// writeSummaries retrieves status summaries from the supplied
// NodeStatusRecorder and persists them to the cockroach data store.
func (n *Node) writeSummaries() error {
var err error
n.stopper.RunTask(func() {
nodeStatus := n.recorder.GetStatusSummary()
if nodeStatus != nil {
key := keys.NodeStatusKey(int32(nodeStatus.Desc.NodeID))
// We use PutInline to store only a single version of the node
// status. There's not much point in keeping the historical
// versions as we keep all of the constituent data as
// timeseries. Further, due to the size of the build info in the
// node status, writing one of these every 10s will generate
// more versions than will easily fit into a range over the
// course of a day.
if err = n.ctx.DB.PutInline(key, nodeStatus); err != nil {
return
}
if log.V(2) {
statusJSON, err := json.Marshal(nodeStatus)
if err != nil {
log.Errorf("error marshaling nodeStatus to json: %s", err)
}
log.Infof("node %d status: %s", nodeStatus.Desc.NodeID, statusJSON)
}
}
})
return err
}
// writeSummaries retrieves status summaries from the supplied
// NodeStatusRecorder and persists them to the cockroach data store.
func (s *Server) writeSummaries() error {
nodeStatus, storeStatuses := s.recorder.GetStatusSummaries()
if nodeStatus != nil {
key := keys.NodeStatusKey(int32(nodeStatus.Desc.NodeID))
if err := s.db.Put(key, nodeStatus); err != nil {
return err
}
if log.V(1) {
statusJSON, err := json.Marshal(nodeStatus)
if err != nil {
log.Errorf("error marshaling nodeStatus to json: %s", err)
}
log.Infof("node %d status: %s", nodeStatus.Desc.NodeID, statusJSON)
}
}
for _, ss := range storeStatuses {
key := keys.StoreStatusKey(int32(ss.Desc.StoreID))
if err := s.db.Put(key, &ss); err != nil {
return err
}
if log.V(1) {
statusJSON, err := json.Marshal(&ss)
if err != nil {
log.Errorf("error marshaling storeStatus to json: %s", err)
}
log.Infof("store %d status: %s", ss.Desc.StoreID, statusJSON)
}
}
return nil
}
// handleNodeStatus handles GET requests for a single node's status. If no id is
// available, it calls handleNodesStatus to return all node's statuses.
func (s *statusServer) handleNodeStatus(w http.ResponseWriter, r *http.Request, ps httprouter.Params) {
id, err := strconv.ParseInt(ps.ByName("id"), 10, 64)
if err != nil {
log.Error(err)
w.WriteHeader(http.StatusInternalServerError)
return
}
key := keys.NodeStatusKey(int32(id))
nodeStatus := &status.NodeStatus{}
if err := s.db.GetProto(key, nodeStatus); err != nil {
log.Error(err)
w.WriteHeader(http.StatusInternalServerError)
return
}
b, contentType, err := util.MarshalResponse(r, nodeStatus, []util.EncodingType{util.JSONEncoding})
if err != nil {
log.Error(err)
w.WriteHeader(http.StatusInternalServerError)
return
}
w.Header().Set(util.ContentTypeHeader, contentType)
w.Write(b)
}
// writeSummaries retrieves status summaries from the supplied
// NodeStatusRecorder and persists them to the cockroach data store.
func (s *Server) writeSummaries() (err error) {
s.stopper.RunTask(func() {
nodeStatus, storeStatuses := s.recorder.GetStatusSummaries()
if nodeStatus != nil {
key := keys.NodeStatusKey(int32(nodeStatus.Desc.NodeID))
if err = s.db.Put(key, nodeStatus); err != nil {
return
}
if log.V(1) {
log.Infof("recorded status for node %d", nodeStatus.Desc.NodeID)
}
}
for _, ss := range storeStatuses {
key := keys.StoreStatusKey(int32(ss.Desc.StoreID))
if err = s.db.Put(key, &ss); err != nil {
return
}
}
if log.V(1) {
log.Infof("recorded status for %d stores", len(storeStatuses))
}
})
return nil
}
// writeSummaries retrieves status summaries from the supplied
// NodeStatusRecorder and persists them to the cockroach data store.
