galaxy-game/notification/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
// notification store; the interface keeps the worker decoupled from the
// store package.
type SQLRetentionStore interface {
	// DeleteRecordsOlderThan removes records rows whose accepted_at predates
	// cutoff. Cascading FKs drop routes and dead_letters owned by the deleted
	// rows.
	DeleteRecordsOlderThan(ctx context.Context, cutoff time.Time) (int64, error)

	// DeleteMalformedIntentsOlderThan removes malformed-intent rows whose
	// recorded_at predates cutoff.
	DeleteMalformedIntentsOlderThan(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

	// RecordRetention bounds how long records (and their cascaded routes and
	// dead_letters) survive after acceptance.
	RecordRetention time.Duration

	// MalformedIntentRetention bounds how long malformed-intent rows survive
	// after recorded_at.
	MalformedIntentRetention 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 records and malformed-intent rows
// whose retention window has expired. The worker replaces the per-key
// Redis EXPIRE eviction that maintained TTLs on the previous Redis-backed
// notification keyspace.
type SQLRetentionWorker struct {
	store                    SQLRetentionStore
	recordRetention          time.Duration
	malformedIntentRetention 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.RecordRetention <= 0:
		return nil, errors.New("new sql retention worker: non-positive record retention")
	case cfg.MalformedIntentRetention <= 0:
		return nil, errors.New("new sql retention worker: non-positive malformed intent 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,
		recordRetention:          cfg.RecordRetention,
		malformedIntentRetention: cfg.MalformedIntentRetention,
		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",
		"record_retention", worker.recordRetention.String(),
		"malformed_intent_retention", worker.malformedIntentRetention.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()

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

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