Exemplo n.º 1
0
// Start starts the test cluster by bootstrapping an in-memory store
// (defaults to maximum of 50M). The server is started, launching the
// node RPC server and all HTTP endpoints. Use the value of
// TestServer.Addr after Start() for client connections. Use Stop()
// to shutdown the server after the test completes.
func (ltc *LocalTestCluster) Start(t util.Tester, baseCtx *base.Config, initSender InitSenderFn) {
	ambient := log.AmbientContext{Tracer: tracing.NewTracer()}
	nc := &base.NodeIDContainer{}
	ambient.AddLogTag("n", nc)

	nodeID := roachpb.NodeID(1)
	nodeDesc := &roachpb.NodeDescriptor{NodeID: nodeID}

	ltc.tester = t
	ltc.Manual = hlc.NewManualClock(0)
	ltc.Clock = hlc.NewClock(ltc.Manual.UnixNano)
	ltc.Stopper = stop.NewStopper()
	rpcContext := rpc.NewContext(ambient, baseCtx, ltc.Clock, ltc.Stopper)
	server := rpc.NewServer(rpcContext) // never started
	ltc.Gossip = gossip.New(ambient, nc, rpcContext, server, nil, ltc.Stopper, metric.NewRegistry())
	ltc.Eng = engine.NewInMem(roachpb.Attributes{}, 50<<20)
	ltc.Stopper.AddCloser(ltc.Eng)

	ltc.Stores = storage.NewStores(ambient, ltc.Clock)

	ltc.Sender = initSender(nodeDesc, ambient.Tracer, ltc.Clock, ltc.Latency, ltc.Stores, ltc.Stopper,
		ltc.Gossip)
	if ltc.DBContext == nil {
		dbCtx := client.DefaultDBContext()
		ltc.DBContext = &dbCtx
	}
	ltc.DB = client.NewDBWithContext(ltc.Sender, *ltc.DBContext)
	transport := storage.NewDummyRaftTransport()
	cfg := storage.TestStoreConfig()
	if ltc.RangeRetryOptions != nil {
		cfg.RangeRetryOptions = *ltc.RangeRetryOptions
	}
	cfg.AmbientCtx = ambient
	cfg.Clock = ltc.Clock
	cfg.DB = ltc.DB
	cfg.Gossip = ltc.Gossip
	cfg.Transport = transport
	cfg.MetricsSampleInterval = metric.TestSampleInterval
	ltc.Store = storage.NewStore(cfg, ltc.Eng, nodeDesc)
	if err := ltc.Store.Bootstrap(roachpb.StoreIdent{NodeID: nodeID, StoreID: 1}); err != nil {
		t.Fatalf("unable to start local test cluster: %s", err)
	}
	ltc.Stores.AddStore(ltc.Store)
	if err := ltc.Store.BootstrapRange(nil); err != nil {
		t.Fatalf("unable to start local test cluster: %s", err)
	}
	if err := ltc.Store.Start(context.Background(), ltc.Stopper); err != nil {
		t.Fatalf("unable to start local test cluster: %s", err)
	}
	nc.Set(context.TODO(), nodeDesc.NodeID)
	if err := ltc.Gossip.SetNodeDescriptor(nodeDesc); err != nil {
		t.Fatalf("unable to set node descriptor: %s", err)
	}
}
Exemplo n.º 2
0
// initStores initializes the Stores map from ID to Store. Stores are
// added to the local sender if already bootstrapped. A bootstrapped
// Store has a valid ident with cluster, node and Store IDs set. If
// the Store doesn't yet have a valid ident, it's added to the
// bootstraps list for initialization once the cluster and node IDs
// have been determined.
func (n *Node) initStores(
	ctx context.Context, engines []engine.Engine, stopper *stop.Stopper, bootstrapped bool,
) error {
	var bootstraps []*storage.Store

	if len(engines) == 0 {
		return errors.Errorf("no engines")
	}
	for _, e := range engines {
		s := storage.NewStore(n.storeCfg, e, &n.Descriptor)
		log.Eventf(ctx, "created store for engine: %s", e)
		if bootstrapped {
			s.NotifyBootstrapped()
		}
		// Initialize each store in turn, handling un-bootstrapped errors by
		// adding the store to the bootstraps list.
		if err := s.Start(ctx, stopper); err != nil {
			if _, ok := err.(*storage.NotBootstrappedError); ok {
				log.Infof(ctx, "store %s not bootstrapped", s)
				bootstraps = append(bootstraps, s)
				continue
			}
			return errors.Errorf("failed to start store: %s", err)
		}
		if s.Ident.ClusterID == *uuid.EmptyUUID || s.Ident.NodeID == 0 {
			return errors.Errorf("unidentified store: %s", s)
		}
		capacity, err := s.Capacity()
		if err != nil {
			return errors.Errorf("could not query store capacity: %s", err)
		}
		log.Infof(ctx, "initialized store %s: %+v", s, capacity)
		n.addStore(s)
	}

