Merge pull request #143 from chaunceyjiang/mutex

reduce race condition. replace mutex to RWmutex

pkg/scheduler/nodes.go

@@ -49,7 +49,7 @@ type NodeUsage struct {
 
 type nodeManager struct {
 	nodes map[string]*NodeInfo
-	mutex sync.Mutex
+	mutex sync.RWMutex
 }
 
 func (m *nodeManager) init() {
@@ -101,14 +101,16 @@ func (m *nodeManager) rmNodeDevice(nodeID string, nodeInfo *NodeInfo) {
 }
 
 func (m *nodeManager) GetNode(nodeID string) (*NodeInfo, error) {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
+	m.mutex.RLock()
+	defer m.mutex.RUnlock()
 	if n, ok := m.nodes[nodeID]; ok {
 		return n, nil
 	}
 	return &NodeInfo{}, fmt.Errorf("node %v not found", nodeID)
 }
 
 func (m *nodeManager) ListNodes() (map[string]*NodeInfo, error) {
+	m.mutex.RLock()
+	defer m.mutex.RUnlock()
 	return m.nodes, nil
 }

pkg/scheduler/pods.go

@@ -36,7 +36,7 @@ type podInfo struct {
 
 type podManager struct {
 	pods  map[k8stypes.UID]*podInfo
-	mutex sync.Mutex
+	mutex sync.RWMutex
 }
 
 func (m *podManager) init() {
@@ -65,6 +65,8 @@ func (m *podManager) delPod(pod *corev1.Pod) {
 }
 
 func (m *podManager) GetScheduledPods() (map[k8stypes.UID]*podInfo, error) {
+	m.mutex.RLock()
+	defer m.mutex.RUnlock()
 	klog.Infof("Getting all scheduled pods with %d nums", len(m.pods))
 	return m.pods, nil
 }

pkg/scheduler/score.go

@@ -88,7 +88,7 @@ func fitInCertainDevice(node *NodeUsage, request util.ContainerDeviceRequest, an
 	originReq := k.Nums
 	prevnuma := -1
 	klog.InfoS("Allocating device for container request", "pod", klog.KObj(pod), "card request", k)
-	tmpDevs := []util.ContainerDevice{}
+	var tmpDevs []util.ContainerDevice
 	for i := len(node.Devices) - 1; i >= 0; i-- {
 		klog.InfoS("scoring pod", "pod", klog.KObj(pod), "Memreq", k.Memreq, "MemPercentagereq", k.MemPercentagereq, "Coresreq", k.Coresreq, "Nums", k.Nums, "device index", i, "device", node.Devices[i].Id)
 		found, numa := checkType(annos, *node.Devices[i], k)