galaxy-game/mail/internal/worker/sqlretention.go

package worker

import (
	"context"
	"errors"
	"fmt"
	"log/slog"
	"time"
)

// SQLRetentionStore performs the durable DELETE statements applied by the
// retention worker. Implementations are typically the umbrella PostgreSQL
// mail store; the interface keeps the worker decoupled from the store
// package.
type SQLRetentionStore interface {
	// DeleteDeliveriesOlderThan removes deliveries whose created_at predates
	// cutoff. Cascading FKs drop attempts, dead_letters, delivery_payloads,
	// and delivery_recipients owned by the deleted rows.
	DeleteDeliveriesOlderThan(ctx context.Context, cutoff time.Time) (int64, error)

	// DeleteMalformedCommandsOlderThan removes malformed-command rows whose
	// recorded_at predates cutoff.
	DeleteMalformedCommandsOlderThan(ctx context.Context, cutoff time.Time) (int64, error)
}

// SQLRetentionConfig stores the dependencies and policy used by
// SQLRetentionWorker.
type SQLRetentionConfig struct {
	// Store applies the durable DELETE statements.
	Store SQLRetentionStore

	// DeliveryRetention bounds how long deliveries (and their cascaded
	// attempts/dead_letters/payloads/recipients) survive after creation.
	DeliveryRetention time.Duration

	// MalformedCommandRetention bounds how long malformed-command rows
	// survive after recorded_at.
	MalformedCommandRetention time.Duration

	// CleanupInterval stores the wall-clock period between two retention
	// passes.
	CleanupInterval time.Duration

	// Clock provides the wall-clock used to compute cutoff timestamps.
	Clock Clock
}

// SQLRetentionWorker periodically deletes deliveries and malformed-command
// rows whose retention window has expired. The worker replaces the previous
// Redis index_cleaner that maintained secondary index keys; PostgreSQL
// indexes are maintained by the engine, so the worker only needs to enforce
// retention.
type SQLRetentionWorker struct {
	store                     SQLRetentionStore
	deliveryRetention         time.Duration
	malformedCommandRetention time.Duration
	cleanupInterval           time.Duration
	clock                     Clock
	logger                    *slog.Logger
}

// NewSQLRetentionWorker constructs the periodic retention worker.
func NewSQLRetentionWorker(cfg SQLRetentionConfig, logger *slog.Logger) (*SQLRetentionWorker, error) {
	switch {
	case cfg.Store == nil:
		return nil, errors.New("new sql retention worker: nil store")
	case cfg.DeliveryRetention <= 0:
		return nil, errors.New("new sql retention worker: non-positive delivery retention")
	case cfg.MalformedCommandRetention <= 0:
		return nil, errors.New("new sql retention worker: non-positive malformed command retention")
	case cfg.CleanupInterval <= 0:
		return nil, errors.New("new sql retention worker: non-positive cleanup interval")
	case cfg.Clock == nil:
		return nil, errors.New("new sql retention worker: nil clock")
	}
	if logger == nil {
		logger = slog.Default()
	}

	return &SQLRetentionWorker{
		store:                     cfg.Store,
		deliveryRetention:         cfg.DeliveryRetention,
		malformedCommandRetention: cfg.MalformedCommandRetention,
		cleanupInterval:           cfg.CleanupInterval,
		clock:                     cfg.Clock,
		logger:                    logger.With("component", "sql_retention_worker"),
	}, nil
}

// Run starts the retention loop and blocks until ctx is canceled.
func (worker *SQLRetentionWorker) Run(ctx context.Context) error {
	if ctx == nil {
		return errors.New("run sql retention worker: nil context")
	}
	if err := ctx.Err(); err != nil {
		return err
	}
	if worker == nil {
		return errors.New("run sql retention worker: nil worker")
	}

	worker.logger.Info("sql retention worker started",
		"delivery_retention", worker.deliveryRetention.String(),
		"malformed_command_retention", worker.malformedCommandRetention.String(),
		"cleanup_interval", worker.cleanupInterval.String(),
	)
	defer worker.logger.Info("sql retention worker stopped")

	// First pass runs immediately so a freshly started service does not wait
	// one full interval before evicting stale rows.
	worker.runOnce(ctx)

	ticker := time.NewTicker(worker.cleanupInterval)
	defer ticker.Stop()

	for {
		select {
		case <-ctx.Done():
			return ctx.Err()
		case <-ticker.C:
			worker.runOnce(ctx)
		}
	}
}

// Shutdown stops the retention worker within ctx.
func (worker *SQLRetentionWorker) Shutdown(ctx context.Context) error {
	if ctx == nil {
		return errors.New("shutdown sql retention worker: nil context")
	}
	return nil
}

func (worker *SQLRetentionWorker) runOnce(ctx context.Context) {
	now := worker.clock.Now().UTC()

	deliveryCutoff := now.Add(-worker.deliveryRetention)
	if deleted, err := worker.store.DeleteDeliveriesOlderThan(ctx, deliveryCutoff); err != nil {
		worker.logger.Warn("delete expired deliveries failed",
			"cutoff", deliveryCutoff,
			"error", fmt.Sprintf("%v", err),
		)
	} else if deleted > 0 {
		worker.logger.Info("expired deliveries deleted",
			"cutoff", deliveryCutoff,
			"deleted", deleted,
		)
	}

	malformedCutoff := now.Add(-worker.malformedCommandRetention)
	if deleted, err := worker.store.DeleteMalformedCommandsOlderThan(ctx, malformedCutoff); err != nil {
		worker.logger.Warn("delete expired malformed commands failed",
			"cutoff", malformedCutoff,
			"error", fmt.Sprintf("%v", err),
		)
	} else if deleted > 0 {
		worker.logger.Info("expired malformed commands deleted",
			"cutoff", malformedCutoff,
			"deleted", deleted,
		)
	}
}