#!/bin/bash

: << =cut

=head1 NAME

  zfs_arcstats - Munin multi-graph plugin to monitor Solaris ZFS ARC statistics

  These functions are implemented:
    size     : to monitor ARC size
    activity : to monitor ARC activities
    actlist  : to monitor ARC activities by cache list (MFU/MRU)
    actdata  : to monitor ARC activities by data type (Demand/Prefetch)
    hitratio : to monitor ARC hit ratio

  Tested with Solaris 10 and 11.
  This plugin is inspired by arcstat.pl [https://github.com/mharsch/arcstat]

=head1 CONFIGURATION

  Make symlink:
    cd /path/to/munin/etc/plugins
    ln -s /path/to/munin/lib/plugins/zfs_arcstats .

=head1 ENVIRONMENT VARIABLES
 
  None

=head1 AUTHOR

  K.Cima https://github.com/shakemid

=head1 LICENSE

  GPLv2

=head1 Magic markers

  #%# family=contrib
  #%# capabilities=autoconf

=cut

# Include plugin.sh
. "${MUNIN_LIBDIR:-}/plugins/plugin.sh"
is_multigraph "$@"

# Shell options
set -o nounset

# Set global variables
plugin_name=zfs_arcstats
functions='size activity actlist actdata hitratio'

# Functions

preconfig() {
    local func=$1

    # data_attr format: field type draw label
    #   label can contain white-spaces.

    case $func in
    size)
        global_attr="
            graph_title ZFS ARC - Size 
            graph_category fs
            graph_args --base 1024 --lower-limit 0
            graph_vlabel Bytes
            graph_info ZFS ARC - Size
        "
        data_attr="
            data_size          GAUGE AREASTACK Data size
            prefetch_meta_size GAUGE AREASTACK Prefetch meta size
            buf_size           GAUGE AREASTACK Buf size
            other_size         GAUGE AREASTACK Other size
            c                  GAUGE LINE      Target size
            p                  GAUGE LINE      Target MRU size
        "
        ;;
    activity)
        global_attr="
            graph_title ZFS ARC - Activities
            graph_category fs
            graph_args --base 1000 --lower-limit 0
            graph_vlabel misses (-) / hits (+) per second
            graph_info ZFS ARC - Activities

            hits.negative misses
            l2_hits.negative l2_misses
        "
        data_attr="
            misses    DERIVE LINE dummy
            hits      DERIVE LINE ARC
            l2_misses DERIVE LINE dummy
            l2_hits   DERIVE LINE L2ARC
        "
        ;;
    actlist)
        global_attr="
            graph_title ZFS ARC - Activities by cache list
            graph_category fs
            graph_args --base 1000 --lower-limit 0
            graph_vlabel ghost hits (-) / hits (+) per second
            graph_info ZFS ARC - Activities by cache list

            mfu_hits.negative mfu_ghost_hits
            mru_hits.negative mru_ghost_hits
        "
        data_attr="
            mfu_ghost_hits DERIVE LINE dummy
            mfu_hits       DERIVE LINE MFU 
            mru_ghost_hits DERIVE LINE dummy
            mru_hits       DERIVE LINE MRU
        "
        ;;
    actdata)
        global_attr="
            graph_title ZFS ARC - Activities by data type
            graph_category fs
            graph_args --base 1000 --lower-limit 0
            graph_vlabel misses (-) / hits (+) per second
            graph_info ZFS ARC - Activities by data type

            demand_data_hits.negative demand_data_misses
            demand_metadata_hits.negative demand_metadata_misses
            prefetch_data_hits.negative prefetch_data_misses
            prefetch_metadata_hits.negative prefetch_metadata_misses
        "
        data_attr="
            demand_data_misses       DERIVE LINE dummy
            demand_data_hits         DERIVE LINE D data
            demand_metadata_misses   DERIVE LINE dummy
            demand_metadata_hits     DERIVE LINE D meta
            prefetch_data_misses     DERIVE LINE dummy
            prefetch_data_hits       DERIVE LINE P data
            prefetch_metadata_misses DERIVE LINE dummy
            prefetch_metadata_hits   DERIVE LINE P meta
        "
        ;;
    hitratio)
        global_attr="
            graph_title ZFS ARC - Hit ratio
            graph_category fs
            graph_args --base 1000 --lower-limit 0 --upper-limit 100 --rigid
            graph_vlabel % hits
            graph_info ZFS ARC - Hit ratio - The graph shows cache hit ratio between munin-update intervals (usually 5 minutes).

