#!/usr/lib64/linuxfabrik-monitoring-plugins/venv/bin/python
# -*- coding: utf-8; py-indent-offset: 4 -*-
#
# Author:  Linuxfabrik GmbH, Zurich, Switzerland
# Contact: info (at) linuxfabrik (dot) ch
#          https://www.linuxfabrik.ch/
# License: The Unlicense, see LICENSE file.

# https://github.com/Linuxfabrik/monitoring-plugins/blob/main/CONTRIBUTING.md

"""See the check's README for more details."""

import argparse
import json
import re
import sys

import lib.args
import lib.base
import lib.huawei_dorado
import lib.human
import lib.lftest
import lib.txt
from lib.globals import STATE_CRIT, STATE_OK, STATE_UNKNOWN, STATE_WARN

__author__ = 'Linuxfabrik GmbH, Zurich/Switzerland'
__version__ = '2026081102'

DESCRIPTION = """Checks the health status of all backup power modules (BBU) on a Huawei OceanStor
Dorado storage system via the REST API (/backup_power endpoint). Alerts when any
module reports a non-normal health state, when it runs out of remaining service life,
and optionally when its voltage leaves the range you expect.
Supports extended reporting via --lengthy."""

DEFAULT_CACHE_EXPIRE = 15  # minutes; default session timeout period is 20 minutes
DEFAULT_CRIT = '30:'
DEFAULT_CRIT_VOLTAGE = ''
DEFAULT_DEVICE_ID = ''  # the appliance reports its own at login
DEFAULT_INSECURE = True
DEFAULT_NO_MATCH_SEVERITY = 'ok'
DEFAULT_NO_PROXY = False
DEFAULT_SCOPE = '0'
DEFAULT_TIMEOUT = 3
DEFAULT_WARN = '180:'
DEFAULT_WARN_VOLTAGE = ''

# Highest `HEALTHSTATUS` code the REST Interface References document (offline). It
# bounds the metric so a graph scales to the whole enumeration; the thresholds stay
# out of the performance data, because whether a code is a fault depends on the
# object and is decided in `lib.huawei_dorado.get_health_status_state()`.
HEALTH_STATUS_MAX = 18

# Fields `--match` is applied to.
MATCH_FIELDS = ('UUID', 'LOCATION')

# RUNNINGSTATUS codes a healthy backup power module reports: normal (1), running (2),
# online (27), charging (48) and charging completed (49). A module cycles through the
# charging states in normal operation, so they are not a reason to alert.
OK_RUNNING_STATUS = (1, 2, 27, 48, 49)


def parse_args():
    """Parse command line arguments using argparse."""
    parser = argparse.ArgumentParser(
        description=DESCRIPTION,
        epilog=lib.args.epilog(__file__),
        formatter_class=lib.args.HelpFormatter,
    )

    parser.add_argument(
        '-V',
        '--version',
        action='version',
        version=f'%(prog)s: v{__version__} by {__author__}',
    )

    parser.add_argument(
        '--always-ok',
        help=lib.args.help('--always-ok'),
        dest='ALWAYS_OK',
        action='store_true',
        default=False,
    )

    parser.add_argument(
        '--cache-expire',
        help=lib.args.help('--cache-expire') + ' Default: %(default)s',
        dest='CACHE_EXPIRE',
        type=int,
        default=DEFAULT_CACHE_EXPIRE,
    )

    parser.add_argument(
        '-c',
        '--critical',
        help='CRIT threshold for the remaining life of a backup power module, as a Nagios '
        'range in days. Default: %(default)s',
        dest='CRIT',
        default=DEFAULT_CRIT,
    )

    parser.add_argument(
        '--critical-voltage',
        help=lib.args.help('--critical-voltage')
        + ' Supports Nagios ranges. '
        'Off by default, because the healthy range depends on the module and on how many '
        'cells it has; read the label or watch the graph first. '
        'Example: `--critical-voltage=13.5:18.5`',
        dest='CRIT_VOLTAGE',
        default=DEFAULT_CRIT_VOLTAGE,
    )

    parser.add_argument(
        '--device-id',
        help='Huawei OceanStor Dorado API device ID. '
        'Optional: the appliance reports its own at login, so this is only '
        'needed to override that answer.',
        dest='DEVICE_ID',
        default=DEFAULT_DEVICE_ID,
    )

