = 0.01            # same scale as above
#cpu_index_tuple_cost = 0.005        # same scale as above
#cpu_operator_cost = 0.0025        # same scale as above
#parallel_setup_cost = 1000.0    # same scale as above
#parallel_tuple_cost = 0.1        # same scale as above
#min_parallel_table_scan_size = 8MB
#min_parallel_index_scan_size = 512kB
#effective_cache_size = 4GB

#jit_above_cost = 100000        # perform JIT compilation if available
                    # and query more expensive than this;
                    # -1 disables
#jit_inline_above_cost = 500000        # inline small functions if query is
                    # more expensive than this; -1 disables
#jit_optimize_above_cost = 500000    # use expensive JIT optimizations if
                    # query is more expensive than this;
                    # -1 disables

# - Genetic Query Optimizer -

#geqo = on
#geqo_threshold = 12
#geqo_effort = 5            # range 1-10
#geqo_pool_size = 0            # selects default based on effort
#geqo_generations = 0            # selects default based on effort
#geqo_selection_bias = 2.0        # range 1.5-2.0
#geqo_seed = 0.0            # range 0.0-1.0

#enable_appendorpath = on

# - Other Planner Options -

#default_statistics_target = 100    # range 1-10000
#constraint_exclusion = partition    # on, off, or partition
#cursor_tuple_fraction = 0.1        # range 0.0-1.0
#from_collapse_limit = 8
#jit = on                # allow JIT compilation
#join_collapse_limit = 8        # 1 disables collapsing of explicit
                    # JOIN clauses
#plan_cache_mode = auto            # auto, force_generic_plan or
                    # force_custom_plan
#recursive_worktable_factor = 10.0    # range 0.001-1000000


#------------------------------------------------------------------------------
# REPORTING AND LOGGING
#------------------------------------------------------------------------------

# - Where to Log -

#log_destination = 'stderr'        # Valid values are combinations of
                    # stderr, csvlog, jsonlog, syslog, and
                    # eventlog, depending on platform.
                    # csvlog and jsonlog require
                    # logging_collector to be on.

# This is used when logging to stderr:
logging_collector = on # Enable capturing of stderr, jsonlog,
                    # and csvlog into log files. Required
                    # to be on for csvlogs and jsonlogs.
                    # (change requires restart)

# These are only used if logging_collector is on:
log_directory = 'logss'            # directory where log files are written,
                    # can be absolute or relative to PGDATA
log_filename = 'postgresql-%Y-%m-%d_%H%M%S.log'    # log file name pattern,
                    # can include strftime() escapes
#log_file_mode = 0600            # creation mode for log files,
                    # begin with 0 to use octal notation
#log_rotation_age = 1d            # Automatic rotation of logfiles will
                    # happen after that time.  0 disables.
#log_rotation_size = 10MB        # Automatic rotation of logfiles will
                    # happen after that much log output.
                    # 0 disables.
#log_truncate_on_rotation = off        # If on, an existing log file with the
                    # same name as the new log file will be
                    # truncated rather than appended to.
                    # But such truncation only occurs on
                    # time-driven rotation, not on restarts
                    # or size-driven rotation.  Default is
                    # off, meaning append to existing files
                    # in all cases.

# These are relevant when logging to syslog:
#syslog_facility = 'LOCAL0'
#syslog_ident = 'postgres'
#syslog_sequence_numbers = on
#syslog_split_messages = on

# This is only relevant when logging to eventlog (Windows):
# (change requires restart)
#event_source = 'PostgreSQL'

# - When to Log -

#log_min_messages = warning        # values in order of decreasing detail:
                    #   debug5
                    #   debug4
                    #   debug3
                    #   debug2
                    #   debug1
                    #   info
                    #   notice
                    #   warning
                    #   error
                    #   log
                    #   fatal
                    #   panic

