munin-contrib/plugins/rsnapshot/rsnapshot_duration

#!/bin/sh

: <<=cut

=head1 NAME

rsnapshot_duration - monitor duration of rsnapshot backups in detail


=head1 APPLICABLE SYSTEMS

Any system with local access to an rsnapshot log file.


=head1 CONFIGURATION

rsnapshot's configuration file (e.g. /etc/rsnapshot.conf) needs to
contain the directive "loglevel 3" in order to write the required
log messages to the log file.


The following configuration represents the plugin's defaults:

  [rsnapshot_duration]
  env.rsnapshot_log_file        /var/log/rsnapshot
  env.rsnapshot_operation_name  beta
  env.rsnapshot_description     Backup Duration

"rsnapshot_description" is used for the title of the graph.

"rsnapshot_operation_name" refers to the interval/retain/level of the
rsnapshot operation to be monitored.  This should be the lowest
executed backup interval (e.g. "daily" or "beta").  In case of the
rsnapshot setting "sync_first 1", it should be "sync".

Multiple rsnapshot log files (with different sets of backup sources)
can be monitored.  Simply create one plugin symlink for each log file
and specify suitable plugin configurations for each symlink name.


=head1 INTERPRETATION

The backup duration for each single "backup" task is calculated based
on the difference of the timestamps in the log file.


=head1 AUTHOR

Lars Kruse <devel@sumpfralle.de>


=head1 LICENSE

SPDX-License-Identifier: GPL-3.0-or-later


=head1 MAGIC MARKERS

 #%# family=auto
 #%# capabilities=autoconf


=cut


set -eu


. "$MUNIN_LIBDIR/plugins/plugin.sh"


LOG_FILE=${rsnapshot_log_file:-/var/log/rsnapshot.log}
RSNAPSHOT_OPERATION_NAME=${rsnapshot_operation_name:-beta}
RSNAPSHOT_DESCRIPTION=${rsnapshot_description:-Backup Duration}


# retrieve the latest set of log files for a complete backup operation
get_latest_process_log() {
    tac "$LOG_FILE" \
        | awk '
            BEGIN { in_process = 0; is_finished = 0; }
            / '"$RSNAPSHOT_OPERATION_NAME"': completed[, ]/ { if (is_finished == 0) in_process = 1; }
            / '"$RSNAPSHOT_OPERATION_NAME"': started$/ { if (in_process == 1) { in_process = 0; is_finished = 1; }}
            { if (in_process == 1) print($0); }' \
        | tac
}


# parse rsnapshot log lines
# output format:
#     BACKUP_NAME DURATION_SECONDS [ERROR_MESSAGES]
get_backups_with_duration() {
    get_latest_process_log | while read -r timestamp command details; do
        parsed_timestamp=$(date +%s --date "$(printf '%s' "$timestamp" | tr -d '[]')")
        if [ -n "${backup_name:-}" ] && printf '%s' "$command" | grep -q "/rsnapshot$" && printf '%s' "$details" | grep -q "ERROR:"; then
            backup_errors=$(printf '%s' "$details" | sed 's/^.*ERROR: //')
        elif printf '%s' "$command" | grep -q "/rsync$"; then
            if [ -n "${backup_name:-}" ]; then
                printf '%s\t%d\t%s\n' "$backup_name" "$((parsed_timestamp - backup_start))" "${backup_errors:-}"
            fi
            backup_name=$(printf '%s' "$details" | sed 's#/$##' | sed 's#^.*/##')
            backup_start=$parsed_timestamp
            backup_errors=
        elif printf '%s' "$command" | grep -q "/rm$"; then
            # the backup is finished
            if [ -n "${backup_name:-}" ]; then
                printf '%s\t%d\t%s\n' "$backup_name" "$((parsed_timestamp - backup_start))" "${backup_errors:-}"
            fi
            cat >/dev/null
        fi
    done | sort
}


do_autoconf() {
    if [ -e "$LOG_FILE" ]; then
        if [ -r "$LOG_FILE" ]; then
            if command -v "tac" >/dev/null; then
                echo "yes"
            else
                echo "no (executable 'tac' (coreutils) is missing)"
            fi
        else
            echo "no (rsnapshot log file is not readable: $LOG_FILE)"
        fi
    else
        echo "no (rsnapshot log file missing: $LOG_FILE)"
    fi
    exit 0
}


get_backup_fieldname() {
    local backup_name="$1"
    clean_fieldname "backup_${backup_name}"
}


print_backup_details() {
    local backup_name="$1"
    local backup_duration="$2"
    local backup_messages="$3"
    local fieldname
    fieldname=$(get_backup_fieldname "$backup_name")
    printf '%s.value %s\n' "$fieldname" "$backup_duration"
    if [ -n "$backup_messages" ]; then
        printf '%s.extinfo %s\n' "$fieldname" "$backup_messages"
    fi
}


do_config() {
    local do_emit_values="${1:-0}"
    echo "graph_title rsnapshot - $RSNAPSHOT_DESCRIPTION"
    echo 'graph_vlabel Duration of backup in minutes'
    echo 'graph_category backup'
    echo 'graph_scale no'
    get_backups_with_duration | while read -r name duration messages; do
        fieldname=$(clean_fieldname "backup_$name")
        printf '%s.label %s\n' "$fieldname" "$name"
        printf '%s.draw %s\n' "$fieldname" "AREASTACK"
        # The duration is stored as an SI unit (seconds).
        # The visualization as the number of minutes should be suitable for most backups.
        printf '%s.cdef %s,60,/\n' "$fieldname" "$fieldname"
        if [ "$do_emit_values" = "1" ]; then
            print_backup_details "$name" "$duration" "$messages"
        fi
    done
}


do_fetch() {
    get_backups_with_duration | while read -r name duration messages; do
        print_backup_details "$name" "$duration" "$messages"
    done
}


case ${1:-} in
    autoconf)
        do_autoconf
        ;;
    config)
        do_config "${MUNIN_CAP_DIRTYCONFIG:-0}"
        ;;
    "")
        do_fetch
        ;;
    *)
        echo >&2 "Unknown command: $1"
        exit 1
        ;;
esac