    parser.add_argument(
        '--ignore',
        help='Skip backup power modules. '
        + lib.args.help('--ignore-regex')
        + ' The regex is anchored at the start of the string (Python `re.match`) and is '
        'matched against `UUID`, `LOCATION`, so prefix with `.*` to match anywhere.',
        dest='IGNORE',
        action='append',
        default=None,
    )

    parser.add_argument(
        '--insecure',
        help=lib.args.help('--insecure'),
        dest='INSECURE',
        action='store_true',
        default=DEFAULT_INSECURE,
    )

    parser.add_argument(
        '--lengthy',
        help=lib.args.help('--lengthy'),
        dest='LENGTHY',
        action='store_true',
        default=False,
    )

    parser.add_argument(
        '--no-insecure',
        help=lib.args.help('--no-insecure'),
        dest='INSECURE',
        action='store_false',
        default=DEFAULT_INSECURE,
    )

    parser.add_argument(
        '--match',
        help='Limit to backup power modules. '
        + lib.args.help('--match')
        + ' The regex is anchored at the start of the string (Python `re.match`) and is '
        'matched against `UUID`, `LOCATION`, so prefix with `.*` to match anywhere.',
        dest='MATCH',
        action='append',
        default=None,
    )

    parser.add_argument(
        '--no-match-severity',
        help=lib.args.help('--no-match-severity') + ' Default: %(default)s',
        dest='NO_MATCH_SEVERITY',
        choices=['ok', 'warn', 'crit', 'unknown'],
        default=DEFAULT_NO_MATCH_SEVERITY,
    )

    parser.add_argument(
        '--no-perfdata',
        help=lib.args.help('--no-perfdata'),
        dest='NO_PERFDATA',
        action='store_true',
        default=False,
    )

    parser.add_argument(
        '--no-proxy',
        help=lib.args.help('--no-proxy'),
        dest='NO_PROXY',
        action='store_true',
        default=DEFAULT_NO_PROXY,
    )

    parser.add_argument(
        '--password',
        help='Huawei OceanStor Dorado API password.',
        dest='PASSWORD',
    )

    parser.add_argument(
        '--password-file',
        help=lib.args.help('--password-file'),
        dest='PASSWORD_FILE',
    )

    parser.add_argument(
        '--scope',
        help='Huawei OceanStor Dorado API scope. Default: %(default)s',
        dest='SCOPE',
        default=DEFAULT_SCOPE,
    )

    parser.add_argument(
        '--test',
        help=lib.args.help('--test'),
        dest='TEST',
        type=lib.args.csv,
    )

    parser.add_argument(
        '--timeout',
        help=lib.args.help('--timeout') + ' Default: %(default)s (seconds)',
        dest='TIMEOUT',
        type=int,
        default=DEFAULT_TIMEOUT,
    )

    parser.add_argument(
        '-u',
        '--url',
        help='Huawei OceanStor Dorado API URL.',
        dest='URL',
        required=True,
    )

    parser.add_argument(
        '--username',
        help='Huawei OceanStor Dorado API username.',
        dest='USERNAME',
        required=True,
    )

    parser.add_argument(
        '-w',
        '--warning',
        help='WARN threshold for the remaining life of a backup power module, as a Nagios '
        'range in days. Default: %(default)s',
        dest='WARN',
        default=DEFAULT_WARN,
    )

    parser.add_argument(
        '--warning-voltage',
        help=lib.args.help('--warning-voltage')
        + ' Supports Nagios ranges. '
        'Off by default, because the healthy range depends on the module and on how many '
        'cells it has; read the label or watch the graph first. '
        'Example: `--warning-voltage=15:17`',
        dest='WARN_VOLTAGE',
        default=DEFAULT_WARN_VOLTAGE,
    )

    parser.add_argument(
        '-v',
        '--verbose',
        help=lib.args.help('--verbose')
        + ' Appends what every API request returned, so the appliance\'s own answers '
        'can be read while working out how it reports something. Session tokens are '
        'redacted. The output is as long as those answers are, so this is a debugging '
        'aid rather than something to leave switched on.',
        dest='VERBOSE',
        action='store_true',
        default=False,
    )

    args, _ = parser.parse_known_args()
    return args


def get_remaining_life_days(power):
    """Return a module's remaining life in days, or `None` if it does not report one."""
    # A module that does not track its remaining life answers with zero or a negative
    # value. Read as a measurement that would alert on every such module forever.
    try:
        days = int(power.get('REMAINLIFEDAYS'))
    except (TypeError, ValueError):
        return None
    if days <= 0:
        return None
    return days


def main():
    """The main function. This is where the magic happens."""