#log_min_error_statement = error    # values in order of decreasing detail:
                    #   debug5
                    #   debug4
                    #   debug3
                    #   debug2
                    #   debug1
                    #   info
                    #   notice
                    #   warning
                    #   error
                    #   log
                    #   fatal
                    #   panic (effectively off)

log_min_duration_statement = 5000    # -1 is disabled, 0 logs all statements
                    # and their durations, > 0 logs only
                    # statements running at least this number
                    # of milliseconds

#log_min_duration_sample = -1        # -1 is disabled, 0 logs a sample of statements
                    # and their durations, > 0 logs only a sample of
                    # statements running at least this number
                    # of milliseconds;
                    # sample fraction is determined by log_statement_sample_rate

#log_statement_sample_rate = 1.0    # fraction of logged statements exceeding
                    # log_min_duration_sample to be logged;
                    # 1.0 logs all such statements, 0.0 never logs


#log_transaction_sample_rate = 0.0    # fraction of transactions whose statements
                    # are logged regardless of their duration; 1.0 logs all
                    # statements from all transactions, 0.0 never logs

#log_startup_progress_interval = 10s    # Time between progress updates for
                    # long-running startup operations.
                    # 0 disables the feature, > 0 indicates
                    # the interval in milliseconds.

# - What to Log -

#debug_print_parse = off
#debug_print_rewritten = off
#debug_print_plan = off
#debug_pretty_print = on
#log_autovacuum_min_duration = 10min    # log autovacuum activity;
                    # -1 disables, 0 logs all actions and
                    # their durations, > 0 logs only
                    # actions running at least this number
                    # of milliseconds.
#log_checkpoints = on
#log_connections = off
#log_disconnections = off
log_duration = on
#log_error_verbosity = default        # terse, default, or verbose messages
#log_hostname = off
#log_line_prefix = '%m [%p] '        # special values:
                    #   %a = application name
                    #   %u = user name
                    #   %d = database name
                    #   %r = remote host and port
                    #   %h = remote host
                    #   %b = backend type
                    #   %p = process ID
                    #   %P = process ID of parallel group leader
                    #   %t = timestamp without milliseconds
                    #   %m = timestamp with milliseconds
                    #   %n = timestamp with milliseconds (as a Unix epoch)
                    #   %Q = query ID (0 if none or not computed)
                    #   %i = command tag
                    #   %e = SQL state
                    #   %c = session ID
                    #   %l = session line number
                    #   %s = session start timestamp
                    #   %v = virtual transaction ID
                    #   %x = transaction ID (0 if none)
                    #   %q = stop here in non-session
                    #        processes
                    #   %% = '%'
                    # e.g. '<%u%%%d> '
#log_lock_waits = off            # log lock waits >= deadlock_timeout
#log_recovery_conflict_waits = off    # log standby recovery conflict waits
                    # >= deadlock_timeout
#log_parameter_max_length = -1        # when logging statements, limit logged
                    # bind-parameter values to N bytes;
                    # -1 means print in full, 0 disables
#log_parameter_max_length_on_error = 0    # when logging an error, limit logged
                    # bind-parameter values to N bytes;
                    # -1 means print in full, 0 disables
log_statement = 'all'            # none, ddl, mod, all
#log_replication_commands = off
#log_temp_files = -1            # log temporary files equal or larger
                    # than the specified size in kilobytes;
                    # -1 disables, 0 logs all temp files
log_timezone = 'Asia/Yekaterinburg'

# - Process Title -

#cluster_name = ''            # added to process titles if nonempty
                    # (change requires restart)
#update_process_title = on


#------------------------------------------------------------------------------
# STATISTICS
#------------------------------------------------------------------------------

# - Cumulative Query and Index Statistics -

#track_activities = on
#track_activity_query_size = 1024    # (change requires restart)
#track_counts = on
#track_io_timing = off
#track_wal_io_timing = off
#track_functions = none            # none, pl, all
#stats_fetch_consistency = cache    # cache, none, snapshot


# - Monitoring -

#compute_query_id = auto
#log_statement_stats = off
#log_parser_stats = off
#log_planner_stats = off
#log_executor_stats = off


