From d83efe24ef051786d6839b2c920fb70e3ef27db2 Mon Sep 17 00:00:00 2001
From: Denozordec <shatskiy.blog@gmail.com>
Date: Tue, 7 Apr 2026 00:03:46 +0700
Subject: [PATCH] refactor: optimize job listing and counting logic for
 concurrency safety. Separate tenant filtering and status checks to reduce
 lock contention, improving performance and ensuring accurate job status
 retrieval.

---
 internal/jobs/job.go | 43 ++++++++++++++++++++++++++++++-------------
 1 file changed, 30 insertions(+), 13 deletions(-)

diff --git a/internal/jobs/job.go b/internal/jobs/job.go
index b33f5a3..da70f73 100644
--- a/internal/jobs/job.go
+++ b/internal/jobs/job.go
@@ -229,22 +229,32 @@ func (r *Registry) List(tenantID, statusFilter, kindFilter, cursor string, limit
 	if limit <= 0 {
 		limit = 50
 	}
+	// Build candidate set under r.mu, but read mutable status fields (j.Status) outside
+	// of r.mu so we don't race with Job.mu-protected writes (MarkRunning/Succeed/Fail).
 	r.mu.RLock()
-	var all []*Job
+	all := make([]*Job, 0, len(r.byID))
 	for _, j := range r.byID {
-		if j.TenantID != tenantID {
-			continue
+		if j.TenantID == tenantID {
+			all = append(all, j)
 		}
-		if statusFilter != "" && j.Status != statusFilter {
-			continue
-		}
-		if kindFilter != "" && j.Kind != kindFilter {
-			continue
-		}
-		all = append(all, j)
 	}
 	r.mu.RUnlock()
 
+	// Apply filters outside of r.mu to synchronize with Job.mu.
+	if statusFilter != "" || kindFilter != "" {
+		filtered := all[:0]
+		for _, j := range all {
+			if kindFilter != "" && j.Kind != kindFilter {
+				continue
+			}
+			if statusFilter != "" && j.statusLocked() != statusFilter {
+				continue
+			}
+			filtered = append(filtered, j)
+		}
+		all = filtered
+	}
+
 	sort.Slice(all, func(i, j int) bool {
 		return all[i].CreatedAt.After(all[j].CreatedAt)
 	})
@@ -273,9 +283,9 @@ func (r *Registry) CountOtherActiveModuleRefresh(tenantID, excludeJobID string)
 	if r == nil {
 		return 0
 	}
+	// Two-phase: snapshot job pointers under r.mu, then read j.Status under Job.mu.
 	r.mu.RLock()
-	defer r.mu.RUnlock()
-	n := 0
+	candidates := make([]*Job, 0, 8)
 	for _, j := range r.byID {
 		if j.TenantID != tenantID || j.Kind != KindModuleRefresh {
 			continue
@@ -283,7 +293,14 @@ func (r *Registry) CountOtherActiveModuleRefresh(tenantID, excludeJobID string)
 		if j.ID == excludeJobID {
 			continue
 		}
-		if j.Status == StatusQueued || j.Status == StatusRunning {
+		candidates = append(candidates, j)
+	}
+	r.mu.RUnlock()
+
+	n := 0
+	for _, j := range candidates {
+		st := j.statusLocked()
+		if st == StatusQueued || st == StatusRunning {
 			n++
 		}
 	}