Telegraf plugin to get summary CPU/mem usage for a tree of processes

sumproc.sh

#!/bin/bash
#
# Telegraf input plugin to get summary cpu/mem/io usage for
# a tree of processes. Identify top process(es) by
# pid file, pgrep pattern or owner.
#
# Note: it requires root to be able to read 'smaps'.
# Note: pattern match (-p) can catch the plugins itself
#
# Output fields:
# cpu_user [ticks] - sum(utime)/100 (see 'man 5 proc' for /proc/<pid>/stat)
# cpu_system [ticks] - sum(stime)/100 (see 'man 5 proc' for /proc/<pid>/stat)
# nprocess [count] - number of processes (including all children)
# mem_pss/rss [B] - sum(pss)*1024 (see 'man 5 proc' for /proc/<pid>/smaps)
# io_read/write [B] - sum(read/write_bytes) (see 'man 5 proc' for /proc/<pid>/io)
# io_syscr/syscw [count] - sum(syscr/syscw) (see 'man 5 proc' for /proc/<pid>/io)
# fd_count [count] - file-descriptor count
#

function usage() {
    echo "usage: `basename $0` <options>"
    echo "  -f PIDFILE  get pids from PIDFILE"
    echo "  -p PATTERN  get pids from pgrep PATTERN"
    echo "  -u USER     get pids owned by USER"
    echo "  -h          this help screen"
}

function descendants() {
    # return all process descendants from a given list of pids
    # (including pids from input)

    declare -a pidlist=($*)
    for ((i=0; i<${#pidlist[@]}; i++)) {
        p=${pidlist[$i]}
        [ -z "$p" -o "$p" -eq "$$" ] && continue
        # append all direct children of $p
        pidlist+=(`pgrep -P "$p" 2>/dev/null`)
    }
    echo ${pidlist[@]} | tr ' ' '\n' | sort -u
}

function cpustat() {
    for p in $*; do
        cat "/proc/$p/stat" 2>/dev/null
    done | awk '
BEGIN { utime = stime = count = 0 }
{ utime += $14; stime += $15; count++ }
END { printf "%.2f %.2f %d\n", utime/100, stime/100, count }'
}

function memstat() {
    for p in $*; do
        cat "/proc/$p/smaps" 2>/dev/null
    done | awk '
BEGIN { pss = rss = cache = 0 }
$1 ~ /^[0-9a-f]/ {
    if ($4 ~ /ca:/ ) {
        cache = 1
    } else {
        cache = 0
    }
}
$1 == "Pss:" && !cache { pss += $2 }
$1 == "Rss:" && !cache { rss += $2 }
END { printf "%.0f %.0f\n", pss * 1024, rss * 1024 }'
}

function iostat() {
    for p in $*; do
        cat "/proc/$p/io" 2>/dev/null
    done | awk '
BEGIN { read_bytes = write_bytes = syscr = syscw = 0 }
$1 == "read_bytes:"  { read_bytes  += $2 }
$1 == "write_bytes:" { write_bytes += $2 }
$1 == "syscr:"       { syscr       += $2 }
$1 == "syscw:"       { syscw       += $2 }
END { printf "%d %d %d %d\n", read_bytes, write_bytes, syscr, syscw }'
}

function fdstat() {
    for p in $*; do
        ls -1 "/proc/$p/fd" 2>/dev/null
    done | wc -l
}

pidfile=
pattern=
user=
while getopts "f:p:u:h" opt; do
    case $opt in
        f) pidfile=$OPTARG ;;
        p) pattern=$OPTARG ;;
        u) user=$OPTARG ;;
        h) usage; exit 0 ;;
        *) exit 2 ;;
    esac
done
shift $((OPTIND - 1))

pidlist=
if [ -n "$pidfile" ]; then
    pidlist=$(cat "$pidfile" 2>/dev/null)
elif [ -n "$pattern" ]; then
    pidlist=$(pgrep -f "$pattern")
elif [ -n "$user" ]; then
    pidlist=$(pgrep -U "$user")
else
    echo "Specify either <pidfile>, <pattern> or <user>" 1>&2
    usage
    exit -1
fi

if [ -z "$pidlist" ]; then
    # no match
    exit 0
fi

pidlist2=$(descendants "$pidlist")

declare -a cpu=(`cpustat $pidlist2`)
declare -a mem=(`memstat $pidlist2`)
declare -a io=(`iostat $pidlist2`)
fd_count=`fdstat $pidlist2`
echo -n "sumproc cpu_user=${cpu[0]},cpu_system=${cpu[1]},nprocess=${cpu[2]}i,"
echo -n "mem_pss=${mem[0]}i,mem_rss=${mem[1]}i,"
echo -n "io_read=${io[0]}i,io_write=${io[1]}i,io_syscr=${io[2]}i,io_syscw=${io[3]}i,"
echo    "fd_count=${fd_count}i"