#------------------------------------------------------------------------------
# AUTOVACUUM
#------------------------------------------------------------------------------

#autovacuum = on            # Enable autovacuum subprocess?  'on'
                    # requires track_counts to also be on.
#autovacuum_max_workers = 3        # max number of autovacuum subprocesses
                    # (change requires restart)
#autovacuum_naptime = 1min        # time between autovacuum runs
#autovacuum_vacuum_threshold = 50    # min number of row updates before
                    # vacuum
#autovacuum_vacuum_insert_threshold = 1000    # min number of row inserts
                    # before vacuum; -1 disables insert
                    # vacuums
#autovacuum_analyze_threshold = 50    # min number of row updates before
                    # analyze
#autovacuum_vacuum_scale_factor = 0.2    # fraction of table size before vacuum
#autovacuum_vacuum_insert_scale_factor = 0.2    # fraction of inserts over table
                    # size before insert vacuum
#autovacuum_analyze_scale_factor = 0.1    # fraction of table size before analyze
#autovacuum_freeze_max_age = 200000000    # maximum XID age before forced vacuum
                    # (change requires restart)
#autovacuum_multixact_freeze_max_age = 400000000    # maximum multixact age
                    # before forced vacuum
                    # (change requires restart)
#autovacuum_vacuum_cost_delay = 2ms    # default vacuum cost delay for
                    # autovacuum, in milliseconds;
                    # -1 means use vacuum_cost_delay
#autovacuum_vacuum_cost_limit = -1    # default vacuum cost limit for
                    # autovacuum, -1 means use
                    # vacuum_cost_limit


#------------------------------------------------------------------------------
# CLIENT CONNECTION DEFAULTS
#------------------------------------------------------------------------------

# - Statement Behavior -

#client_min_messages = notice        # values in order of decreasing detail:
                    #   debug5
                    #   debug4
                    #   debug3
                    #   debug2
                    #   debug1
                    #   log
                    #   notice
                    #   warning
                    #   error
#search_path = '"$user", public'    # schema names
#row_security = on
#default_table_access_method = 'heap'
#default_tablespace = ''        # a tablespace name, '' uses the default
#default_toast_compression = 'pglz'    # 'pglz' or 'lz4'
#temp_tablespaces = ''            # a list of tablespace names, '' uses
                    # only default tablespace
#check_function_bodies = on
#default_transaction_isolation = 'read committed'
#default_transaction_read_only = off
#default_transaction_deferrable = off
#session_replication_role = 'origin'
#statement_timeout = 0            # in milliseconds, 0 is disabled
#lock_timeout = 0            # in milliseconds, 0 is disabled
#idle_in_transaction_session_timeout = 0    # in milliseconds, 0 is disabled
#idle_session_timeout = 0        # in milliseconds, 0 is disabled
#vacuum_freeze_table_age = 150000000
#vacuum_freeze_min_age = 50000000
#vacuum_failsafe_age = 1600000000
#vacuum_multixact_freeze_table_age = 150000000
#vacuum_multixact_freeze_min_age = 5000000
#vacuum_multixact_failsafe_age = 1600000000
#bytea_output = 'hex'            # hex, escape
#xmlbinary = 'base64'
#xmloption = 'content'
#gin_pending_list_limit = 4MB
#createrole_self_grant = ''        # set and/or inherit

# - Locale and Formatting -

datestyle = 'iso, dmy'
#intervalstyle = 'postgres'
timezone = 'Asia/Yekaterinburg'
#timezone_abbreviations = 'Default'     # Select the set of available time zone
                    # abbreviations.  Currently, there are
                    #   Default
                    #   Australia (historical usage)
                    #   India
                    # You can create your own file in
                    # share/timezonesets/.
#extra_float_digits = 1            # min -15, max 3; any value >0 actually
                    # selects precise output mode
#client_encoding = sql_ascii        # actually, defaults to database
                    # encoding

