set node status before remove node (#645)

* set node status before remove node

* fix metric help message

* fix UT issue

Author: yuyang0

cluster/calcium/node.go

@@ -87,6 +87,12 @@ func (c *Calcium) RemoveNode(ctx context.Context, nodename string) error {
 		return utils.Txn(ctx,
 			// if: remove node metadata
 			func(ctx context.Context) error {
+				// we need set node status here, consider the following scenery:
+				// the node is down, so the node status doesn't exist in ETCD,
+				// if we don't set node status here, other core instances will not be notified when the node is removed
+				if err = c.store.SetNodeStatus(ctx, node, 90); err != nil {
+					logger.Warnf(ctx, "failed to set node status: %s", err)
+				}
 				if err := c.store.RemoveNode(ctx, node); err != nil {
 					return err
 				}

cluster/calcium/node_test.go

@@ -90,12 +90,14 @@ func TestRemoveNode(t *testing.T) {
 
 	// fail by store.RemoveNode
 	store.On("ListNodeWorkloads", mock.Anything, mock.Anything, mock.Anything).Return([]*types.Workload{}, nil)
+	store.On("SetNodeStatus", mock.Anything, mock.Anything, mock.Anything).Return(nil).Once()
 	store.On("RemoveNode", mock.Anything, mock.Anything).Return(types.ErrMockError).Once()
 	assert.Error(t, c.RemoveNode(ctx, name))
 
 	// success
+	store.On("SetNodeStatus", mock.Anything, mock.Anything, mock.Anything).Return(nil).Once()
 	store.On("RemoveNode", mock.Anything, mock.Anything).Return(nil)
-	store.On("SetNodeStatus", mock.Anything, mock.Anything, int64(-1)).Return(nil)
+	store.On("SetNodeStatus", mock.Anything, mock.Anything, int64(-1)).Return(nil).Once()
 	rmgr := c.rmgr.(*resourcemocks.Manager)
 	rmgr.On("RemoveNode", mock.Anything, mock.Anything).Return(nil)
 	assert.NoError(t, c.RemoveNode(ctx, name))

metrics/metrics.go

@@ -225,7 +225,7 @@ func InitMetrics(ctx context.Context, config types.Config, metricsDescriptions [
 
 	Client.Collectors[podNodeStatusName] = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: podNodeStatusName,
-		Help: "number of up nodes",
+		Help: "node status",
 	}, []string{"hostname", "podname", "nodename"})
 
 	once.Do(func() {