rdev/internal/worker/resource_gc.go

package worker

import (
	"context"
	"database/sql"
	"sync"
	"time"

	"github.com/orchard9/rdev/internal/logging"
)

// resourceReconciler abstracts the K8s operations needed by the GC worker.
// The concrete deployer.Deployer satisfies this interface implicitly.
type resourceReconciler interface {
	ListProjectLabels(ctx context.Context) ([]string, error)
	GetOldestResourceTime(ctx context.Context, projectName string) (time.Time, bool, error)
	UndeployAll(ctx context.Context, projectName string) error
}

// projectChecker determines whether a project exists in the database.
type projectChecker interface {
	projectExists(ctx context.Context, projectID string) (bool, error)
}

// dbProjectChecker checks project existence via SQL.
type dbProjectChecker struct {
	db *sql.DB
}

func (c *dbProjectChecker) projectExists(ctx context.Context, projectID string) (bool, error) {
	var exists bool
	err := c.db.QueryRowContext(ctx, "SELECT EXISTS(SELECT 1 FROM projects WHERE id = $1)", projectID).Scan(&exists)
	return exists, err
}

// ResourceGCConfig holds configuration for the GC worker.
type ResourceGCConfig struct {
	// MinAge is the minimum resource age before deletion. Default: 1 hour.
	MinAge time.Duration
	// ReconcileInterval is how often to run reconciliation. Default: 15 minutes.
	ReconcileInterval time.Duration
}

// DefaultResourceGCConfig returns sensible defaults.
func DefaultResourceGCConfig() *ResourceGCConfig {
	return &ResourceGCConfig{
		MinAge:            1 * time.Hour,
		ReconcileInterval: 15 * time.Minute,
	}
}

// ResourceGC periodically finds K8s resources whose project label doesn't
// match any project in the database, and deletes them after a safety window.
type ResourceGC struct {
	reconciler resourceReconciler
	checker    projectChecker
	config     *ResourceGCConfig

	ctx    context.Context
	cancel context.CancelFunc
	wg     sync.WaitGroup
}

// NewResourceGC creates a new resource GC worker.
// The reconciler must implement ListProjectLabels, GetOldestResourceTime, and UndeployAll
// (deployer.Deployer satisfies this). If cfg is nil, defaults are used.
func NewResourceGC(reconciler resourceReconciler, db *sql.DB, cfg *ResourceGCConfig) *ResourceGC {
	if cfg == nil {
		cfg = DefaultResourceGCConfig()
	}
	ctx, cancel := context.WithCancel(context.Background())
	return &ResourceGC{
		reconciler: reconciler,
		checker:    &dbProjectChecker{db: db},
		config:     cfg,
		ctx:        ctx,
		cancel:     cancel,
	}
}

// newResourceGCWithChecker creates a ResourceGC with a custom project checker (for testing).
func newResourceGCWithChecker(reconciler resourceReconciler, checker projectChecker, cfg *ResourceGCConfig) *ResourceGC {
	if cfg == nil {
		cfg = DefaultResourceGCConfig()
	}
	ctx, cancel := context.WithCancel(context.Background())
	return &ResourceGC{
		reconciler: reconciler,
		checker:    checker,
		config:     cfg,
		ctx:        ctx,
		cancel:     cancel,
	}
}

// Start begins the reconciliation loop.
func (g *ResourceGC) Start() {
	log := logging.FromContext(g.ctx).WithWorker("resource-gc")
	log.Info("resource GC started",
		"min_age", g.config.MinAge,
		"reconcile_interval", g.config.ReconcileInterval,
	)

	g.wg.Add(1)
	go g.reconcileLoop()
}

// Stop gracefully shuts down the GC worker.
func (g *ResourceGC) Stop() {
	log := logging.FromContext(g.ctx).WithWorker("resource-gc")
	log.Info("resource GC stopping")
	g.cancel()
	g.wg.Wait()
	log.Info("resource GC stopped")
}

func (g *ResourceGC) reconcileLoop() {
	defer g.wg.Done()

	// Run immediately on start
	g.runReconciliation()

	ticker := time.NewTicker(g.config.ReconcileInterval)
	defer ticker.Stop()

	for {
		select {
		case <-g.ctx.Done():
			return
		case <-ticker.C:
			g.runReconciliation()
		}
	}
}

func (g *ResourceGC) runReconciliation() {
	ctx, cancel := context.WithTimeout(g.ctx, TimeoutWorkExecution)
	defer cancel()

	log := logging.FromContext(ctx).WithWorker("resource-gc")

	labels, err := g.reconciler.ListProjectLabels(ctx)
	if err != nil {
		log.Error("failed to list project labels", logging.FieldError, err)
		return
	}

	var deleted, skipped, errCount int
	var firstErr error
	for _, project := range labels {
		exists, err := g.checker.projectExists(ctx, project)
		if err != nil {
			log.Error("failed to check project existence",
				logging.FieldProjectID, project,
				logging.FieldError, err,
			)
			continue
		}
		if exists {
			continue
		}

		// Project not in DB — check resource age before deleting
		oldest, found, err := g.reconciler.GetOldestResourceTime(ctx, project)
		if err != nil {
			log.Error("failed to get resource age",
				logging.FieldProjectID, project,
				logging.FieldError, err,
			)
			continue
		}
		if !found {
			continue
		}

		age := time.Since(oldest)
		if age < g.config.MinAge {
			log.Info("skipping young orphan",
				logging.FieldProjectID, project,
				"age", age,
				"min_age", g.config.MinAge,
			)
			skipped++
			continue
		}

		log.Info("deleting orphaned resources",
			logging.FieldProjectID, project,
			"age", age,
		)
		if err := g.reconciler.UndeployAll(ctx, project); err != nil {
			log.Error("failed to delete orphaned resources",
				logging.FieldProjectID, project,
				logging.FieldError, err,
			)
			errCount++
			if firstErr == nil {
				firstErr = err
			}
			continue
		}
		deleted++
	}

	if deleted > 0 || skipped > 0 || errCount > 0 {
		log.Info("reconciliation complete",
			"deleted", deleted,
			"skipped", skipped,
			"errors", errCount,
			"total_labels", len(labels),
		)
	}

}