# These settings are initialized by initdb, but they can be changed.
lc_messages = 'ru_RU.UTF-8'        # locale for system error message
                    # strings
lc_monetary = 'ru_RU.UTF-8'        # locale for monetary formatting
lc_numeric = 'ru_RU.UTF-8'        # locale for number formatting
lc_time = 'ru_RU.UTF-8'            # locale for time formatting

#icu_validation_level = warning        # report ICU locale validation
                    # errors at the given level

# default configuration for text search
default_text_search_config = 'pg_catalog.russian'

# - Shared Library Preloading -

#local_preload_libraries = ''
#session_preload_libraries = ''
#shared_preload_libraries = ''    # (change requires restart)
#jit_provider = 'llvmjit'        # JIT library to use

# - Other Defaults -

#dynamic_library_path = '$libdir'
#gin_fuzzy_search_limit = 0


#------------------------------------------------------------------------------
# LOCK MANAGEMENT
#------------------------------------------------------------------------------

#deadlock_timeout = 1s
#max_locks_per_transaction = 64        # min 10
                    # (change requires restart)
#max_pred_locks_per_transaction = 64    # min 10
                    # (change requires restart)
#max_pred_locks_per_relation = -2    # negative values mean
                    # (max_pred_locks_per_transaction
                    #  / -max_pred_locks_per_relation) - 1
#max_pred_locks_per_page = 2            # min 0


#------------------------------------------------------------------------------
# VERSION AND PLATFORM COMPATIBILITY
#------------------------------------------------------------------------------

# - Previous PostgreSQL Versions -

#array_nulls = on
#backslash_quote = safe_encoding    # on, off, or safe_encoding
#escape_string_warning = on
#lo_compat_privileges = off
#quote_all_identifiers = off
#standard_conforming_strings = on
#synchronize_seqscans = on

# - Other Platforms and Clients -

#transform_null_equals = off


#------------------------------------------------------------------------------
# ERROR HANDLING
#------------------------------------------------------------------------------

#exit_on_error = off            # terminate session on any error?
#restart_after_crash = on        # reinitialize after backend crash?
#data_sync_retry = off            # retry or panic on failure to fsync
                    # data?
                    # (change requires restart)
#recovery_init_sync_method = fsync    # fsync, syncfs (Linux 5.8+)


#------------------------------------------------------------------------------
# CONFIG FILE INCLUDES
#------------------------------------------------------------------------------

# These options allow settings to be loaded from files other than the
# default postgresql.conf.  Note that these are directives, not variable
# assignments, so they can usefully be given more than once.

#include_dir = '...'            # include files ending in '.conf' from
                    # a directory, e.g., 'conf.d'
#include_if_exists = '...'        # include file only if it exists
#include = '...'            # include file


#------------------------------------------------------------------------------
# CUSTOMIZED OPTIONS
#------------------------------------------------------------------------------

# Add settings for extensions here
listen_addresses = '*'
shared_buffers = 64359MB # 25% of RAM
temp_buffers = 128MB
max_files_per_process = 10000
max_parallel_workers_per_gather = 0
max_parallel_maintenance_workers = 6 # Количество CPU/4, минимум 2, максимум 6
commit_delay = 1000
max_wal_size = 40GB
min_wal_size = 10GB
checkpoint_timeout = 15min
effective_cache_size = 193077MB # 75% of RAM
work_mem=512MB
maintenance_work_mem = 1024MB
from_collapse_limit = 8
join_collapse_limit = 8
autovacuum_max_workers = 16 # Количество CPU/2, минимум 2
vacuum_cost_limit = 1600 # 100* autovacuum_max_workers
autovacuum_naptime = 20s
autovacuum_vacuum_scale_factor = 0.01
autovacuum_analyze_scale_factor = 0.005
max_locks_per_transaction = 256
escape_string_warning = off
standard_conforming_strings = off
shared_preload_libraries = 'online_analyze, plantuner'
online_analyze.threshold = 50
online_analyze.scale_factor = 0.1
online_analyze.enable = on
online_analyze.verbose = off
online_analyze.min_interval = 10000
online_analyze.table_type = 'temporary'
plantuner.fix_empty_table = on

А железо?

