upstream prematurely closed connection

BaHbka bahbka at gmail.com
Thu Jul 17 22:00:56 MSD 2008


Здравствуйте.

Имеются 1 фронтэнд, 4 бэкенда и 2 сервера mysql
Никак не могу справиться с одной проблемой: периодически в логе ngnix'а появляются строки следующего вида

2008/07/17 14:34:40 [error] 14415#0: *3460073 upstream prematurely closed connection while reading response header from upstream, ... upstream: "fastcgi://192.168.0.1:9000"
2008/07/17 14:34:40 [error] 14415#0: *3460070 upstream prematurely closed connection while reading response header from upstream, ... upstream: "fastcgi://192.168.0.3:9000"
2008/07/17 14:34:40 [error] 14415#0: *3460069 upstream prematurely closed connection while reading response header from upstream, ... upstream: "fastcgi://192.168.0.4:9000"
2008/07/17 14:34:40 [error] 14410#0: *3460247 upstream prematurely closed connection while reading response header from upstream, ... upstream: "fastcgi://192.168.0.4:9000"

после чего зачастую проскакивает no live upstreams while connecting to upstream

иногда бывают такие:

2008/07/17 14:54:21 [error] 14412#0: *3760898 kevent() reported about an closed connection (54: Connection reset by peer) while reading response header from upstream, ... upstream: "fastcgi://192.168.0.2:9000"
2008/07/17 14:54:21 [error] 14412#0: *3760900 kevent() reported about an closed connection (54: Connection reset by peer) while reading response header from upstream, ... upstream: "fastcgi://192.168.0.3:9000"
2008/07/17 14:54:21 [error] 14412#0: *3746025 kevent() reported about an closed connection (54: Connection reset by peer) while reading response header from upstream, ... upstream: "fastcgi://192.168.0.3:9000"

Причем появляются эти строчки пачками в одну секунду штук по 5-50 с периодичностью раз в 5-10 минут (в зависимости от нагрузки), в основном со всех бэкендов.
Утром например, когда кол-во посетителей небольшое этого не происходит.
Причем в реквесте часто бывают скрипты, которые вообще не используют mysql. В одном из таких тестовых скриптов вообще пхпшного кода нет (скрипт из одной строчки вида <input id=asd value=1>) :) Но бэкенд все равно разрывает соединение.
Я думаю (но могу ошибаться) mysql тут ни при чем, а раз в одну секунду разрывается соединение со всеми бэками, то и они ни при чем.
Подозреваю, что клиенты в этот момент получают 502-ю, т.к. сам бывало на нее нарывался.

В дебаг логе nginx'а следующее (не уверен, все ли что нужно захватил)

