菜单

大量LAST_ACK 分析过程

2011年09月9日 - apache

分析的很仔细,并且我在实际中用到了~ 转过来保留
net.ipv4.tcp_fin_timeout=30
net.ipv4.tcp_max_syn_backlog = 65536
net.core.netdev_max_backlog = 32768

net.core.somaxconn = 32768
net.core.wmem_default = 8388608
net.core.rmem_default = 8388608
net.core.rmem_max = 16777216
net.core.wmem_max = 16777216
net.ipv4.tcp_timestamps = 0
net.ipv4.tcp_synack_retries = 2
net.ipv4.tcp_syn_retries = 2
net.ipv4.tcp_tw_recycle = 1
#net.ipv4.tcp_tw_len = 1
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_mem = 94500000 915000000 927000000
net.ipv4.tcp_max_orphans = 32768
net.ipv4.ip_local_port_range = 1024  65535
以下解决了LAST_ACK过多的问题
# 当keepalive打开的情况下,TCP发送keepalive消息的频率。(默认值是7200(2小时),服务器建议设为1800)
net.ipv4.tcp_keepalive_time = 1800
# 在近端丢弃TCP连接之前﹐要进行多少次重试。默认值是7个﹐在大负载服务器上建议调整为3)
net.ipv4.tcp_orphan_retries = 3
# 探测消息发送的频率,乘以tcp_keepalive_probes就得到对于从开始探测以来没有响应的连接杀除的时间。(默认值为75秒,推荐设为15秒)
net.ipv4.tcp_keepalive_intvl = 15优化前:             优化后:
LAST_ACK 374         LAST_ACK 50
SYN_RECV 91          SYN_RECV 33
ESTABLISHED 216      ESTABLISHED 185
FIN_WAIT1 63        FIN_WAIT1 8 

——————————————————————————————————————————————–

现象:在netstat的时候发现大量处于LAST_ACK状态的TCP连接,达到在ESTABLISHED状态的90%以上 [root@ccsafe ~]# netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c
6 CLOSE_WAIT
7 CLOSING
6838 ESTABLISHED
1037 FIN_WAIT1
357 FIN_WAIT2
5830 LAST_ACK
2 LISTEN
276 SYN_RECV
71 TIME_WAIT
[root@ccsafe ~]#
看看系统 状态,性能 都花在系统中断和上下文切换
[root@ccsafe ~]# vmstat 2
procs ———–memory———- —swap– —–io—- –system– —–cpu——
r b    swpd    free    buff cache     si    so     bi     bo    in    cs us sy id wa st
1 0       0 3091812 363032 284132     0     0      0      0     1     1 0 0 100 0 0
0 0       0 3091812 363032 284132     0     0      0      0 13750 3174 0 5 94 0 0
0 0       0 3091936 363032 284132     0     0      0      0 13666 3057 1 5 94 0 0
0 0       0 3092060 363032 284132     0     0      0     16 13749 3030 0 5 95 0 0
0 0       0 3092060 363032 284132     0     0      0      0 13822 3144 0 5 95 0 0
0 0       0 3092060 363032 284132     0     0      0      0 13390 2961 0 5 95 0 0
0 0       0 3092060 363032 284132     0     0      0      0 13541 3182 0 6 94 0 0