func (s *Server) writeSummaries() error {
if !s.stopper.StartTask() {
return nil
}
defer s.stopper.FinishTask()
nodeStatus, storeStatuses := s.recorder.GetStatusSummaries()
if nodeStatus != nil {
key := keys.NodeStatusKey(int32(nodeStatus.Desc.NodeID))
if err := s.db.Put(key, nodeStatus); err != nil {
return err
}
if log.V(1) {
log.Infof("recorded status for node %d", nodeStatus.Desc.NodeID)
}
}
for _, ss := range storeStatuses {
key := keys.StoreStatusKey(int32(ss.Desc.StoreID))
if err := s.db.Put(key, &ss); err != nil {
return err
}
}
if log.V(1) {
log.Infof("recorded status for %d stores", len(storeStatuses))
}
return nil
}
// writeSummaries retrieves status summaries from the supplied
// NodeStatusRecorder and persists them to the cockroach data store.
func (n *Node) writeSummaries() error {
var err error
n.stopper.RunTask(func() {
nodeStatus := n.recorder.GetStatusSummary()
if nodeStatus != nil {
key := keys.NodeStatusKey(int32(nodeStatus.Desc.NodeID))
if pErr := n.ctx.DB.Put(key, nodeStatus); pErr != nil {
err = pErr.GoError()
return
}
if log.V(1) {
statusJSON, err := json.Marshal(nodeStatus)
if err != nil {
log.Errorf("error marshaling nodeStatus to json: %s", err)
}
log.Infof("node %d status: %s", nodeStatus.Desc.NodeID, statusJSON)
}
}
})
return err
}
// handleNodeStatus handles GET requests for a single node's status.
func (s *statusServer) Node(ctx context.Context, req *serverpb.NodeRequest) (*status.NodeStatus, error) {
nodeID, _, err := s.parseNodeID(req.NodeId)
if err != nil {
return nil, grpc.Errorf(codes.InvalidArgument, err.Error())
}
key := keys.NodeStatusKey(int32(nodeID))
b := &client.Batch{}
b.Get(key)
if err := s.db.Run(b); err != nil {
log.Error(ctx, err)
return nil, grpc.Errorf(codes.Internal, err.Error())
}
var nodeStatus status.NodeStatus
if err := b.Results[0].Rows[0].ValueProto(&nodeStatus); err != nil {
err = errors.Errorf("could not unmarshal NodeStatus from %s: %s", key, err)
log.Error(ctx, err)
return nil, grpc.Errorf(codes.Internal, err.Error())
}
return &nodeStatus, nil
}
// compareNodeStatus ensures that the actual node status for the passed in
// node is updated correctly. It checks that the Node Descriptor, StoreIDs,
// RangeCount, StartedAt, ReplicatedRangeCount and are exactly correct and that
// the bytes and counts for Live, Key and Val are at least the expected value.
// And that UpdatedAt has increased.
// The latest actual stats are returned.
func compareNodeStatus(t *testing.T, ts *TestServer, expectedNodeStatus *status.NodeStatus, testNumber int) *status.NodeStatus {
nodeStatusKey := keys.NodeStatusKey(int32(ts.node.Descriptor.NodeID))
nodeStatus := &status.NodeStatus{}
if err := ts.db.GetProto(nodeStatusKey, nodeStatus); err != nil {
t.Fatalf("%v: failure getting node status: %s", testNumber, err)
}
// These values must be equal.
if a, e := nodeStatus.RangeCount, expectedNodeStatus.RangeCount; a != e {
t.Errorf("%d: RangeCount does not match expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Desc, expectedNodeStatus.Desc; !reflect.DeepEqual(a, e) {
t.Errorf("%d: Descriptor does not match expected.\nexpected: %s\nactual: %s", testNumber, e, a)
}
if a, e := nodeStatus.ReplicatedRangeCount, expectedNodeStatus.ReplicatedRangeCount; a != e {
t.Errorf("%d: ReplicatedRangeCount does not match expected.\nexpected: %d actual: %d", testNumber, e, a)
}
// These values must >= to the older value.