2008/07/16 16:17:52 [debug] 35120#0: *405143 http init upstream, client timer: 0
2008/07/16 16:17:52 [debug] 35120#0: *405143 kevent set event: 534: ft:-2 fl:0025
2008/07/16 16:17:52 [debug] 35120#0: *405143 http script copy: ""
...
2008/07/16 16:17:52 [debug] 35120#0: *405143 fastcgi param: "X_FORWARDED_FOR: 123.123.123.123"
2008/07/16 16:17:52 [debug] 35120#0: *405143 http script copy: ""
2008/07/16 16:17:52 [debug] 35120#0: *405143 fastcgi param: "SSL_CLIENT_SERIAL: 0"
2008/07/16 16:17:52 [debug] 35120#0: *405143 http cleanup add: 00000000006EDF40
2008/07/16 16:17:52 [debug] 35120#0: *405143 get rr peer, try: 4
2008/07/16 16:17:52 [debug] 35120#0: *405143 get rr peer, current: 0 2
2008/07/16 16:17:52 [debug] 35120#0: *405143 socket 237
2008/07/16 16:17:52 [debug] 35120#0: *405143 connect to 192.168.0.2:9000, fd:237 #405294
2008/07/16 16:17:52 [debug] 35120#0: *405143 kevent set event: 237: ft:-1 fl:0025
2008/07/16 16:17:52 [debug] 35120#0: *405143 kevent set event: 237: ft:-2 fl:0025
2008/07/16 16:17:52 [debug] 35120#0: *405143 http upstream connect: -2
2008/07/16 16:17:52 [debug] 35120#0: *405143 event timer add: 237: 75000:1216210747795
2008/07/16 16:17:58 [debug] 35120#0: *405143 http upstream check client, write event:1, "/blablabla.php"
2008/07/16 16:17:58 [debug] 35120#0: *405143 http upstream process header
2008/07/16 16:17:58 [debug] 35120#0: *405143 malloc: 000000000110E000:32768
2008/07/16 16:17:58 [debug] 35120#0: *405143 recv: eof:1, avail:0, err:0
2008/07/16 16:17:58 [error] 35120#0: *405143 upstream prematurely closed connection while reading response header from upstream, .....
2008/07/16 16:17:58 [debug] 35120#0: *405143 http next upstream, 2
2008/07/16 16:17:58 [debug] 35120#0: *405143 free rr peer 4 4
2008/07/16 16:17:58 [debug] 35120#0: *405143 free rr peer failed: 0 -5
2008/07/16 16:17:58 [debug] 35120#0: *405143 close http upstream connection: 237
2008/07/16 16:17:58 [debug] 35120#0: *405143 event timer del: 237: 1216210747795
2008/07/16 16:17:58 [debug] 35120#0: *405143 get rr peer, try: 3
2008/07/16 16:17:58 [debug] 35120#0: *405143 socket 140
2008/07/16 16:17:58 [debug] 35120#0: *405143 connect to 192.168.0.3:9000, fd:140 #406707
2008/07/16 16:17:58 [debug] 35120#0: *405143 kevent set event: 140: ft:-1 fl:0025
2008/07/16 16:17:58 [debug] 35120#0: *405143 kevent set event: 140: ft:-2 fl:0025
2008/07/16 16:17:58 [debug] 35120#0: *405143 http upstream connect: -2

или вот, например

2008/07/16 16:17:49 [debug] 35117#0: *404981 connect to 192.168.0.3:9000, fd:560 #405020
2008/07/16 16:17:49 [debug] 35117#0: *404981 kevent set event: 560: ft:-1 fl:0025
2008/07/16 16:17:49 [debug] 35117#0: *404981 kevent set event: 560: ft:-2 fl:0025
2008/07/16 16:17:49 [debug] 35117#0: *404981 http upstream connect: -2
2008/07/16 16:17:49 [debug] 35117#0: *404981 event timer add: 560: 75000:1216210744998
2008/07/16 16:17:59 [debug] 35117#0: *404981 post event 0000000000DA58A0
2008/07/16 16:17:59 [debug] 35117#0: *404981 post event 0000000000AFAA20
2008/07/16 16:17:59 [debug] 35117#0: *404981 post event 0000000000DA6A20
2008/07/16 16:17:59 [debug] 35117#0: *404981 delete posted event 0000000000DA6A20
2008/07/16 16:17:59 [debug] 35117#0: *404981 http upstream send request handler
2008/07/16 16:17:59 [debug] 35117#0: *404981 http upstream send request
2008/07/16 16:17:59 [debug] 35117#0: *404981 chain writer buf fl:0 s:1320
2008/07/16 16:17:59 [debug] 35117#0: *404981 chain writer in: 000000000106EF68
2008/07/16 16:17:59 [debug] 35117#0: *404981 writev: 1320 of 1320
2008/07/16 16:17:59 [debug] 35117#0: *404981 chain writer out: 0000000000000000
2008/07/16 16:17:59 [debug] 35117#0: *404981 event timer del: 560: 1216210744998
2008/07/16 16:17:59 [debug] 35117#0: *404981 event timer add: 560: 180000:1216210859644
2008/07/16 16:17:59 [debug] 35117#0: *404981 http upstream process header
2008/07/16 16:17:59 [debug] 35117#0: *404981 malloc: 00000000006E9000:32768
2008/07/16 16:17:59 [debug] 35117#0: *404981 recv: eof:1, avail:0, err:0
2008/07/16 16:17:59 [error] 35117#0: *404981 upstream prematurely closed connection while reading response header from upstream, .....
2008/07/16 16:17:59 [debug] 35117#0: *404981 http next upstream, 2
2008/07/16 16:17:59 [debug] 35117#0: *404981 free rr peer 4 4
2008/07/16 16:17:59 [debug] 35117#0: *404981 free rr peer failed: 2 -11
2008/07/16 16:17:59 [debug] 35117#0: *404981 close http upstream connection: 560

ktrace'ом бэкендов ничего не смог выловить. Видимо просто не умею его готовить :)