	// If there are no initialized stores and no gossip resolvers,
	// bootstrap this node as the seed of a new cluster.
	if n.stores.GetStoreCount() == 0 {
		resolvers := n.storeCfg.Gossip.GetResolvers()
		// Check for the case of uninitialized node having only itself specified as join host.
		switch len(resolvers) {
		case 0:
			return errNeedsBootstrap
		case 1:
			if resolvers[0].Addr() == n.Descriptor.Address.String() {
				return errCannotJoinSelf
			}
		}
	}

	// Verify all initialized stores agree on cluster and node IDs.
	if err := n.validateStores(); err != nil {
		return err
	}
	log.Event(ctx, "validated stores")

	// Set the stores map as the gossip persistent storage, so that
	// gossip can bootstrap using the most recently persisted set of
	// node addresses.
	if err := n.storeCfg.Gossip.SetStorage(n.stores); err != nil {
		return fmt.Errorf("failed to initialize the gossip interface: %s", err)
	}

	// Connect gossip before starting bootstrap. For new nodes, connecting
	// to the gossip network is necessary to get the cluster ID.
	n.connectGossip(ctx)
	log.Event(ctx, "connected to gossip")

	// If no NodeID has been assigned yet, allocate a new node ID by
	// supplying 0 to initNodeID.
	if n.Descriptor.NodeID == 0 {
		n.initNodeID(0)
		n.initialBoot = true
		log.Eventf(ctx, "allocated node ID %d", n.Descriptor.NodeID)
	}

	// Bootstrap any uninitialized stores asynchronously.
	if len(bootstraps) > 0 {
		if err := stopper.RunAsyncTask(ctx, func(ctx context.Context) {
			n.bootstrapStores(ctx, bootstraps, stopper)
		}); err != nil {
			return err
		}
	}

	return nil
}
Exemplo n.º 3
0
// bootstrapCluster bootstraps a multiple stores using the provided
// engines and cluster ID. The first bootstrapped store contains a
// single range spanning all keys. Initial range lookup metadata is
// populated for the range. Returns the cluster ID.
func bootstrapCluster(engines []engine.Engine, txnMetrics kv.TxnMetrics) (uuid.UUID, error) {
	clusterID := uuid.MakeV4()
	stopper := stop.NewStopper()
	defer stopper.Stop()

	cfg := storage.StoreConfig{}
	cfg.ScanInterval = 10 * time.Minute
	cfg.MetricsSampleInterval = time.Duration(math.MaxInt64)
	cfg.ConsistencyCheckInterval = 10 * time.Minute
	cfg.Clock = hlc.NewClock(hlc.UnixNano)
	cfg.AmbientCtx.Tracer = tracing.NewTracer()
	// Create a KV DB with a local sender.
	stores := storage.NewStores(cfg.AmbientCtx, cfg.Clock)
	sender := kv.NewTxnCoordSender(cfg.AmbientCtx, stores, cfg.Clock, false, stopper, txnMetrics)
	cfg.DB = client.NewDB(sender)
	cfg.Transport = storage.NewDummyRaftTransport()
	for i, eng := range engines {
		sIdent := roachpb.StoreIdent{
			ClusterID: clusterID,
			NodeID:    FirstNodeID,
			StoreID:   roachpb.StoreID(i + 1),
		}

		// The bootstrapping store will not connect to other nodes so its
		// StoreConfig doesn't really matter.
		s := storage.NewStore(cfg, eng, &roachpb.NodeDescriptor{NodeID: FirstNodeID})

		// Verify the store isn't already part of a cluster.
		if s.Ident.ClusterID != *uuid.EmptyUUID {
			return uuid.UUID{}, errors.Errorf("storage engine already belongs to a cluster (%s)", s.Ident.ClusterID)
		}

		// Bootstrap store to persist the store ident.
		if err := s.Bootstrap(sIdent); err != nil {
			return uuid.UUID{}, err
		}
		// Create first range, writing directly to engine. Note this does
		// not create the range, just its data. Only do this if this is the
		// first store.
		if i == 0 {
			initialValues := GetBootstrapSchema().GetInitialValues()
			if err := s.BootstrapRange(initialValues); err != nil {
				return uuid.UUID{}, err
			}
		}
		if err := s.Start(context.Background(), stopper); err != nil {
			return uuid.UUID{}, err
		}

		stores.AddStore(s)

		ctx := context.TODO()
		// Initialize node and store ids.  Only initialize the node once.
		if i == 0 {
			if nodeID, err := allocateNodeID(ctx, cfg.DB); nodeID != sIdent.NodeID || err != nil {
				return uuid.UUID{}, errors.Errorf("expected to initialize node id allocator to %d, got %d: %s",
					sIdent.NodeID, nodeID, err)
			}
		}
		if storeID, err := allocateStoreIDs(ctx, sIdent.NodeID, 1, cfg.DB); storeID != sIdent.StoreID || err != nil {
			return uuid.UUID{}, errors.Errorf("expected to initialize store id allocator to %d, got %d: %s",
				sIdent.StoreID, storeID, err)
		}
	}
	return clusterID, nil
}