            hitratio.cdef                   hits,DUP,DUP,misses,+,/,100,*,hitratio,IF
            l2_hitratio.cdef                l2_hits,DUP,DUP,l2_misses,+,/,100,*,l2_hitratio,IF
            demand_data_hitratio.cdef       demand_data_hits,DUP,DUP,demand_data_misses,+,/,100,*,demand_data_hitratio,IF
            demand_metadata_hitratio.cdef   demand_metadata_hits,DUP,DUP,demand_metadata_misses,+,/,100,*,demand_metadata_hitratio,IF
            prefetch_data_hitratio.cdef     prefetch_data_hits,DUP,DUP,prefetch_data_misses,+,/,100,*,prefetch_data_hitratio,IF
            prefetch_metadata_hitratio.cdef prefetch_metadata_hits,DUP,DUP,prefetch_metadata_misses,+,/,100,*,prefetch_metadata_hitratio,IF
        "
            # Note: hitratio = hits > 0 ? hits / ( hits + misses ) * 100 : 0

            # Bug?: If '0' is set at IF ELSE of CDEF, RRD cur value always become 0. 
            #   NG: hits,DUP,DUP,misses,+,/,100,*,0,IF
            #   OK: hits,DUP,DUP,misses,+,/,100,*,hitratio,IF

        data_attr="
            hits                       DERIVE LINE  dummy
            misses                     DERIVE LINE  dummy
            l2_hits                    DERIVE LINE  dummy
            l2_misses                  DERIVE LINE  dummy
            demand_data_hits           DERIVE LINE  dummy
            demand_data_misses         DERIVE LINE  dummy
            demand_metadata_hits       DERIVE LINE  dummy
            demand_metadata_misses     DERIVE LINE  dummy
            prefetch_data_hits         DERIVE LINE  dummy
            prefetch_data_misses       DERIVE LINE  dummy
            prefetch_metadata_hits     DERIVE LINE  dummy
            prefetch_metadata_misses   DERIVE LINE  dummy
            hitratio                   GAUGE  LINE2 ARC hits
            l2_hitratio                GAUGE  LINE  L2ARC hits
            demand_data_hitratio       GAUGE  LINE  Demand data hits
            demand_metadata_hitratio   GAUGE  LINE  Demand metadata hits
            prefetch_data_hitratio     GAUGE  LINE  Prefetch data hits
            prefetch_metadata_hitratio GAUGE  LINE  Prefetch metadata hits
        "
        ;;
    *)
        echo "Unknown function: $func"
        exit 1
        ;;
    esac 
}

do_config() {
    local func=$1
    local label_max_length=45 
    local field type draw label

    preconfig "$func"
    echo "multigraph ${plugin_name}_${func}"

    # print global attributes
    echo "$global_attr" | sed -e 's/^  *//' -e '/^$/d'

    # print data source attributes
    echo "$data_attr" | while read -r field type draw label
    do
        [ -z "$field" ] && continue

        echo "${field}.type ${type}"
        echo "${field}.draw ${draw}"
        echo "${field}.label ${label:0:${label_max_length}}"
        if [ "$type" = 'DERIVE' ]; then
            echo "${field}.min 0"
        fi
        if [ "$label" = 'dummy' ]; then
            echo "${field}.graph no"
        fi
    done

    echo
}

get_stats() {
    local stat value
    
    while read -r stat value
    do 
        printf -v "arcstats_${stat}" "%s" "$value" 
        # printf -v means indirect variable assignment (similar to eval)
    done < <( kstat -p 'zfs:0:arcstats' | sed -e 's/:/ /g' | awk '{ print $4,$5 }' )

    # kstat output example:
    #   $ kstat -p zfs:0:arcstats:c
    #   zfs:0:arcstats:c        4135233544
    #   ...

    # Note: It should be possible to monitor ZFS on FreeBSD/Linux by 
    #   imprementing get_stats function as well.

    #   FreeBSD: sysctl kstat.zfs.misc.arcstats ...
    #   Linux: cat /proc/spl/kstat/zfs/arcstats ...
}

do_fetch() {
    local func=$1
    local field type draw label value ref

    preconfig "$func"
    echo "multigraph ${plugin_name}_${func}"

    echo "$data_attr" | while read -r field type draw label
    do
        [ -z "$field" ] && continue

        ref="arcstats_${field}" 
        value=${!ref:-0} 
        # ${!varname} means indirect evaluation (similar to eval)

        echo "${field}.value ${value}"
    done

    echo
}

autoconf() {
    if which kstat >/dev/null ; then
        echo yes
    else
        echo "no (failed to find executable 'kstat')"
    fi
}

config() {
    local func

    for func in $functions
    do
        do_config "$func"
    done
}

fetch() {
    local func

    get_stats

    for func in $functions
    do
        do_fetch "$func"
    done
}

# Main
case ${1:-} in
autoconf)
    autoconf
    ;;
config)
    config
    [ "${MUNIN_CAP_DIRTYCONFIG:-}" = "1" ] && fetch
    ;;
*)
    fetch
    ;;
esac

exit 0