thermal

// The temperatures and fans of a server, by the Redfish service of its management card (DMTF Redfish): the
// reading of each temperature sensor and each fan, and its health, as the card judges it against its own
// thresholds. Any card that serves Redfish: Dell iDRAC, HPE iLO, Supermicro, Lenovo XCC, OpenBMC.
// A chassis is read by its Thermal resource, which the cards serve, deprecated since schema 2020.4; a chassis
// that serves only the ThermalSubsystem of the later schemas is read by it. Thermal is preferred when both
// are served: it gives each temperature sensor its own health.
need('protocol:http');
$address = need('host:address');
$user = need('protocol:http:user');
$password = need('protocol:http:password');

per('sensor', 'Sensors');
output('health', 'text', array('label' => 'Health', 'per' => 'sensor'));
output('temperature', 'number', array('label' => 'Temperature', 'unit' => '°C', 'per' => 'sensor'));
output('fan-speed', 'number', array('label' => 'Fan speed', 'unit' => 'rpm', 'per' => 'sensor'));
output('fan-percent', 'number', array('label' => 'Fan speed', 'unit' => '%', 'per' => 'sensor'));

watch.state('unhealthy', 'Sensor health degraded', 'health', array('warning' => array('warning'), 'critical' => array('critical')));
watch.above('hot', 'High temperature', 'temperature', array(
    'help'    => 'Raises above the temperature set on the probe. The card already judges each sensor against its own thresholds, in its health.',
    'enabled' => false,
));
watch.below('fan-slow', 'Fan slow', 'fan-speed', array(
    'help'    => 'Raises below the speed set on the probe, in rpm.',
    'enabled' => false,
));
watch.missing('missing', 'Sensor missing', 'health', array('enabled' => false));

// Returns one Redfish resource, read by its path (/redfish/v1/...), the account in Basic. The password only
// goes in the Basic header: the answer is not secret.
function redfish_get($address, $path, $user, $password) {
    $answer = http.get('https://' . $address . $path, headers=array('Accept' => 'application/json'), user=$user, password=$password);
    if ($answer['status'] == 401 || $answer['status'] == 403) {
        fail('misconfigured', 'The Redfish service refuses the account, code ' . $answer['status'], $path);
    }
    if ($answer['status'] != 200) {
        fail('unreachable', 'The Redfish service answers code ' . $answer['status'], $path);
    }
    return json.read($answer['body']);
}

// Returns the health of a resource, in lower case (ok, warning, critical): the health it rolls up from what
// is under it when it gives one, its own otherwise, unknown when it gives neither.
function health_of($resource) {
    $status = list.get($resource, 'Status', array());
    $health = list.get($status, 'HealthRollup', '');
    if ($health == '') {
        $health = list.get($status, 'Health', '');
    }
    if ($health == '') {
        return 'unknown';
    }
    return text.lower($health);
}

// Returns the instance name of each item: its name, or, when another item bears the same name, its name
// followed by its id.
function instance_names($items) {
    $counts = array();
    foreach ($items as $item) {
        $counts[$item['name']] = list.get($counts, $item['name'], 0) + 1;
    }
    $names = array();
    foreach ($items as $key => $item) {
        $names[$key] = $item['name'];
        if ($item['name'] == '' || $counts[$item['name']] > 1) {
            $names[$key] = text.trim($item['name'] . ' ' . $item['id']);
        }
        // A card that gives neither a name nor an id (both null, read as '').
        if ($names[$key] == '') {
            $names[$key] = 'unnamed ' . ($key + 1);
        }
    }
    return $names;
}

// Each sensor, as array('name', 'id'), and what it reads: 'health', 'temperature', 'rpm', 'percent'.
$items = array();
$chassis_list = redfish_get($address, '/redfish/v1/Chassis', $user, $password);
foreach (list.get($chassis_list, 'Members', array()) as $member) {
    $chassis = redfish_get($address, $member['@odata.id'], $user, $password);
    $chassis_id = list.get($chassis, 'Id', $member['@odata.id']);

