Merge pull request #559 from domcyrus/fix-race-on-table-file

fix race condition on table.file #557

internal/table/file.go

@@ -124,24 +124,31 @@ func (f *File) reloader() {
 	for {
 		select {
 		case <-t.C:
+			f.reload()
+
+		case <-f.forceReload:
+			f.reload()
+
+		case <-f.stopReloader:
+			f.stopReloader <- struct{}{}
+			return
+		}
+	}
+}
+
+func (f *File) reload() {
 	info, err := os.Stat(f.file)
 	if err != nil {
 		if os.IsNotExist(err) {
 			f.mLck.Lock()
 			f.m = map[string][]string{}
-					f.mStamp = time.Now()
 			f.mLck.Unlock()
-					continue
+			return
 		}
 		f.log.Error("os stat", err)
 	}
-			if info.ModTime().Before(f.mStamp) {
-				continue // reload not necessary
-			}
-		case <-f.forceReload:
-		case <-f.stopReloader:
-			f.stopReloader <- struct{}{}
-			return
+	if info.ModTime().Before(f.mStamp) || time.Since(info.ModTime()) < (reloadInterval/2) {
+		return // reload not necessary
 	}
 
 	f.log.Debugf("reloading")
@@ -150,19 +157,29 @@ func (f *File) reloader() {
 	if err := readFile(f.file, newm); err != nil {
 		if os.IsNotExist(err) {
 			f.log.Printf("ignoring non-existent file: %s", f.file)
-				continue
+			return
 		}
 
 		f.log.Println(err)
-			continue
+		return
+	}
+	// after reading we need to check whether file has changed in between
+	info2, err := os.Stat(f.file)
+	if err != nil {
+		f.log.Println(err)
+		return
+	}
+
+	if !info2.ModTime().Equal(info.ModTime()) {
+		// file has changed in the meantime
+		return
 	}
 
 	f.mLck.Lock()
 	f.m = newm
-		f.mStamp = time.Now()
+	f.mStamp = info.ModTime()
 	f.mLck.Unlock()
 }
-}
 
 func (f *File) Close() error {
 	f.stopReloader <- struct{}{}

internal/table/file_test.go

@@ -111,27 +111,26 @@ func TestFileReload(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	// This delay is somehow important. Not sure why.
-	time.Sleep(500 * time.Millisecond)
+	// ensure it is correctly loaded at first time.
+	m.mLck.RLock()
+	if m.m["cat"] == nil {
+		t.Fatalf("wrong content loaded, new m were not loaded, %v", m.m)
+	}
+	m.mLck.RUnlock()
 
-	if err := ioutil.WriteFile(f.Name(), []byte("dog: cat"), os.ModePerm); err != nil {
+	for i := 0; i < 100; i++ {
+		// try to provoke race condition on file writing
+		if i%2 == 0 {
+			if err := os.WriteFile(f.Name(), []byte("dog: cat"), os.ModePerm); err != nil {
 				t.Fatal(err)
 			}
-
-	for i := 0; i < 10; i++ {
-		time.Sleep(reloadInterval + 50*time.Millisecond)
-		m.mLck.RLock()
-		if m.m["dog"] != nil {
-			m.mLck.RUnlock()
-			break
 		}
-		m.mLck.RUnlock()
-	}
-
+		time.Sleep(reloadInterval + 5*time.Millisecond)
 		m.mLck.RLock()
-	defer m.mLck.RUnlock()
 		if m.m["dog"] == nil {
-		t.Fatal("New m were not loaded")
+			t.Fatalf("wrong content loaded, new m were not loaded, %v", m.m)
+		}
+		m.mLck.RUnlock()
 	}
 }
 
@@ -208,9 +207,6 @@ func TestFileReload_Removed(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	// This delay is somehow important. Not sure why.
-	time.Sleep(250 * time.Millisecond)
-
 	os.Remove(f.Name())
 
 	time.Sleep(3 * reloadInterval)
@@ -223,5 +219,5 @@ func TestFileReload_Removed(t *testing.T) {
 }
 
 func init() {
-	reloadInterval = 250 * time.Millisecond
+	reloadInterval = 10 * time.Millisecond
 }