// If StartedAt is 0, we skip this test as we don't have the base value yet.
if a, e := nodeStatus.StartedAt, expectedNodeStatus.StartedAt; e > 0 && e != a {
t.Errorf("%d: StartedAt does not match expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Stats.LiveBytes, expectedNodeStatus.Stats.LiveBytes; a < e {
t.Errorf("%d: LiveBytes is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Stats.KeyBytes, expectedNodeStatus.Stats.KeyBytes; a < e {
t.Errorf("%d: KeyBytes is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Stats.ValBytes, expectedNodeStatus.Stats.ValBytes; a < e {
t.Errorf("%d: ValBytes is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Stats.LiveCount, expectedNodeStatus.Stats.LiveCount; a < e {
t.Errorf("%d: LiveCount is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Stats.KeyCount, expectedNodeStatus.Stats.KeyCount; a < e {
t.Errorf("%d: KeyCount is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.Stats.ValCount, expectedNodeStatus.Stats.ValCount; a < e {
t.Errorf("%d: ValCount is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
if a, e := nodeStatus.UpdatedAt, expectedNodeStatus.UpdatedAt; a < e {
t.Errorf("%d: UpdatedAt is not greater or equal to expected.\nexpected: %d actual: %d", testNumber, e, a)
}
// Compare the store ids.
var actualStoreIDs, expectedStoreIDs sort.IntSlice
for _, id := range nodeStatus.StoreIDs {
actualStoreIDs = append(actualStoreIDs, int(id))
}
sort.Sort(actualStoreIDs)
for _, id := range expectedNodeStatus.StoreIDs {
expectedStoreIDs = append(expectedStoreIDs, int(id))
}
sort.Sort(expectedStoreIDs)
if !reflect.DeepEqual(actualStoreIDs, expectedStoreIDs) {
t.Errorf("%d: actual Store IDs don't match expected.\nexpected: %d actual: %d", testNumber, expectedStoreIDs, actualStoreIDs)
}
return nodeStatus
}
// compareNodeStatus ensures that the actual node status for the passed in
// node is updated correctly. It checks that the Node Descriptor, StoreIDs,
// RangeCount, StartedAt, ReplicatedRangeCount and are exactly correct and that
// the bytes and counts for Live, Key and Val are at least the expected value.
// And that UpdatedAt has increased.
// The latest actual stats are returned.
func compareStoreStatus(t *testing.T, node *Node, expectedNodeStatus *status.NodeStatus, testNumber int) *status.NodeStatus {
nodeStatusKey := keys.NodeStatusKey(int32(node.Descriptor.NodeID))
request := &proto.GetRequest{
RequestHeader: proto.RequestHeader{
Key: nodeStatusKey,
},
}
ns := (*nodeServer)(node)
response := &proto.GetResponse{}
if err := ns.Get(request, response); err != nil {
t.Fatalf("%v: failure getting node status: %s", testNumber, err)
}
if response.Value == nil {
t.Errorf("%v: could not find node status at: %s", testNumber, nodeStatusKey)
}
nodeStatus := &status.NodeStatus{}
if err := gogoproto.Unmarshal(response.Value.GetBytes(), nodeStatus); err != nil {
t.Fatalf("%v: could not unmarshal store status: %+v", testNumber, response)
}
// There values must be equal.
if expectedNodeStatus.RangeCount != nodeStatus.RangeCount {
t.Errorf("%v: RangeCount does not match expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if !reflect.DeepEqual(expectedNodeStatus.Desc, nodeStatus.Desc) {
t.Errorf("%v: Description does not match expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if expectedNodeStatus.ReplicatedRangeCount != nodeStatus.ReplicatedRangeCount {
t.Errorf("%v: ReplicatedRangeCount does not match expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
// There values must >= to the older value.
