Я запускаю сервер memcached под большой нагрузкой от нескольких веб-серверов. Изучая странное поведение на уровне приложения, мы обнаружили, что netstat -s
в окне memcached было много «сбросов соединений из-за непредвиденных данных».
Под нагрузкой значение увеличивается на несколько сотен в секунду.
Что это означает и что я могу с этим поделать?
Пример вывода:
Ip:
147533037 total packets received
2 with invalid addresses
0 forwarded
0 incoming packets discarded
147533033 incoming packets delivered
154883954 requests sent out
5 outgoing packets dropped
Icmp:
27 ICMP messages received
4 input ICMP message failed.
ICMP input histogram:
destination unreachable: 27
0 ICMP messages sent
0 ICMP messages failed
ICMP output histogram:
IcmpMsg:
InType3: 27
Tcp:
8010699 active connections openings
8086268 passive connection openings
6146 failed connection attempts
5030 connection resets received
195 connections established
134422776 segments received
141824483 segments send out
32169 segments retransmited
5 bad segments received.
1711889 resets sent
Udp:
13110231 packets received
0 packets to unknown port received.
9 packet receive errors
13129049 packets sent
UdpLite:
TcpExt:
251 resets received for embryonic SYN_RECV sockets
1949831 TCP sockets finished time wait in fast timer
4587602 time wait sockets recycled by time stamp
1005360 delayed acks sent
1055 delayed acks further delayed because of locked socket
Quick ack mode was activated 259 times
650992 packets directly queued to recvmsg prequeue.
60990 bytes directly in process context from backlog
6099061 bytes directly received in process context from prequeue
47684327 packet headers predicted
550456 packets header predicted and directly queued to user
26594156 acknowledgments not containing data payload received
44593118 predicted acknowledgments
2 bad SACK blocks received
22777 congestion windows recovered without slow start after partial ack
1 timeouts after reno fast retransmit
19 timeouts after SACK recovery
1 timeouts in loss state
406 retransmits in slow start
28152 other TCP timeouts
259 DSACKs sent for old packets
4234 DSACKs received
1130771 connections reset due to unexpected data
143 connections reset due to early user close
482 connections aborted due to timeout
TCPDSACKIgnoredNoUndo: 18
TCPSackShiftFallback: 4233
TCPChallengeACK: 5
TCPSYNChallenge: 5
IpExt:
InOctets: -414538291
OutOctets: -630924797
Из https://serverfault.com/a/527670/442063:
Вы увеличиваете лимит FD? Вы можете получить некоторую информацию здесь http://www.cyberciti.biz/faq/linux-increase-the-maximum-number-of-open-files/