性能测试--检查服务器配置脚本

LANG="en_US.UTF-8"
# accelerated_cpus are the cpus that app running on
accelerated_cpus="1 2 3"

# 获取当前时间
ttime=`date +"%Y-%m-%d_%H-%M"`

echo "Testing_time:$ttime"
# 执行性能测试的设备
declare -i test_device_name
test_device_name=0
device_name="swiftn$test_device_name"
device_interface=0
echo "Default Test device_name=$device_name"
echo "Default Test device_interface=$device_interface"

# 判断是否安装驱动 为获得版本号
if ! command -v yusur_ctl >/dev/null 2>&1; then
    echo "yusur_ctl not found or cannot be started"
else
    first_row=$((`yusur_ctl -v| grep -n "Device swiftn[0-9]*" | awk 'NR==1' | awk -F: '{print  $1}'`))
    lat_row=$((`yusur_ctl -v| grep -n "Device swiftn[0-9]*" | awk 'NR==2' | awk -F: '{print  $1}'`))
    max_row=$((`yusur_ctl -v| wc -l`))

    # 判断一张板卡设备的详情一共有多少行
    if [ $lat_row == 0 ];then
        diff_row=$(($max_row - $first_row))
        # echo "m      $diff_row"
    else
        diff_row=$(($lat_row - $first_row - 1))
        # echo "l      $diff_row"
    fi
    # 获得版本号
    net_card_msg=`yusur_ctl -v| grep -A$diff_row "$device_name"`
    net_card_firmware_version=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep "Firmware version" | awk '{print $3}'`
    net_card_swiftn_version=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep "swiftn.ko" | awk '{print $3}'`
    net_card_instanta_version=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep "instanta.ko" | awk '{print $3}'`
    net_card_libinstanta_version=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep "libinstanta.so" | awk '{print $3}'`
    # 版本号输出排版
    net_card_version="Firmware=$net_card_firmware_version"
    swiftn_driver="swiftn.ko=$net_card_swiftn_version"
    kernel_module="instanta.ko=$net_card_instanta_version"
    lib="libinstanta.so=$net_card_libinstanta_version"

    # 输出 操作系统 检查结果记录
    echo ""
    echo "---OS version check------------------"
    sa_version="`uname -m`"
    os_version="`cat /etc/os-release | grep "PRETTY_NAME=" | awk -F'"' '{print  $2}'`"
    os_version=(${os_version// /_})
    cpu_model="`cat /proc/cpuinfo | grep "model name" | awk 'NR==1' | cut -d" " -f3-`"
    cpu_model=(${cpu_model// /_})
    printf "SA_version=\e[1;31m%s\e[0m\n" $sa_version
    echo "Cpu_model_name=$cpu_model"
    printf "OS_version=\e[1;31m%s\e[0m\n" $os_version

    # 输出 lspci 检查结果记录
    echo ""
    echo "---Yusur Network Card paramet check--"
    yusur_nic_pci_bdf=`lspci -D | grep -E "Yusur|Xilinx|1f47" | awk '{print $1}'`
    declare -i i_nr
    declare -i card_num
    i_nr=1
    card_num=0
    # 遍历含有yusur信息的设备
    for nic_bdf in $yusur_nic_pci_bdf
    do
        echo "Yusur NIC pci number:$nic_bdf"
        if [ $card_num == $test_device_name ]
        then
            test_card_bdf=$nic_bdf
        fi
        echo "    Numa: `cat /sys/bus/pci/devices/$nic_bdf/numa_node`"
        nic_bdf_int=$((16#${nic_bdf:5:2}+0))
        device_name_pci=` yusur_ctl | grep -E "swiftn[0-9]*:" | awk "NR==$i_nr" `
        device_name_pci=${device_name_pci:0:-1}
        echo "    Device_name:$device_name_pci"
        echo "    Base_addr[size=16M]:`lspci -v -s $nic_bdf | grep Memory | awk '{print $3}' | awk 'NR==1'`"
        echo "    LnkSta: `lspci -vv -s $nic_bdf | grep LnkSta: | awk '{print $2, $3, $5, $6}'`"
        i_nr+=1
        card_num+=1
    done
    declare -i cpu_num
    # 遍历中断状态
    echo "Interruption Status:"
    nic_irq_num=`cat /proc/interrupts | grep swiftn | wc -l`
    for ((i=0;i<nic_irq_num;i++))
    do
        nic_irq=`cat /proc/interrupts | grep swiftn| awk "NR==$((i+1))" |awk '{print $1}'| awk '{t=length($0)}END{print substr($0, 0, t-1)}' `
        nic_irq_name=`cat /proc/interrupts | grep swiftn| awk "NR==$((i+1))" |awk '{print $NF}'`
        if [ ! -n "$nic_irq" ] ; then
            echo "    Interrupt is Null"
        else
            echo "    $nic_irq_name Interrupt: $nic_irq"
            echo "    Irq affinity: `cat /proc/irq/$nic_irq/smp_affinity_list`"
        fi
    done
    cpu_num=`cat /proc/cpuinfo | grep processor | wc -l`
    interrupts_leng=` cat /proc/interrupts | grep swiftn | awk "NR==1" | awk '{print NF}' `
    other_info_leng=$(($interrupts_leng-$cpu_num-1))
    info_leng=$(($other_info_leng+$cpu_num))