// If StartedAt is 0, we skip this test as we don't have the base value yet.
if expectedNodeStatus.StartedAt > 0 && expectedNodeStatus.StartedAt != nodeStatus.StartedAt {
t.Errorf("%v: StartedAt does not match expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.Stats.LiveBytes < expectedNodeStatus.Stats.LiveBytes {
t.Errorf("%v: LiveBytes is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.Stats.KeyBytes < expectedNodeStatus.Stats.KeyBytes {
t.Errorf("%v: KeyBytes is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.Stats.ValBytes < expectedNodeStatus.Stats.ValBytes {
t.Errorf("%v: ValBytes is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.Stats.LiveCount < expectedNodeStatus.Stats.LiveCount {
t.Errorf("%v: LiveCount is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.Stats.KeyCount < expectedNodeStatus.Stats.KeyCount {
t.Errorf("%v: KeyCount is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.Stats.ValCount < expectedNodeStatus.Stats.ValCount {
t.Errorf("%v: ValCount is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
if nodeStatus.UpdatedAt < expectedNodeStatus.UpdatedAt {
t.Errorf("%v: UpdatedAt is not greater or equal to expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
}
// Compare the store ids.
storeIDs := make(map[proto.StoreID]int)
for _, id := range expectedNodeStatus.StoreIDs {
storeIDs[id]++
}
for _, id := range nodeStatus.StoreIDs {
storeIDs[id]--
}
for _, count := range storeIDs {
if count != 0 {
t.Errorf("%v: actual Store IDs don't match expected\nexpected: %+v\nactual: %v\n", testNumber, expectedNodeStatus, nodeStatus)
break
}
}
return nodeStatus
}
// compareNodeStatus ensures that the actual node status for the passed in
// node is updated correctly. It checks that the Node Descriptor, StoreIDs,
// RangeCount, StartedAt, ReplicatedRangeCount and are exactly correct and that
// the bytes and counts for Live, Key and Val are at least the expected value.
// And that UpdatedAt has increased.
// The latest actual stats are returned.
func compareNodeStatus(t *testing.T, ts *TestServer, expectedNodeStatus *status.NodeStatus, testNumber int) *status.NodeStatus {
// ========================================
// Read NodeStatus from server and validate top-level fields.
// ========================================
nodeStatusKey := keys.NodeStatusKey(int32(ts.node.Descriptor.NodeID))
nodeStatus := &status.NodeStatus{}
if err := ts.db.GetProto(nodeStatusKey, nodeStatus); err != nil {
t.Fatalf("%d: failure getting node status: %s", testNumber, err)
}
// Descriptor values should be exactly equal to expected.
if a, e := nodeStatus.Desc, expectedNodeStatus.Desc; !reflect.DeepEqual(a, e) {
t.Errorf("%d: Descriptor does not match expected.\nexpected: %s\nactual: %s", testNumber, e, a)
}
// ========================================
// Ensure all expected stores are represented in the node status.
// ========================================
storesToMap := func(ns *status.NodeStatus) map[roachpb.StoreID]status.StoreStatus {
strMap := make(map[roachpb.StoreID]status.StoreStatus, len(ns.StoreStatuses))
for _, str := range ns.StoreStatuses {
strMap[str.Desc.StoreID] = str
}
return strMap
}
actualStores := storesToMap(nodeStatus)
expectedStores := storesToMap(expectedNodeStatus)
if a, e := len(actualStores), len(expectedStores); a != e {
t.Errorf("%d: actual status contained %d stores, expected %d", testNumber, a, e)
}
for key := range expectedStores {
if _, ok := actualStores[key]; !ok {
t.Errorf("%d: actual node status did not contain expected store %d", testNumber, key)
}
}
if t.Failed() {
t.FailNow()
}
// ========================================
// Ensure all metric sets (node and store level) are consistent with
// expected status.