    # parse the command line
    try:
        args = parse_args()
    except SystemExit:
        sys.exit(STATE_UNKNOWN)

    if args.PASSWORD_FILE:
        args.PASSWORD = lib.args.load_secret(args.PASSWORD_FILE)
    if not args.PASSWORD:
        lib.base.cu('Provide the API password via --password or --password-file.')

    # set default values for append parameters that were not specified
    if args.IGNORE is None:
        args.IGNORE = []
    if args.MATCH is None:
        args.MATCH = []

    if not args.URL.startswith('http'):
        lib.base.cu('--url parameter has to start with "http://" or https://".')

    # fetch data
    if args.TEST is None:
        result = lib.huawei_dorado.get_data('backup_power', args)
    else:
        # do not call the command, put in test data
        stdout, _stderr, _retc = lib.lftest.test(args.TEST)
        result = json.loads(stdout)

    # no valuable result?
    lib.huawei_dorado.assert_ok(result, 'the backup power modules')

    # An appliance always has backup power modules, so an empty list is a query that
    # never reached them rather than an inventory that is genuinely empty.
    # Reporting OK here would hide the fault behind a green check.
    if not result.get('data'):
        lib.base.oao(
            f'{args.URL} reported no backup power modules.'
            ' Verify that the API user is allowed to query them.',
            STATE_UNKNOWN,
        )

    # init some vars
    msg = ''
    state = STATE_OK
    perfdata = ''
    compiled_match_regex = [
        lib.base.coe(item) for item in lib.txt.compile_regex(args.MATCH, '--match')
    ]
    compiled_ignore_regex = [
        lib.base.coe(item)
        for item in lib.txt.compile_regex(args.IGNORE, '--ignore')
    ]

    # analyze data
    table_data = []
    for power in result.get('data') or []:
        power['UUID'] = lib.huawei_dorado.get_uuid(power)
        # Perfdata labels carry the object's own TYPE:ID, which stays the same
        # when hardware is moved between slots, unlike its location.
        label = re.sub(r'\W+', '_', power['UUID'])

        if args.MATCH and not any(
            lib.base.coe(lib.txt.match_regex(pattern, str(power.get(field, ''))))
            for pattern in compiled_match_regex
            for field in MATCH_FIELDS
        ):
            continue

        if args.IGNORE and any(
            lib.base.coe(lib.txt.match_regex(pattern, str(power.get(field, ''))))
            for pattern in compiled_ignore_regex
            for field in MATCH_FIELDS
        ):
            continue

        health_state = lib.huawei_dorado.get_health_status_state(
            power.get('HEALTHSTATUS')
        )
        state = lib.base.get_worst(state, health_state)

        running_state = lib.huawei_dorado.get_running_status_state(
            power.get('RUNNINGSTATUS'), OK_RUNNING_STATUS
        )
        state = lib.base.get_worst(state, running_state)

        remaining_life_days = get_remaining_life_days(power)
        if remaining_life_days is not None:
            state = lib.base.get_worst(
                state,
                lib.base.get_state(
                    remaining_life_days, args.WARN, args.CRIT, _operator='range'
                ),
            )
            power['REMAINLIFEDAYS'] = remaining_life_days * 24 * 60 * 60
        else:
            power['REMAINLIFEDAYS'] = None
        # The REST Interface References document this field as "Unit: V/10", so a healthy
        # module reads as 16.1 rather than 161.
        voltage = lib.huawei_dorado.as_code(power.get('VOLTAGE'))
        power['VOLTAGE'] = None if voltage is None else round(voltage / 10, 1)
        # A module that has lost a cell, or one being trickle-charged back up, sits
        # outside its normal band long before its health status changes.
        voltage_state = STATE_OK
        if power['VOLTAGE'] is not None and (
            args.WARN_VOLTAGE or args.CRIT_VOLTAGE
        ):
            voltage_state = lib.base.get_state(
                power['VOLTAGE'],
                args.WARN_VOLTAGE or None,
                args.CRIT_VOLTAGE or None,
                _operator='range',
            )
            state = lib.base.get_worst(state, voltage_state)

