Deploymentcenter/src/Modules/Watchdog/MetricStore.php

<?php

declare(strict_types=1);

namespace Deploymentcenter\Modules\Watchdog;

use Deploymentcenter\Core\Logger;
use PDO;

/**
 * Verlauf der Heartbeat-Metriken.
 *
 * Bisher wurde metrics_json bei jedem Heartbeat ueberschrieben - es gab immer
 * nur den letzten Moment. Damit laesst sich "die Platte laeuft seit drei Tagen
 * voll" nicht erkennen, sondern nur "die Platte ist voll".
 *
 * Bewusst schmal gehalten: Das hier ist die Datengrundlage fuer Sparklines und
 * Abweichungsalarme im Dashboard, keine Zeitreihendatenbank. Wer echte Analyse
 * braucht, ist mit Prometheus besser bedient.
 */
final class MetricStore
{
    /** Aufbewahrung der Rohwerte in Tagen. */
    private const RETENTION_DAYS = 14;

    /** Maximale Anzahl Metriken je Heartbeat - schuetzt vor Ausreissern. */
    private const MAX_KEYS_PER_BEAT = 25;

    private PDO $db;

    public function __construct(PDO $db)
    {
        $this->db = $db;
    }

    /**
     * Nimmt die numerischen Werte eines Heartbeats auf.
     * Nicht numerische Werte werden uebergangen.
     *
     * @param mixed $metrics
     */
    public function record(string $source, string $instance, $metrics): int
    {
        if (!is_array($metrics) && !is_object($metrics)) {
            return 0;
        }

        $flat = self::flatten((array)$metrics);
        if ($flat === []) {
            return 0;
        }

        try {
            $stmt = $this->db->prepare('
                INSERT INTO watchdog_metrics (source, instance, metric_key, metric_value, recorded_utc)
                VALUES (:source, :instance, :metric_key, :metric_value, UTC_TIMESTAMP())
            ');

            $written = 0;
            foreach ($flat as $key => $value) {
                if ($written >= self::MAX_KEYS_PER_BEAT) {
                    break;
                }

                $stmt->execute([
                    ':source'       => mb_substr($source, 0, 100),
                    ':instance'     => mb_substr($instance, 0, 100),
                    ':metric_key'   => mb_substr($key, 0, 64),
                    ':metric_value' => $value,
                ]);
                $written++;
            }

            return $written;
        } catch (\Throwable $e) {
            // Der Verlauf ist Beiwerk; ein Heartbeat darf daran nicht scheitern.
            Logger::warning('Metriken nicht gespeichert', [
                'source' => $source,
                'error'  => $e->getMessage(),
            ]);
            return 0;
        }
    }

    /**
     * Verlauf einer Metrik, auf Zeitfenster verdichtet.
     *
     * @return list<array{bucket:string,avg:float,min:float,max:float,samples:int}>
     */
    public function history(
        string $source,
        string $metricKey,
        int $hours = 24,
        string $instance = 'default',
        int $bucketMinutes = 15
    ): array {
        $hours = max(1, min($hours, 24 * self::RETENTION_DAYS));
        $bucketMinutes = max(1, min($bucketMinutes, 1440));

        try {
            // Zeitstempel auf das Raster runden, damit gleichmaessige
            // Stuetzstellen entstehen.
            $stmt = $this->db->prepare('
                SELECT
                    FROM_UNIXTIME(FLOOR(UNIX_TIMESTAMP(recorded_utc) / (' . $bucketMinutes . ' * 60))
                                  * (' . $bucketMinutes . ' * 60)) AS bucket,
                    AVG(metric_value) AS avg_value,
                    MIN(metric_value) AS min_value,
                    MAX(metric_value) AS max_value,
                    COUNT(*)          AS samples
                FROM watchdog_metrics
                WHERE source = :source
                  AND instance = :instance
                  AND metric_key = :metric_key
                  AND recorded_utc > (UTC_TIMESTAMP() - INTERVAL ' . $hours . ' HOUR)
                GROUP BY bucket
                ORDER BY bucket ASC
            ');
            $stmt->execute([
                ':source'     => $source,
                ':instance'   => $instance,
                ':metric_key' => $metricKey,
            ]);

            $out = [];
            foreach ($stmt->fetchAll() ?: [] as $row) {
                $out[] = [
                    'bucket'  => (string)$row['bucket'],
                    'avg'     => (float)$row['avg_value'],
                    'min'     => (float)$row['min_value'],
                    'max'     => (float)$row['max_value'],
                    'samples' => (int)$row['samples'],
                ];
            }

            return $out;
        } catch (\Throwable $e) {
            Logger::warning('Metrik-Verlauf nicht abrufbar', ['error' => $e->getMessage()]);
            return [];
        }
    }