2 проца по 16 потоков
256 гигов оперативы
рейд из 2 ssd samsung nvme на 3.2 ТБ суммарно

Есть подозрения что затык в дисковой системе

#random_page_cost = 4.0            # same scale as above


https://its.1c.ru/db/metod8dev/content/5866/hdoc

(69)+ для NVMe смело ставь
random_page_cost = 1.1

(67) 2 проца? Возможно стандартное ядро не подойдет.
В архитектуре процессоры разбиты на 2 ноды?
Отключите 1 процессор и протестируйте снова

(71) Ну судя по синтетическим тестам, система корректно работает с процессорами, тот же гилев например, выдает нормальный результат. графики нагрузки, тоже отображают результат, аналогичный тестам на windows server

Включить(временно) auto_explain
shared_preload_libraries = 'online_analyze, plantuner, auto_exlplain'

auto_explain.log_min_duration = '10s'
auto_explain.log_analyze = true

Посмотреть что в плане запроса у которого duration: 12905.457 ms

(72) Так синтетика не подразумевает нагрузку на оперативку. В многопроцессорных архитектурах память самое сложное в настройке.

Попробуй поиграться с параметром effective_cache_size
Сталкивался, что иногда его надо ставить значительно меньше

(0) Может планировщик работы процессора неверно работает...
Попробуй поменять с помощью cpufreq-set.
А процы точно на максимуме работают? Помониторь.

#!/bin/bash
watch -n1 "cat /proc/cpuinfo | grep \"MHz\""

Кстати, а analyze на базе запускали?

/tmp как смонтирован? памяти у тебя много, монтируй в tmpfs, кроме повышения скорости еще и износ дисков значительно уменьшится.
stats_temp_directory тоже стоит на tmpfs перевести, ну потеряется что-то при перезагрузке, ну и хрен с ним, не уверен насчет 15-16 PGSQL, но 14й в этот каталог пишет  очень активно.
Немного прироста еще дает указание сервера не как localhost, а как /tmp Но это всего 5% наверное, хотя даже столько уже хорошо.

Читайте - наслаждайтесь
https://habr.com/ru/articles/505108/
https://wiki.astralinux.ru/tandocs/nastrojka-postgresql-tantor-dlya-raboty-1s-294394904.html#:~:text=kernel.numa_balancing%3D0,%D0%B5%D0%B5%20%D0%BD%D0%B5%20%D0%BF%D0%BE%D0%B4%D0%B4%D0%B5%D1%80%D0%B6%D0%B8%D0%B2%D0%B0%D0%B5%D1%82.

(77) Это все проверять надо, причем последовательно, это очень долго...
У меня на 14 PGSQL включение geqo, которое должно вроде бы ускорять запросы, замедляет проведение раз в 10.
Ага, а вдруг он в 15 включен по-умолчанию? (0) а если попровать geqo=off

(76)
А суть этого в чем?
Вместо через сеть начинает как то иначе 1С с PgSQL общаться?

(78) Для начала достаточно NUMA отключить и посмотреть на результат.

(79) Да, так работает иначе. Это передача данных напрямую через сокет/память, забыл как в MSSQL это называется. Это работает, и это проверено.

(81) Почему именно /tmp? Возможно он много где использоваться может.
Может лучше сделать свой - уникальный симлинк?

(82) http://www.gilev.ru/unixsocket/

(83) Жутковатое объяснение... Вообще ни чёрта не понятно, слишком заумно! Но так это работает и работает быстрее.

online_analyze.enable = off

(83) ну так в сервисе системд или в конфе постгри лучше симлинк уникальным сделать
Интересно что будет если сделать
cp data.qwerty /tmp

(65) У вас shared_buffers дважды записан:

# - Memory -

shared_buffers = 128MB            # min 128kB

# CUSTOMIZED OPTIONS
#-----------------------------------------------------------

# Add settings for extensions here
listen_addresses = '*'
shared_buffers = 64359MB # 25% of RAM