查看socket队列信息
[root@ccsafe ~]# sar -n SOCK 5
Linux 2.6.18-53.1.13.el5PAE (ccsafe)      10/21/2008
06:31:43 PM     totsck     tcpsck     udpsck     rawsck    ip-frag     tcp-tw
06:31:48 PM       6951      13868          1          0          0        430
Average:          6951      13868          1          0          0        430
根据TCP状态的变化过程来分析,LAST_ACK属于被动关闭连接过程中的状态
ESTABLISHED->CLOSE_WAIT->(发送ACK)->LAST_ACK->(发送FIN+接收ACK)->CLOSED
现在状态都堆积到LAST_ACK,初步判断问题从上下两个状态着手
调节一下LAST_ACK时间…
[root@ccsafe ~]# sysctl -a |grep last_ack
net.ipv4.netfilter.ip_conntrack_tcp_timeout_last_ack = 30
[root@ccsafe ~]# sysctl -w net.ipv4.netfilter.ip_conntrack_tcp_timeout_last_ack=10
net.ipv4.netfilter.ip_conntrack_tcp_timeout_last_ack = 10
[root@ccsafe ~]# sysctl -p
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 5.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
6 CLOSE_WAIT
9 CLOSING
6420 ESTABLISHED
693 FIN_WAIT1
391 FIN_WAIT2
5081 LAST_ACK
2 LISTEN
203 SYN_RECV
66 TIME_WAIT
检查一下LAST_ACK所对应的应用
[root@ccsafe ~]# netstat -ant|fgrep “LAST_ACK”|cut -b 49-75|cut -d “:” -f1|sort |uniq -c|sort -nr –key=1,7|head -5
101 220.160.210.6
46 222.75.65.69
31 221.0.91.118
24 222.210.8.160
22 60.161.81.28
[root@ccsafe ~]#
[root@ccsafe ~]# netstat -an|grep “220.160.210.6”
tcp         0 17280 10.1.1.145:80            220.160.210.6:52787          ESTABLISHED
tcp         1 14401 10.1.1.145:80            220.160.210.6:52513          LAST_ACK
tcp         1 14401 10.1.1.145:80            220.160.210.6:52769          LAST_ACK
tcp         1 14401 10.1.1.145:80            220.160.210.6:52768          LAST_ACK
tcp         0    8184 10.1.1.145:80            220.160.210.6:52515          LAST_ACK
tcp         1 14401 10.1.1.145:80            220.160.210.6:52514          LAST_ACK
tcp         0    8184 10.1.1.145:80            220.160.210.6:52781          LAST_ACK

是TCP80端口的应用,调节一下nginx 的keepalive时间…
[root@ccsafe ~]# /usr/local/nginx/sbin/nginx -t -c /usr/local/nginx/conf/nginx.conf
2008/10/21 19:15:31 [info] 21352#0: the configuration file /usr/local/nginx/conf/nginx.conf syntax is ok
2008/10/21 19:15:31 [info] 21352#0: the configuration file /usr/local/nginx/conf/nginx.conf was tested successfully
[root@ccsafe ~]# ps aux|egrep ‘(PID|nginx)’
USER        PID %CPU %MEM     VSZ    RSS TTY       STAT START    TIME COMMAND
root       8290 0.0 0.0    7572 1124 ?         Ss    Oct04    0:00 nginx: master process /usr/local/nginx/sbin/nginx
nobody     8291 0.2 0.3 19704 13776 ?         S     Oct04 71:35 nginx: worker process
nobody     8292 0.3 0.2 17604 11680 ?         S     Oct04 77:26 nginx: worker process
nobody     8293 0.2 0.4 22528 16636 ?         S     Oct04 58:13 nginx: worker process
nobody     8294 0.3 0.4 24944 19020 ?         S     Oct04 94:07 nginx: worker process
nobody     8295 0.3 0.5 27496 21508 ?         S     Oct04 84:41 nginx: worker process
nobody     8296 0.3 0.1 13388 7496 ?         S     Oct04 84:14 nginx: worker process
nobody     8297 0.2 0.0    9196 3268 ?         S     Oct04 58:21 nginx: worker process
nobody     8298 0.3 0.2 15392 9504 ?         S     Oct04 75:16 nginx: worker process
root      21354 0.0 0.0    3896    720 pts/0     S+    19:15    0:00 egrep (PID|nginx)
(动态加载新配置)
[root@ccsafe ~]# kill -HUP 8290
[root@ccsafe ~]#
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90 |sort |uniq -c
1 CLOSE_WAIT
1138 CLOSING
7161 ESTABLISHED
1427 FIN_WAIT1
396 FIN_WAIT2
5740 LAST_ACK
2 LISTEN
350 SYN_RECV
148 TIME_WAIT