        perfdata += lib.base.get_perfdata(
            f'{label}_health_status',
            power.get('HEALTHSTATUS'),
            uom=None,
            _min=0,
            _max=HEALTH_STATUS_MAX,
        )
        perfdata += lib.base.get_perfdata(
            f'{label}_running_status',
            power.get('RUNNINGSTATUS'),
            _min=0,
        )
        # No performance data for CHARGETIMES, the number of discharge cycles the module
        # has been through. It only ever counts up, and a cumulative counter aggregates
        # wrong in every Grafana panel that touches it
        # ([#320](https://github.com/Linuxfabrik/monitoring-plugins/issues/320)).
        # The value stays in the table, where it reads as the wear of the module.
        perfdata += (
            lib.base.get_perfdata(
                f'{label}_remaining_life',
                power['REMAINLIFEDAYS'],
                uom='s',
                _min=0,
            )
            if power['REMAINLIFEDAYS']
            else ''
        )
        perfdata += lib.base.get_perfdata(
            f'{label}_voltage',
            power.get('VOLTAGE'),
            uom=None,
            _min=0,
        )

        power['health'] = lib.huawei_dorado.get_health_status(
            power.get('HEALTHSTATUS')
        )
        power['running'] = lib.huawei_dorado.get_running_status(
            power.get('RUNNINGSTATUS')
        )
        # One state per row, in the last column. IcingaWeb replaces a state with an
        # icon, which shifts everything to the right of it, so a second one mid-row
        # would break the table. What decided it stays readable in the two columns
        # in front of it.
        power['state'] = lib.base.state2str(
            lib.base.get_worst(health_state, running_state), empty_ok=False
        )
        power['VOLTAGE'] = (
            '--'
            if power['VOLTAGE'] is None
            else f'{power["VOLTAGE"]}'
            f'{lib.base.state2str(voltage_state, prefix=" ")}'
        )
        # A module that does not track its remaining life gets a dash, as everywhere
        # else in the family. A literal `0` would read as a dead battery.
        power['REMAINLIFEDAYS'] = (
            '--'
            if not power['REMAINLIFEDAYS']
            else lib.human.seconds2human(power['REMAINLIFEDAYS'])
        )

        table_data.append(power)

    # The appliance listed backup power modules and the filter selected none of them.
    if not table_data:
        lib.base.oao(
            f'No backup power modules matched `{", ".join(args.MATCH)}`.',
            lib.base.str2state(args.NO_MATCH_SEVERITY),
            always_ok=args.ALWAYS_OK,
            no_perfdata=args.NO_PERFDATA,
        )

    # build the message
    if state == STATE_CRIT:
        msg += 'There are critical errors.'
    elif state == STATE_WARN:
        msg += 'There are warnings.'
    else:
        msg += 'Everything is ok.'
    msg += '\n\n'

    # build table output
    if table_data:
        if args.LENGTHY:
            keys = [
                'UUID',
                'LOCATION',
                'MANUFACTUREDDATE',
                'CONTROLLERID',
                'CHARGETIMES',
                'REMAINLIFEDAYS',
                'VOLTAGE',
                'health',
                'running',
                'state',
            ]
            headers = [
                'UUID',
                'Location',
                'Produced',
                'ControllerID',
                '#Discharged',
                'Remain',
                'Volt',
                'Health',
                'Running',
                'State',
            ]
        else:
            keys = [
                'UUID',
                'LOCATION',
                'REMAINLIFEDAYS',
                'VOLTAGE',
                'health',
                'running',
                'state',
            ]
            headers = [
                'UUID',
                'Location',
                'Remain',
                'Volt',
                'Health',
                'Running',
                'State',
            ]

        msg += lib.base.get_table(
            table_data, keys, header=headers, missing='--', hide_empty=True
        )

    if args.VERBOSE:
        msg += '\n\n' + lib.huawei_dorado.format_responses()

    # over and out
    lib.base.oao(
        msg, state, perfdata, always_ok=args.ALWAYS_OK, no_perfdata=args.NO_PERFDATA
    )


if __name__ == '__main__':
    try:
        main()
    except Exception:
        lib.base.cu()