    i=0
    printf "IRQ Details:\n"
    for cpu_i in `cat /proc/interrupts | grep swiftn`
    do
        if (( $i < $info_leng))
        then
            ((i++))
            if (( i == 1 ))
            then
                echo "    $cpu_i"
            elif (( i > 1 )) &&  ((i < $((cpu_num+2)) ))
            then
                k=$(($cpu_i))
                # k=$cpu_i
                interrupt=$(($i-2))
                # printf "%s:%s " $interrupt $k
                if test $k -ne 0;then
                    printf "        code:%s-IRQ_num:%s \n" $interrupt $k
                fi
            else
                sleep 0.01
            fi
        else
            echo "    *** BOTTOM ****************"
            i=0
        fi
    done

    # 输出 yusur_ctl 检查结果记录
    echo ""
    echo "---Yusur Network Card config check---"

    net_card_num=$((`yusur_ctl -v| grep -n "Device swiftn.*" | wc -l`))
    echo "Number of network cards:$net_card_num"
    # 输出执行性能测试设备的版本信息
    declare -i i
    i=0
    ls /dev/swiftn*| awk -F'/' '{print $3}'| while read line
    do
        echo "    dev_name[$i]:$line"
        i=i+1
    done

    echo "`yusur_ctl -v| grep "$device_name"` "
    echo "    $net_card_version"
    echo "    $swiftn_driver"
    echo "    $kernel_module"
    echo "    $lib"

    # test_card_bdf 文件头设置的测试网卡bdf
    interface_0=` ls /sys/bus/pci/devices/$test_card_bdf/net | awk "NR==1" `
    interface_1=` ls /sys/bus/pci/devices/$test_card_bdf/net | awk "NR==2" `
    # 收集测试网卡的接口信息
    echo " `yusur_ctl -v| grep -A$diff_row "$device_name" | grep "$interface_0" ` "
    mtu_0=`ifconfig $interface_0 |grep mtu |awk '{print $NF}'`
    echo "        MTU:$mtu_0"
    if [ ` ethtool -c $interface_0 >/dev/null 2>&1 || echo $? ` ];then
        echo "        rx-usecs:N/A"
    else
        rx_usecs=` ethtool -c $interface_0 | grep "rx-usecs:" | cut -d" " -f2- `
        echo "        rx-usecs:$rx_usecs"
    fi
    loopback_mode=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_0" | grep "Loopback mode:" | awk '{print $3}'`
    echo "        Loopback_mode:$loopback_mode"
    # promiscuous_mode=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_0" | grep "Promiscuous mode:" | awk '{print $3}'`
    # echo "        Promiscuous_mode:$promiscuous_mode"
    Kernel_bypass_mode=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_0" | grep "Kernel-bypass mode:" | awk '{print $3}'`
    echo "        Kernel-bypass_mode:$Kernel_bypass_mode"
    ip=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_0" | grep "inet:" | awk '{print $2}'`
    echo "        IP address:$ip"

    echo "  `yusur_ctl -v| grep -A$diff_row "$device_name" | grep "$interface_1" ` "
    mtu_1=`ifconfig $interface_1 |grep mtu |awk '{print $NF}'`
    echo "        MTU:$mtu_1"
    if [ ` ethtool -c $interface_1 >/dev/null 2>&1 || echo $? ` ];then
        echo "        rx-usecs:N/A"
    else
        rx_usecs=` ethtool -c $interface_1 | grep "rx-usecs:" | cut -d" " -f2- `
        echo "        rx-usecs:$rx_usecs"
    fi
    loopback_mode=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_1" | grep "Loopback mode:" | awk '{print $3}'`
    echo "        Loopback_mode:$loopback_mode"
    # promiscuous_mode=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_1" | grep "Promiscuous mode:" | awk '{print $3}'`
    # echo "        Promiscuous_mode:$promiscuous_mode"
    Kernel_bypass_mode=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_1" | grep "Kernel-bypass mode:" | awk '{print $3}'`
    echo "        Kernel-bypass_mode:$Kernel_bypass_mode"
    ip=`yusur_ctl -v| grep -A$diff_row "$device_name" | grep -A 12 "$interface_1" | grep "inet:" | awk '{print $2}'`
    echo "        IP address:$ip"
fi

