From 6f676d1bd412ca89355cfa7b4df741cd88fb4453 Mon Sep 17 00:00:00 2001
From: Goon <duy@wearetopgroup.com>
Date: Sun, 31 May 2026 17:01:35 +0700
Subject: [PATCH] fix(cron): wait for job goroutines on stop

---
 internal/cron/service.go           | 12 ++++++++++--
 internal/cron/service_execution.go |  2 ++
 2 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/internal/cron/service.go b/internal/cron/service.go
index 0e8d7a03..eee8765f 100644
--- a/internal/cron/service.go
+++ b/internal/cron/service.go
@@ -13,6 +13,7 @@ type Service struct {
 	onJob     JobHandler
 	running   bool
 	stopChan  chan struct{}
+	wg        sync.WaitGroup
 	mu        sync.Mutex
 	runLog    []RunLogEntry // in-memory run history (last 200 entries)
 	retryCfg  RetryConfig   // retry config for failed jobs
@@ -89,7 +90,11 @@ func (cs *Service) Start() error {
 	// after a previous Stop() returned but the runLoop goroutine hasn't yet
 	// executed its ticker construction.
 	tick := runLoopTickInterval
-	go cs.runLoop(cs.stopChan, tick)
+	cs.wg.Add(1)
+	go func() {
+		defer cs.wg.Done()
+		cs.runLoop(cs.stopChan, tick)
+	}()
 
 	slog.Info("cron service started", "jobs", len(cs.store.Jobs))
 	return nil
@@ -98,14 +103,17 @@ func (cs *Service) Start() error {
 // Stop halts the scheduling loop.
 func (cs *Service) Stop() {
 	cs.mu.Lock()
-	defer cs.mu.Unlock()
 
 	if !cs.running {
+		cs.mu.Unlock()
 		return
 	}
 
 	close(cs.stopChan)
 	cs.running = false
+	cs.mu.Unlock()
+
+	cs.wg.Wait()
 	slog.Info("cron service stopped")
 }
 
diff --git a/internal/cron/service_execution.go b/internal/cron/service_execution.go
index fd8a7a5d..fdd13485 100644
--- a/internal/cron/service_execution.go
+++ b/internal/cron/service_execution.go
@@ -216,7 +216,9 @@ func (cs *Service) checkJobs() {
 	// agent loop stuck), the entire cron scheduler would stop checking for new
 	// due jobs. Now each job runs independently with panic recovery.
 	for _, dj := range dueJobs {
+		cs.wg.Add(1)
 		go func(id string, scheduledAtMS int64) {
+			defer cs.wg.Done()
 			defer safego.Recover(nil, "job_id", id)
 			cs.executeJobByID(id, scheduledAtMS)
 		}(dj.id, dj.scheduledAtMS)