Схема следующая:
---
один фронтэнд - FreeBSD 6.2-RELEASE-p7 amd64
nginx 0.6.32 (до этого был nginx 0.5.34 - ситуация такая же) ставился из портов

netstat -Lan
tcp4  0/0/4096       123.123.123.123.80

netstat -m
6652/9503/16155 mbufs in use (current/cache/total)
2249/7867/10116/32768 mbuf clusters in use (current/cache/total/max)
2231/6351 mbuf+clusters out of packet secondary zone in use (current/cache)
6161K/18109K/24270K bytes allocated to network (current/cache/total)

ngnix.conf:

user                    www www;
worker_processes        8;

...

events {
         worker_connections  25000;
         use kqueue;
}

http {
...
         client_header_timeout           1m;
         client_body_timeout             1m;
         send_timeout                    1m;

         client_header_buffer_size       32k;
         large_client_header_buffers     16 16k;

         gzip                            on;
         gzip_min_length                 1100;
         gzip_buffers                    4 8k;
         gzip_types                      text/plain text/css application/x-javascript;

         output_buffers                  1 32k;
         postpone_output                 1460;

         sendfile                        on;
         sendfile_max_chunk              128k;
         tcp_nopush                      on;
         tcp_nodelay                     on;
         send_lowat                      12000;

         keepalive_timeout               75 20;

         fastcgi_hide_header             x-powered-by;
         fastcgi_ignore_client_abort     off;
         fastcgi_next_upstream           error;

         upstream backend
         {
                 server  192.168.0.1:9000        weight=10;
                 server  192.168.0.2:9000        weight=10;
                 server  192.168.0.3:9000        weight=10;
                 server  192.168.0.4:9000        weight=10;
         }

         server {

                 listen                  123.123.123.123:80;
                 server_name             www.***.ru ***.ru;
                 error_log               /var/log/nginx/nginx-error.log  error;

                 client_body_buffer_size 1024k;
                 client_max_body_size    4m;

...

                location ~ \.php$
                 {
                         access_log                              off;
                         error_log                               /var/log/nginx/php-error.log error;
                         fastcgi_pass                            backend;
                         fastcgi_index                           index.php;
                         fastcgi_connect_timeout                 75;
                         fastcgi_send_timeout                    180;
                         fastcgi_read_timeout                    180;
                         fastcgi_buffer_size                     32k;
                         fastcgi_buffers                         8 256k;
                         fastcgi_busy_buffers_size               256k;
                         fastcgi_temp_file_write_size            256k;
                         fastcgi_param   SCRIPT_FILENAME         /var/www/www.***.ru/php/$fastcgi_script_name;
...

---
четыре бездисковых бэкенда FreeBSD 7.0-RELEASE-p1 amd64 (раньше было FreeBSD 6.2, php 5.2.4 без php-fpm патча - такая же ситуация)
php 5.2.6 + php-fpm 0.5.8-rc1

netstat -Lan
tcp4  0/0/4096       *.9000

netstat -m
262/3578/3840 mbufs in use (current/cache/total)
260/3624/3884/25600 mbuf clusters in use (current/cache/total/max)
260/2812 mbuf+clusters out of packet secondary zone in use (current/cache)
585K/9030K/9616K bytes allocated to
  network (current/cache/total)

php-fpm.conf:
...
                 <section name="pool">
                         <value name="name">first</value>
                         <value name="listen_address">0.0.0.0:9000</value>
...
                                 <value name="max_children">100</value>
...
                         <value name="request_execution_timeout">601s</value>
                         <value name="rlimit_files">2048</value>
                         <value name="rlimit_core">0</value>
...
                         <value name="max_requests">500</value>
...

Где копать? На фронтэнде или все таки бэкенды выеживаются?

Заранее благодарю :)






More information about the nginx-ru mailing list