# 输出 dmidecode 检查结果记录
echo ""
echo "---DIMM check------------------------"

mem=`dmidecode -t memory | grep -A5 "Memory Device" | grep "Size"`
mem_total_slot=`dmidecode -t memory | grep -A5 "Memory Device" | grep "Size" | wc -l`
mem_empty_slot=`dmidecode -t memory | grep -A5 "Memory Device" | grep "Size" | grep "No Module Installed" | wc -l`

if [ $mem_total_slot -ne $mem_empty_slot ]; then
    echo "    DIMM slots (empty/total): $mem_empty_slot/$mem_total_slot"
    echo "    DIMM slots : "
    dmidecode -t memory | grep -A5 "Memory Device" | grep "Size" | while read line
    do
        echo "        $line"
    done
fi

# 输出 lscpu-numa 检查结果记录
echo ""
echo "---Cpu numa check--------------------"
lscpu | grep -E "NUMA node|NUMA 节点" | while read line_cpu
do
    echo "    $line_cpu"
done

# 输出 cmdline 检查结果记录
echo ""
echo "---Cmdline check---------------------"
isolcpus="`awk '{for(i=1;i<=NF;i++) if($i ~ /isolcpus/)   print $i}' /proc/cmdline`"
nohz_full="`awk '{for(i=1;i<=NF;i++) if($i ~ /nohz_full/)  print $i}' /proc/cmdline`"
rcu_nocbs="`awk '{for(i=1;i<=NF;i++) if($i ~ /rcu_nocbs/)  print $i}' /proc/cmdline`"
iommu="`awk '{for(i=1;i<=NF;i++) if($i ~ /iommu/)       print $i}' /proc/cmdline`"
intel_max_cstate="`awk '{for(i=1;i<=NF;i++) if($i ~ /intel_idle.max_cstate/)       print $i}' /proc/cmdline`"
amd_max_cstate="`awk '{for(i=1;i<=NF;i++) if($i ~ /processor.max_cstate/)       print $i}' /proc/cmdline`"
idle="`awk '{for(i=1;i<=NF;i++) if($i ~ /idle=/)       print $i}' /proc/cmdline`"
kthread_cpus="`awk '{for(i=1;i<=NF;i++) if($i ~ /kthread_cpus/)       print $i}' /proc/cmdline`"
irqaffinity="`awk '{for(i=1;i<=NF;i++) if($i ~ /irqaffinity/)       print $i}' /proc/cmdline`"
intel_pstate="`awk '{for(i=1;i<=NF;i++) if($i ~ /intel_pstate/)       print $i}' /proc/cmdline`"
echo "    isolcpus           : $isolcpus"
echo "    nohz_full          : $nohz_full"
echo "    rcu_nocbs          : $rcu_nocbs"
echo "    iommu              : $iommu"
echo "    idle               : $idle"
echo "    kthread_cpus       : $kthread_cpus"
echo "    irqaffinity        : $irqaffinity"
echo "    (Intel) max_cstate : $intel_max_cstate"
echo "    (Amd) max_cstate   : $amd_max_cstate"
echo "    (Intel) pstate     : $intel_pstate"


echo ""
echo "---Cpu freq scaling gonernor check---"
for cpu in $accelerated_cpus
do
    if [ ! -d "/sys/devices/system/cpu/cpu$cpu/cpufreq/" ] ; then
        echo "    Cannot find cpu$cpu"
    else
        echo "    cpu$cpu: `cat "/sys/devices/system/cpu/cpu$cpu/cpufreq/scaling_governor"`"
    fi
done

# 输出 cpu模式 检查结果记录
echo ""
echo "---Cpu freq check--------------------"
for cpu in $accelerated_cpus
do
    if [ ! -d "/sys/devices/system/cpu/cpu$cpu/cpufreq/" ] ; then
        echo "    Cannot find cpu$cpu"
    else
        echo "    cpu$cpu: `cat "/sys/devices/system/cpu/cpu$cpu/cpufreq/scaling_cur_freq"`"
    fi
done

# 输出 cpu频率 检查结果记录
echo ""
echo "---Watchdog check--------------------"
watchdog=`cat /proc/sys/kernel/watchdog`
if [ $watchdog == 1 ];then
    echo "    Watchdog is open"
else
    echo "    Watchdog is closed"
fi

echo ""
echo "Test End"