[root@ccsafe ~]# netstat -ant|fgrep “:”|cut -b 77-90 |sort |uniq -c
1151 CLOSING
8506 ESTABLISHED
1452 FIN_WAIT1
666 FIN_WAIT2
6568 LAST_ACK
2 LISTEN
429 SYN_RECV
92 TIME_WAIT

LAST_ACK不下,而且CLOSING 和FIN_WAIT突增
着重看看可影响主动断开TCP连接时几个参数
tcp_keepalive_intvl:探测消息发送的频率
tcp_keepalive_probes:TCP发送keepalive探测以确定该连接已经断开的次数
tcp_keepalive_time:当keepalive打开的情况下,TCP发送keepalive消息的频率
[root@ccsafe ~]# sysctl -a|grep tcp_keepalive
net.ipv4.tcp_keepalive_intvl = 30
net.ipv4.tcp_keepalive_probes = 2
net.ipv4.tcp_keepalive_time = 160
tcp_retries2:在丢弃激活(已建立通讯状况)的TCP连接之前﹐需要进行多少次重试
[root@ccsafe ~]# sysctl -a |grep tcp_retries
net.ipv4.tcp_retries2 = 15
net.ipv4.tcp_retries1 = 3
加速处理那些等待ACK的LAST_ACK,减少等待ACK的LAST_ACK的重试次数
[root@ccsafe ~]# sysctl -w net.ipv4.tcp_retries2=5
net.ipv4.tcp_retries2 = 5
减少keepalive发送的频率
[root@ccsafe ~]# sysctl -w net.ipv4.tcp_keepalive_intvl=15
net.ipv4.tcp_keepalive_intvl = 15
[root@ccsafe ~]# sysctl -p
排除syncookies的影响
[root@ccsafe ~]# echo “1” >/proc/sys/net/ipv4/tcp_syncookies
[root@ccsafe ~]# sysctl -a|grep tcp_keepalive
net.ipv4.tcp_keepalive_intvl = 30
net.ipv4.tcp_keepalive_probes = 2
net.ipv4.tcp_keepalive_time = 160
[root@ccsafe ~]# sysctl -a|grep syncookies
net.ipv4.tcp_syncookies = 1
延长keepalive检测周期,保留ESTABLISHED数量
[root@ccsafe ~]# echo “1800” >/proc/sys/net/ipv4/tcp_keepalive_time
[root@ccsafe ~]# echo “5” >/proc/sys/net/ipv4/tcp_keepalive_probes
[root@ccsafe ~]# echo “15” >/proc/sys/net/ipv4/tcp_keepalive_intvl
[root@ccsafe ~]# sysctl -a|grep tcp_keepalive
net.ipv4.tcp_keepalive_intvl = 15
net.ipv4.tcp_keepalive_probes = 5
net.ipv4.tcp_keepalive_time = 1800
[root@ccsafe ~]# !watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
1 CLOSE_WAIT
363 CLOSING
5145 ESTABLISHED
1073 FIN_WAIT1
174 FIN_WAIT2
6042 LAST_ACK
2 LISTEN
301 SYN_RECV
85 TIME_WAIT

