ec2-instances

// The EC2 instances of an AWS region, by the Amazon EC2 API with an access key: for each instance, its state
// and the results of its two status checks, the one of the AWS systems it runs on and its own.
// Reads GET https://ec2.<region>.amazonaws.com/?Action=DescribeInstanceStatus, page after page, signed by
// Signature Version 4.
need('protocol:http');
need('protocol:aws');
$region = variable('aws:region', 'text', array(
    'label' => 'Region',
    'help'  => 'The AWS region of the instances, as AWS writes it in an API: eu-west-3, us-east-1.',
));

per('instance', 'Instances');
output('state', 'text', array('label' => 'State', 'per' => 'instance'));
output('system-status', 'text', array('label' => 'System status check', 'per' => 'instance'));
output('instance-status', 'text', array('label' => 'Instance status check', 'per' => 'instance'));

watch.expect('not-running', 'Instance not running', 'state', array('critical' => array('running'), 'for' => '2 runs'));
watch.state('system-impaired', 'System status check failed', 'system-status', array('critical' => array('impaired')));
watch.state('instance-impaired', 'Instance status check failed', 'instance-status', array('critical' => array('impaired')));
watch.missing('missing', 'Instance missing', 'state', array('enabled' => false));

// One signed GET on an AWS API of the Query protocol (EC2, RDS, CloudWatch): the parameters in the address, the
// answer in XML. aws.sign gives the headers that carry the signature: the keys never leave. A refusal is told
// by the error code AWS writes in its answer; a throttled request or a fault of AWS is not one.
function aws_get($service, $region, $query, $what) {
    $url = 'https://' . $service . '.' . $region . '.amazonaws.com/?' . url.query($query);
    $answer = http.get($url, headers=aws.sign('GET', $url, array(), '', $region, $service));
    if ($answer['status'] != 200) {
        $code = '';
        $found = text.match($answer['body'], '<Code>([^<]+)</Code>');
        if (list.count($found) > 0) {
            $code = $found[1];
        }
        if ($code == 'Throttling' || $code == 'RequestLimitExceeded' || $answer['status'] >= 500) {
            fail('unreachable', 'AWS answers the reading of ' . $what . ' with code ' . $answer['status'] . ' ' . $code);
        }
        if ($answer['status'] == 400 || $answer['status'] == 401 || $answer['status'] == 403) {
            fail('misconfigured', 'AWS refuses to read ' . $what . ', code ' . $answer['status'] . ' ' . $code,
                'Check the access key, the permissions of its user or role, and the region.');
        }
        fail('unreachable', 'AWS answers the reading of ' . $what . ' with code ' . $answer['status']);
    }
    return xml.read($answer['body']);
}

// The first child of that name, array() when it is absent or empty. An empty element, such as a list without
// members written <name/>, reads as the text '' and not as an array.
function element($parent, $name) {
    $children = list.get($parent, $name, array());
    if (list.count($children) == 0 || !list.is($children[0])) {
        return array();
    }
    return $children[0];
}

// The text of the first child of that name, '' when it is absent.
function child_text($parent, $name) {
    $children = list.get($parent, $name, array());
    if (list.count($children) == 0) {
        return '';
    }
    return $children[0];
}

// The region goes into the name of the host: it must be a name.
if (list.count(text.match($region, '^[a-z0-9-]+$')) == 0) {
    fail('misconfigured', 'The region holds other characters than lower-case letters, digits and hyphens');
}

// <DescribeInstanceStatusResponse><requestId>…</requestId><instanceStatusSet><item><instanceId>i-0123…</instanceId>
// <instanceState><code>16</code><name>running</name></instanceState><systemStatus><status>ok</status>…
// </systemStatus><instanceStatus><status>ok</status>…</instanceStatus></item></instanceStatusSet>
// <nextToken>…</nextToken></DescribeInstanceStatusResponse>. IncludeAllInstances asks for the stopped
// instances too, which AWS leaves out by default.
$token = '';
while (true) {
    $query = array('Action' => 'DescribeInstanceStatus', 'Version' => '2016-11-15', 'IncludeAllInstances' => 'true');
    if ($token != '') {
        $query['NextToken'] = $token;
    }
    $answer = element(aws_get('ec2', $region, $query, 'the EC2 instances'), 'DescribeInstanceStatusResponse');
    foreach (list.get(element($answer, 'instanceStatusSet'), 'item', array()) as $item) {
        $name = child_text($item, 'instanceId');
        value('state', child_text(element($item, 'instanceState'), 'name'), $name);
        value('system-status', child_text(element($item, 'systemStatus'), 'status'), $name);
        value('instance-status', child_text(element($item, 'instanceStatus'), 'status'), $name);
    }
    $token = child_text($answer, 'nextToken');
    if ($token == '') {
        break;
    }
}