    /**
     * Welche Metriken liefert dieser Monitor ueberhaupt?
     *
     * @return list<string>
     */
    public function keysFor(string $source, string $instance = 'default'): array
    {
        try {
            $stmt = $this->db->prepare('
                SELECT DISTINCT metric_key
                FROM watchdog_metrics
                WHERE source = :source AND instance = :instance
                  AND recorded_utc > (UTC_TIMESTAMP() - INTERVAL 7 DAY)
                ORDER BY metric_key ASC
            ');
            $stmt->execute([':source' => $source, ':instance' => $instance]);
            return $stmt->fetchAll(PDO::FETCH_COLUMN) ?: [];
        } catch (\Throwable $e) {
            return [];
        }
    }

    /**
     * Vergleicht den aktuellen Wert mit dem eigenen Verlauf.
     *
     * Dieser Ansatz braucht kein Projektwissen: Statt fester Schwellwerte je
     * Anwendung wird gemeldet, was deutlich vom bisherigen Verhalten desselben
     * Monitors abweicht.
     *
     * @return array{deviates:bool,current:float,baseline:float,factor:float}|null
     */
    public function deviation(string $source, string $metricKey, string $instance = 'default', float $factor = 3.0): ?array
    {
        try {
            $stmt = $this->db->prepare('
                SELECT
                    (SELECT metric_value FROM watchdog_metrics
                     WHERE source = :s1 AND instance = :i1 AND metric_key = :k1
                     ORDER BY recorded_utc DESC LIMIT 1) AS current_value,
                    (SELECT AVG(metric_value) FROM watchdog_metrics
                     WHERE source = :s2 AND instance = :i2 AND metric_key = :k2
                       AND recorded_utc BETWEEN (UTC_TIMESTAMP() - INTERVAL 7 DAY)
                                            AND (UTC_TIMESTAMP() - INTERVAL 1 HOUR)) AS baseline_value
            ');
            $stmt->execute([
                ':s1' => $source, ':i1' => $instance, ':k1' => $metricKey,
                ':s2' => $source, ':i2' => $instance, ':k2' => $metricKey,
            ]);
            $row = $stmt->fetch();

            if (!is_array($row) || $row['current_value'] === null || $row['baseline_value'] === null) {
                return null;
            }

            $current = (float)$row['current_value'];
            $baseline = (float)$row['baseline_value'];

            if (abs($baseline) < 0.0001) {
                return null;
            }

            $ratio = $current / $baseline;

            return [
                'deviates' => $ratio >= $factor || $ratio <= (1 / $factor),
                'current'  => $current,
                'baseline' => $baseline,
                'factor'   => $ratio,
            ];
        } catch (\Throwable $e) {
            return null;
        }
    }

    /** Entfernt Werte, die aelter als die Aufbewahrungsfrist sind. */
    public function purge(): int
    {
        try {
            $stmt = $this->db->prepare(
                'DELETE FROM watchdog_metrics
                 WHERE recorded_utc < (UTC_TIMESTAMP() - INTERVAL ' . self::RETENTION_DAYS . ' DAY)
                 LIMIT 50000'
            );
            $stmt->execute();
            return $stmt->rowCount();
        } catch (\Throwable $e) {
            Logger::warning('Metrik-Bereinigung fehlgeschlagen', ['error' => $e->getMessage()]);
            return 0;
        }
    }

    /**
     * Verschachtelte Metriken flach klopfen: {"cpu":{"load":1.2}} -> "cpu.load".
     *
     * @return array<string,float>
     */
    private static function flatten(array $metrics, string $prefix = '', int $depth = 0): array
    {
        if ($depth > 3) {
            return [];
        }

        $out = [];
        foreach ($metrics as $key => $value) {
            if (!is_string($key) && !is_int($key)) {
                continue;
            }

            $name = $prefix === '' ? (string)$key : $prefix . '.' . $key;

            if (is_array($value)) {
                $out += self::flatten($value, $name, $depth + 1);
                continue;
            }

            if (is_bool($value)) {
                $out[$name] = $value ? 1.0 : 0.0;
                continue;
            }

            if (is_numeric($value)) {
                $out[$name] = (float)$value;
            }
        }

        return $out;
    }
}