omd tuning script

omd-tuning.sh

#!/usr/bin/env bash
 
# License: BSD
# Author: Florian Heigl (@FlorianHeigl)
# Contributors: Davide Gibilisco (@m3rlinux)
 
set -eu

# Function to convert bytes to human-readable format
human_readable() {
    awk 'function human_readable(bytes) {
        if (bytes >= 1099511627776) {
            suffix = "TB"
            divisor = 1099511627776
        } else if (bytes >= 1073741824) {
            suffix = "GB"
            divisor = 1073741824
        } else if (bytes >= 1048576) {
            suffix = "MB"
            divisor = 1048576
        } else if (bytes >= 1024) {
            suffix = "KB"
            divisor = 1024
        } else {
            suffix = "B"
            divisor = 1
        }
        return sprintf("%.2f %s", bytes / divisor, suffix)
    }
    { print human_readable($1) }'
}

# pull variables if not present
SITECFG=~/etc/omd/site.conf
if [ -r "$SITECFG" ]; then
    source "$SITECFG" || true
fi

# determine total memory (in kilobyte like RSS in ps)
MEM_TOTAL=$(free --kilo | awk '/Mem:/ {print $2}')

# NET Interfaces
ETH_NAMES=($(ip -o link show up | awk '!/lo/ {print $2}' | cut -d ':' -f 1))

# list of test commands
tests=(
    "grep -c ^processor /proc/cpuinfo" # 1
    "echo $((MEM_TOTAL * 1024)) | human_readable" # 2
    "cat /proc/sys/kernel/pid_max" # 3
    "cat /proc/sys/fs/file-max" # 4
    "awk '{print \$1}' /proc/sys/fs/file-nr | human_readable" # 5
    "df --output=source $OMD_ROOT | grep ^/dev | awk '{print \$4}' | human_readable" # 6
    "ps hux -U $OMD_SITE | awk -v total=$MEM_TOTAL '{sum +=\$6} END {printf \"%.2f%%\\n\", sum / total * 100}'" # 7
    "ulimit -n" # 8
    "ulimit -u" # 9
    "lsof -u $OMD_SITE -e /sys/kernel/debug/tracing | wc -l" # 10
    "ps -u $OMD_SITE | grep -v 'PID' | wc -l" # 11
    "grep ^write_bytes /proc/\$(pgrep -u $OMD_SITE rrdcached)/io | awk '{print \$2}' | human_readable" # 12
    "grep ^read_bytes /proc/\$(pgrep -u $OMD_SITE rrdcached)/io | awk '{print \$2}' | human_readable" # 13
    "pgrep -u $OMD_SITE checkhelper | wc -l" # 14
    "pgrep -u $OMD_SITE -f 'python /omd/sites/$OMD_SITE/bin/cmk --keepalive' | wc -l" # 15
    "grep -c 'Resource temp' ~/var/log/cmc.log" # 16
    "pgrep -u $OMD_SITE -f 'python /omd/sites/$OMD_SITE/bin/liveproxyd' | wc -l" # 17
    "grep -c 'Site is considered dead. Closing all connections.' ~/var/log/liveproxyd.log" # 18
    "grep -c -E 'Cannot forward next' ~/var/log/liveproxyd.log" # 19
    "curl -s localhost:${CONFIG_APACHE_TCP_PORT}/server-status | grep -E 'requests currently being processed' | sed 's/<dt>//; s/<\/dt>//'" # 20
    "grep -c 'WARNING: ping-queueing has lasted' ~/var/log/cmc.log" # 21
)

# list of test descriptions
messages=(
    "System: CPU Cores" # 1
    "System: Total Memory" # 2
    "System: process limit" # 3
    "System: file descriptors limit" # 4
    "System: opened file descriptors" # 5
    "System: OMD_ROOT mount info" # 6
    "OMD Site: Total memory used" # 7
    "OMD Site: file limit" # 8
    "OMD Site: processes limit" # 9
    "OMD Site: open files" # 10
    "OMD Site: running processes" # 11
    "rrdcached: written" # 12
    "rrdcached: read" # 13
    "CMC: active check workers" # 14
    "CMC: check_mk workers" # 15
    "CMC: resources exhaustion errors" # 16
    "Liveproxyd: processes" # 17
    "Liveproxyd: remote site conn dead errors" # 17
    "Liveproxyd: remote site query aborted errors" # 18
    "Apache: worker usage" # 20
    "icmphelper: ping-queue over 100ms errors" # 21
)

# Run tests and output values
for index in "${!messages[@]}"; do
    echo "${messages[index]} : $(eval "${tests[index]}")"
done

# Execute ethtool commands for each network interface
for eth in "${ETH_NAMES[@]}"; do
    echo "$eth - RX buffer size : $(ethtool -g $eth | awk '/Current hardware/{found=1} found && /^RX/{print $2; found=0}')"
    echo "$eth - NIC out of buffer errors : $(ethtool -S $eth | awk -F":" '/OOB/ {sum += $2} END {print sum}')"
    echo "$eth - Driver rx/tx drop errors : $(ethtool -S $eth | awk '/drv dropped tx total/ {tx_sum += $NF} /drv dropped rx total/ {rx_sum += $NF} END {print rx_sum"/"tx_sum}')"
done

@m3rlinux hey thank you, that was a lot of cleanups & improvements!