Linux服务器网卡丢包的解决办法

通常情况下，Linux的网卡中断是由一个CPU核心来处理的，当承担高流量的场景下，会出现一些诡异的情况（网卡尚未达到瓶颈，但是却出现丢包的情况）

这种时候，我们最好看下网卡中断是不是缺少调优。

优化3要点：网卡多队列+irq affinity亲缘性设置+关闭irqbalance

目前阿里云官方提供的centos和ubuntu镜像里面，已经自带了优化脚本，内容如下:

centos7的脚本路径在 /usr/sbin/ecs_mq_rps_rfs 具体内容如下：

#!/bin/bash# This is the default setting of networking multiqueue and irq affinity# 1. enable multiqueue if available# 2. irq affinity optimization# 3. stop irqbalance service# set and check multiqueuefunction set_check_multiqueue(){    eth=$1    log_file=$2    queue_num=$(ethtool -l $eth | grep -ia5 'pre-set' | grep -i combined | awk {'print $2'})    if [ $queue_num -gt 1 ]; then        # set multiqueue        ethtool -L $eth combined $queue_num        # check multiqueue setting        cur_q_num=$(ethtool -l $eth | grep -iA5 current | grep -i combined | awk {'print $2'})        if [ "X$queue_num" != "X$cur_q_num" ]; then            echo "Failed to set $eth queue size to $queue_num" >> $log_file            echo "after setting, pre-set queue num: $queue_num , current: $cur_q_num" >> $log_file            return 1        else            echo "OK. set $eth queue size to $queue_num" >> $log_file        fi    else        echo "only support $queue_num queue; no need to enable multiqueue on $eth" >> $log_file    fi}#set irq affinityfunction set_irq_smpaffinity(){    log_file=$1    node_dir=/sys/devices/system/node    for i in $(ls -d $node_dir/node*); do        i=${i/*node/}    done        echo "max node :$i" >> $log_file    node_cpumax=$(cat /sys/devices/system/node/node${i}/cpulist |awk -F- '{print $NF}')    irqs=($(cat /proc/interrupts |grep virtio |grep put | awk -F: '{print $1}'))    core=0    for irq in ${irqs[@]};do        VEC=$core        if [ $VEC -ge 32 ];then            let "IDX = $VEC / 32"            MASK_FILL=""            MASK_ZERO="00000000"            for ((i=1; i<=$IDX;i++))                do                    MASK_FILL="${MASK_FILL},${MASK_ZERO}"                done            let "VEC -= 32 * $IDX"            MASK_TMP=$((1<<$VEC))            MASK=$(printf "%X%s" $MASK_TMP $MASK_FILL)        else            MASK_TMP=$((1<<$VEC))            MASK=$(printf "%X" $MASK_TMP)        fi        echo $MASK > /proc/irq/$irq/smp_affinity        echo "mask:$MASK, irq:$irq" >> $log_file        core=$(((core+1)%(node_cpumax+1)))    done}# stop irqbalance servicefunction stop_irqblance(){    log_file=$1    ret=0    if [ "X" != "X$(ps -ef | grep irqbalance | grep -v grep)" ]; then        if which systemctl;then            systemctl stop irqbalance        else            service irqbalance stop        fi        if [ $? -ne 0 ]; then            echo "Failed to stop irqbalance" >> $log_file            ret=1        fi    else       echo "OK. irqbalance stoped." >> $log_file    fi    return $ret}# main logicfunction main(){    ecs_network_log=/var/log/ecs_network_optimization.log    ret_value=0    echo "running $0" > $ecs_network_log    echo "========  ECS network setting starts $(date +'%Y-%m-%d %H:%M:%S') ========" >> $ecs_network_log    # we assume your NIC interface(s) is/are like eth*    eth_dirs=$(ls -d /sys/class/net/eth*)    if [ "X$eth_dirs" = "X" ]; then        echo "ERROR! can not find any ethX in /sys/class/net/ dir." >> $ecs_network_log        ret_value=1    fi    for i in $eth_dirs    do        cur_eth=$(basename $i)        echo "optimize network performance: current device $cur_eth" >> $ecs_network_log        # only optimize virtio_net device        driver=$(basename $(readlink $i/device/driver))        if ! echo $driver | grep -q virtio; then            echo "ignore device $cur_eth with driver $driver" >> $ecs_network_log            continue        fi        echo "set and check multiqueue on $cur_eth" >> $ecs_network_log        set_check_multiqueue $cur_eth $ecs_network_log        if [ $? -ne 0 ]; then            echo "Failed to set multiqueue on $cur_eth" >> $ecs_network_log            ret_value=1        fi    done    stop_irqblance  $ecs_network_log    set_irq_smpaffinity $ecs_network_log    echo "========  ECS network setting END $(date +'%Y-%m-%d %H:%M:%S')  ========" >> $ecs_network_log    return $ret_value}# program starts heremainexit $?

查询的rps绑定情况的脚本 get_rps.sh

#!/bin/bash# 获取当前rps情况for i in $(ls /sys/class/net/eth0/queues/rx-*/rps_cpus); do   echo $i  cat $idone