|文章出处：http://hi.baidu.com/_wang8
记录一下自己的思想过程
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 ~]# !ec
echo "0" >/proc/sys/net/ipv4/tcp_syncookies
[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 ~]# !wat
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 ~]# !wat
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 ~]# !wat
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 ~]# !wat
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 ~]# !wat
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 ~]# !wat
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