// ========================================
// CompareMetricMaps accepts an actual and expected metric maps, along with
// two lists of string keys. For metrics with keys in the 'equal' map, the
// actual value must be equal to the expected value. For keys in the
// 'greater' map, the actul value must be greater than or equal to the
// expected value.
compareMetricMaps := func(actual, expected map[string]float64, equal, greater []string) {
// Make sure the actual value map contains all values in expected map.
for key := range expected {
if _, ok := actual[key]; !ok {
t.Errorf("%d: actual node status did not contain expected metric %s", testNumber, key)
}
}
if t.Failed() {
return
}
// For each equal key, ensure that the actual value is equal to expected
// key.
for _, key := range equal {
if _, ok := actual[key]; !ok {
t.Errorf("%d, actual node status did not contain expected 'equal' metric key %s", testNumber, key)
continue
}
if a, e := actual[key], expected[key]; a != e {
t.Errorf("%d: %s does not match expected value.\nExpected %f, Actual %f", testNumber, key, e, a)
}
}
for _, key := range greater {
if _, ok := actual[key]; !ok {
t.Errorf("%d: actual node status did not contain expected 'greater' metric key %s", testNumber, key)
continue
}
if a, e := actual[key], expected[key]; a < e {
t.Errorf("%d: %s is not greater than or equal to expected value.\nExpected %f, Actual %f", testNumber, key, e, a)
}
}
}
compareMetricMaps(nodeStatus.Metrics, expectedNodeStatus.Metrics, nil, []string{
"exec.success-count",
"exec.error-count",
})
for key := range actualStores {
// Directly verify a subset of metrics which have predictable output.
compareMetricMaps(actualStores[key].Metrics, expectedStores[key].Metrics,
[]string{
"replicas",
"ranges.replicated",
},
[]string{
"livebytes",
"keybytes",
"valbytes",
"livecount",
"keycount",
"valcount",
})
}
if t.Failed() {
t.FailNow()
}
return nodeStatus
}
// startStoresScanner will walk through all the stores in the node every
// ctx.ScanInterval and store the status in the db.
func (n *Node) startStoresScanner(stopper *util.Stopper) {
stopper.RunWorker(func() {
// Pick the smaller of the two intervals.
var minScanInterval time.Duration
if n.ctx.ScanInterval <= n.ctx.ScanMaxIdleTime || n.ctx.ScanMaxIdleTime == 0 {
minScanInterval = n.ctx.ScanInterval
} else {
minScanInterval = n.ctx.ScanMaxIdleTime
}
// TODO(bram): The number of stores is small. The node status should be
// updated whenever a store status is updated.
for interval := time.Duration(0); true; interval = minScanInterval {
select {
case <-time.After(interval):
if !stopper.StartTask() {
continue
}
// Walk through all the stores on this node.
var rangeCount, leaderRangeCount, replicatedRangeCount, availableRangeCount int32
stats := &engine.MVCCStats{}
accessedStoreIDs := []proto.StoreID{}
// will never error because `return nil` below
_ = n.lSender.VisitStores(func(store *storage.Store) error {
storeStatus, err := store.GetStatus()
if err != nil {
log.Error(err)
return nil
}
if storeStatus == nil {
// The store scanner hasn't run on this node yet.
return nil
}
accessedStoreIDs = append(accessedStoreIDs, store.Ident.StoreID)
rangeCount += storeStatus.RangeCount
leaderRangeCount += storeStatus.LeaderRangeCount
replicatedRangeCount += storeStatus.ReplicatedRangeCount
availableRangeCount += storeStatus.AvailableRangeCount
stats.Add(&storeStatus.Stats)
return nil
})
// Store the combined stats in the db.
now := n.ctx.Clock.Now().WallTime
status := &NodeStatus{
Desc: n.Descriptor,
StoreIDs: accessedStoreIDs,
UpdatedAt: now,
StartedAt: n.startedAt,
RangeCount: rangeCount,
Stats: *stats,
LeaderRangeCount: leaderRangeCount,
ReplicatedRangeCount: replicatedRangeCount,
AvailableRangeCount: availableRangeCount,
}
key := keys.NodeStatusKey(int32(n.Descriptor.NodeID))
if err := n.ctx.DB.Put(key, status); err != nil {
log.Error(err)
}
// Increment iteration count.
n.completedScan.L.Lock()
n.scanCount++
n.completedScan.Broadcast()
n.completedScan.L.Unlock()
if log.V(6) {
log.Infof("store scan iteration completed")
}
stopper.FinishTask()
case <-stopper.ShouldStop():
// Exit the loop.
return
}
}
})
}
