monitor-io

#!/bin/bash

#  Copyright 2021 Red Hat, Inc.
#
#  Purpose:  Monitor block device performance
#  Author:   Vivek Goyal <vgoyal@redhat.com>

set -e

# Sampling interval in seconds
_INTERVAL="5"
# Default average wait thresholds in ms
_R_AWAIT_THRESHOLD=50
_W_AWAIT_THRESHOLD=50
_PRINT_FORMAT="%-24s%-16s%-16s\n"
usage() {
  cat <<-FOE
Usage: $1 [OPTIONS] <block-dev-to-monitor>
Options:
  -h, --help	Print help message
  -r, --rlat	Average read latency threshold in ms (default 50ms)
  -w, --wlat	Average write latency threshold in ms (default 50ms)
  -i		Sampling Interval in seconds (default 5 second)
FOE
}

print_header() {
  #echo "Read latency threshold=$_R_AWAIT_THRESHOLD Write latency threshold=$_W_AWAIT_THRESHOLD"
  #printf "\n\n"
  printf "$_PRINT_FORMAT" "Date/Time" "Read Lat(ms)" "Write Lat(ms)"
}

monitor_device() {
  local bdev_path=$1
  local bdev_base=$(basename $bdev_path)
  local field_dev f_r_await f_w_await r_await w_await
  local timestamp

  /usr/bin/iostat -d $bdev_path -x -y $_INTERVAL | while read line;
  do
    # TODO: Use awk instead
    field_dev=`echo $line | cut -d " " -f1`
    [ "$field_dev" != "$bdev_base" ] && continue

    f_r_await=`echo $line | cut -d " " -f6`
    f_w_await=`echo $line | cut -d " " -f12`
    timestamp=`date +"%x %T"`

    printf "$_PRINT_FORMAT" "$timestamp" "$f_r_await" "$f_w_await"

    r_await=${f_r_await%%.*}
    w_await=${f_w_await%%.*}

    if [ $r_await -gt $_R_AWAIT_THRESHOLD ];then
      printf "%s Read Latency %sms is higher than threshold %sms\n" "$timestamp" "$r_await" "$_R_AWAIT_THRESHOLD"
    fi

    if [ $w_await -gt $_W_AWAIT_THRESHOLD ];then
      printf "%s Write Latency %sms is higher than threshold %sms \n" "$timestamp" "$w_await" "$_W_AWAIT_THRESHOLD"
    fi
  done
}

# Main
_INPUT_STR=""$@
_OPTS=`getopt -o hr:w:i: -l help -l rlat: -l wlat: -- $_INPUT_STR`
eval set -- "$_OPTS"
while true ; do
    case "$1" in
        -h | --help)  usage $(basename $0); exit 0;;
        -r | --rlat)  _R_AWAIT_THRESHOLD=$2; shift 2;;
        -w | --wlat)  _W_AWAIT_THRESHOLD=$2; shift 2;;
        -i)  _INTERVAL=$2; shift 2;;
        --) shift; break;;
    esac
done

if [ $# -ne 1 ];then
  usage $(basename $0)
  exit 1
fi

_BLOCK_DEV=$1
if [ -z $_BLOCK_DEV ]; then
  usage $(basename $0)
  exit 1
fi

if [ ! -b "$_BLOCK_DEV" ]; then
  echo "Error: $_BLOCK_DEV is not a block device"
  exit 1
fi

if [ ! -x "/usr/bin/iostat" ]; then
  echo "Error: /usr/bin/iostat is not available/executable."
  exit 1
fi

print_header
monitor_device "$_BLOCK_DEV"