    // Thermal: {"Temperatures": [{"MemberId": "0", "Name": "CPU1 Temp", "ReadingCelsius": 45, "Status":
    // {"State": "Enabled", "Health": "OK"}}], "Fans": [{"MemberId": "0", "Name": "Fan 1", "Reading": 5880,
    // "ReadingUnits": "RPM", "Status": {...}}]}. A fan is named FanName before Thermal 1.1. A sensor the
    // card says absent is an empty slot.
    if (list.has($chassis, 'Thermal')) {
        $thermal = redfish_get($address, $chassis['Thermal']['@odata.id'], $user, $password);
        foreach (list.get($thermal, 'Temperatures', array()) as $sensor) {
            if (list.get(list.get($sensor, 'Status', array()), 'State', '') == 'Absent') {
                continue;
            }
            $item = array('name' => list.get($sensor, 'Name', 'Temperature'), 'id' => list.get($sensor, 'MemberId', ''),
                'health' => health_of($sensor));
            if (list.get($sensor, 'ReadingCelsius', '') != '') {
                $item['temperature'] = $sensor['ReadingCelsius'];
            }
            $items = list.add($items, $item);
        }
        foreach (list.get($thermal, 'Fans', array()) as $fan) {
            if (list.get(list.get($fan, 'Status', array()), 'State', '') == 'Absent') {
                continue;
            }
            $item = array('name' => list.get($fan, 'Name', list.get($fan, 'FanName', 'Fan')),
                'id' => list.get($fan, 'MemberId', ''), 'health' => health_of($fan));
            if (list.get($fan, 'Reading', '') != '') {
                if (list.get($fan, 'ReadingUnits', 'RPM') == 'Percent') {
                    $item['percent'] = $fan['Reading'];
                } else {
                    $item['rpm'] = $fan['Reading'];
                }
            }
            $items = list.add($items, $item);
        }
    } elseif (list.has($chassis, 'ThermalSubsystem')) {
        // ThermalSubsystem: {"Fans": {"@odata.id": ".../Fans"}, "ThermalMetrics": {"@odata.id": "..."},
        // "Status": {...}}; each fan a resource of its own, {"Id": "1", "Name": "Fan 1", "SpeedPercent":
        // {"Reading": 45, "SpeedRPM": 5880}, "Status": {...}}. ThermalMetrics gives the temperatures without a
        // health of their own: {"TemperatureReadingsCelsius": [{"DeviceName": "CPU1", "Reading": 45,
        // "DataSourceUri": "..."}]}; the subsystem's own health stands for them.
        $subsystem = redfish_get($address, $chassis['ThermalSubsystem']['@odata.id'], $user, $password);
        $items = list.add($items, array('name' => 'Thermal subsystem', 'id' => $chassis_id, 'health' => health_of($subsystem)));
        if (list.has($subsystem, 'Fans')) {
            $fans = redfish_get($address, $subsystem['Fans']['@odata.id'], $user, $password);
            foreach (list.get($fans, 'Members', array()) as $link) {
                $fan = redfish_get($address, $link['@odata.id'], $user, $password);
                if (list.get(list.get($fan, 'Status', array()), 'State', '') == 'Absent') {
                    continue;
                }
                $item = array('name' => list.get($fan, 'Name', 'Fan'), 'id' => list.get($fan, 'Id', ''),
                    'health' => health_of($fan));
                $speed = list.get($fan, 'SpeedPercent', array());
                if (list.get($speed, 'Reading', '') != '') {
                    $item['percent'] = $speed['Reading'];
                }
                if (list.get($speed, 'SpeedRPM', '') != '') {
                    $item['rpm'] = $speed['SpeedRPM'];
                }
                $items = list.add($items, $item);
            }
        }
        if (list.has($subsystem, 'ThermalMetrics')) {
            $metrics = redfish_get($address, $subsystem['ThermalMetrics']['@odata.id'], $user, $password);
            foreach (list.get($metrics, 'TemperatureReadingsCelsius', array()) as $sensor) {
                if (list.get($sensor, 'Reading', '') == '') {
                    continue;
                }
                $items = list.add($items, array('name' => list.get($sensor, 'DeviceName', 'Temperature'),
                    'id' => list.get($sensor, 'DataSourceUri', ''), 'temperature' => $sensor['Reading']));
            }
        }
    }
}

if (list.count($items) == 0) {
    fail('misconfigured', 'The Redfish service serves no temperature and no fan',
        'No chassis has a Thermal or a ThermalSubsystem resource.');
}
$names = instance_names($items);
foreach ($items as $key => $item) {
    $name = $names[$key];
    if (list.has($item, 'health')) {
        value('health', $item['health'], $name);
    }
    if (list.has($item, 'temperature')) {
        value('temperature', $item['temperature'], $name);
    }
    if (list.has($item, 'rpm')) {
        value('fan-speed', $item['rpm'], $name);
    }
    if (list.has($item, 'percent')) {
        value('fan-percent', $item['percent'], $name);
    }
}