LAST_ACK不下,但是CLOSING有所回落
tcp_orphan_retries:在近端丢弃TCP连接之前﹐要进行多少次重试。
[root@ccsafe ~]# sysctl -a|grep tcp_orphan
net.ipv4.tcp_orphan_retries = 0
关键,丢TCP太频繁了,以至于后勤都跟不上。设置 丢弃之前的重试次数
[root@ccsafe ~]# echo “3” >/proc/sys/net/ipv4/tcp_orphan_retries
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
1 CLOSE_WAIT
24 CLOSING
5422 ESTABLISHED
279 FIN_WAIT1
214 FIN_WAIT2
1966 LAST_ACK
2 LISTEN
269 SYN_RECV
74 TIME_WAIT
上下调节该值,找个合适的临界点
[root@ccsafe ~]# echo “7” >/proc/sys/net/ipv4/tcp_orphan_retries
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
1 CLOSE_WAIT
175 CLOSING
5373 ESTABLISHED
436 FIN_WAIT1
209 FIN_WAIT2
3184 LAST_ACK
2 LISTEN
283 SYN_RECV
110 TIME_WAIT
恢复,同时FIN_WAIT1的值过高。考虑减少tcp_fin_timeout时间
[root@ccsafe ~]# echo “2” >/proc/sys/net/ipv4/tcp_orphan_retries
[root@ccsafe ~]# sysctl -a|grep tcp_fin
net.ipv4.tcp_fin_timeout = 10
[root@ccsafe ~]# echo “5” >/proc/sys/net/ipv4/tcp_fin_timeout
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
2 CLOSE_WAIT
17 CLOSING
5665 ESTABLISHED
145 FIN_WAIT1
141 FIN_WAIT2
1068 LAST_ACK
2 LISTEN
287 SYN_RECV
68 TIME_WAIT
相比FIN_WAIT,SYN_RECV的值偏高。加大发送synack的质量
[root@ccsafe ~]# sysctl -a|grep synack
net.ipv4.tcp_synack_retries = 1
[root@ccsafe ~]# echo “2” >/proc/sys/net/ipv4/tcp_synack_retries
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
3 CLOSE_WAIT
16 CLOSING
5317 ESTABLISHED
200 FIN_WAIT1
158 FIN_WAIT2
1001 LAST_ACK
2 LISTEN
303 SYN_RECV
78 TIME_WAIT
[root@ccsafe ~]# sysctl -a|grep keepalive
net.ipv4.tcp_keepalive_intvl = 15
net.ipv4.tcp_keepalive_probes = 5
net.ipv4.tcp_keepalive_time = 1800
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
1 CLOSE_WAIT
7 CLOSING
5356 ESTABLISHED
175 FIN_WAIT1
136 FIN_WAIT2
1045 LAST_ACK
2 LISTEN
345 SYN_RECV
64 TIME_WAIT
减少keepalive的检测周期,LAST_ACK上升
[root@ccsafe ~]# echo “10” >/proc/sys/net/ipv4/tcp_keepalive_intvl
[root@ccsafe ~]# echo “1” >/proc/sys/net/ipv4/tcp_synack_retries
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
1 CLOSE_WAIT
13 CLOSING
5605 ESTABLISHED
212 FIN_WAIT1
131 FIN_WAIT2
1143 LAST_ACK
2 LISTEN
252 SYN_RECV
79 TIME_WAIT
恢复
[root@ccsafe ~]# echo “15” >/proc/sys/net/ipv4/tcp_keepalive_intvl
[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
3 CLOSE_WAIT
14 CLOSING
5862 ESTABLISHED
230 FIN_WAIT1
205 FIN_WAIT2
1064 LAST_ACK
2 LISTEN
244 SYN_RECV
59 TIME_WAIT

[root@ccsafe ~]# watch -n 10 “netstat -ant|fgrep “:”|cut -b 77-90|sort |uniq -c”
Every 10.0s: netstat -ant|fgrep :|cut -b 77-90|sort |uniq -c
3 CLOSE_WAIT
26 CLOSING
6712 ESTABLISHED
270 FIN_WAIT1
230 FIN_WAIT2
994 LAST_ACK
2 LISTEN
254 SYN_RECV
73 TIME_WAIT

[root@ccsafe ~]#
目前LAST_ACK占ESTABLISHED的量在15%左右

发表评论

电子邮件地址不会被公开。 必填项已用*标注