How can I make sure the high availability when I use the go client for dgraph?

(Haitao) #1

Hi,
when I use the go client to query and mutate , I have a question about the high availability, the code for example as below:

func NewGrapher(cfg *GrapherConfig) (*BTCGrapher, error) {
	p := new(BTCGrapher)
	p.cfg = cfg
	conn, err := grpc.Dial(p.cfg.Address+":"+p.cfg.Port, grpc.WithDefaultCallOptions(grpc.MaxCallRecvMsgSize(4<<30),
		grpc.MaxCallSendMsgSize(4<<30)), grpc.WithInsecure())
	if err != nil {
		log.Error("While trying to dial gRPC", err)
		log.DetailError(err)
		return nil, err
	}
	p.grapherClientConn = conn
    ctx := context.Background()
    mu := &api.Mutation{
			CommitNow:  true,
			DeleteJson: pb,
	}

	_, err = dgo.NewDgraphClient(api.NewDgraphClient(p.grapherClientConn)).NewTxn().Mutate(ctx, mu)
		if err != nil {
			log.DetailError(err)
		}
	return p, nil
}

I have three alpha nodes and I have a domain like “test.dgraph.xx.com” point to the three alpha port 9080, I do not clear how can I write codes to relize the high availability like the method GET or POST in HTTP.

Thank You Very Much!

(Haitao) #2

like that ?

func NewGrapher(cfg *GrapherConfig) (*BTCGrapher, error) {
	conn1, err := grpc.Dial(p.cfg.Address1+":"+p.cfg.Port1, grpc.WithDefaultCallOptions(grpc.MaxCallRecvMsgSize(4<<30),
		grpc.MaxCallSendMsgSize(4<<30)), grpc.WithInsecure())
	conn2, err := grpc.Dial(p.cfg.Address2+":"+p.cfg.Port2, grpc.WithDefaultCallOptions(grpc.MaxCallRecvMsgSize(4<<30),
		grpc.MaxCallSendMsgSize(4<<30)), grpc.WithInsecure())
	conn3, err := grpc.Dial(p.cfg.Address3+":"+p.cfg.Port3, grpc.WithDefaultCallOptions(grpc.MaxCallRecvMsgSize(4<<30),
		grpc.MaxCallSendMsgSize(4<<30)), grpc.WithInsecure())
	ctx := context.Background()

	mu := &api.Mutation{
		CommitNow:  true,
		DeleteJson: pb,
	}

	_, err = dgo.NewDgraphClient(api.NewDgraphClient(conn1), api.NewDgraphClient(conn2), api.NewDgraphClient(conn3)).NewTxn().Mutate(ctx, mu)
	if err != nil {
		log.DetailError(err)
	}
	return p, nil
}

but, how can I know which alpha node is alive? If I kill one alpha node , how can I know it?

(Javier Alvarado) #3

The recommended way to ensure high availability is to run Dgraph behind a load balancer which would be responsible for keeping track of which nodes are alive.

The sample code

_, err = dgo.NewDgraphClient(api.NewDgraphClient(conn1), api.NewDgraphClient(conn2), api.NewDgraphClient(conn3)).NewTxn().Mutate(ctx, mu)
	if err != nil {
		log.DetailError(err)
	}

will not work as you intend. When a DgraphClient is initialized with multiple connections, each operation uses only one of them at random. You have to check the error and retry if it the node is down.

(Haitao) #4

Thanks for your reply, I understand your mean, I know I need a load balancer, but I do not know how to client an domain in grpc,can you show me an example code? like that?

conn, err := grpc.Dial("dns:///"+ p.cfg.Domain+":"+p.cfg.Port, grpc.WithDefaultCallOptions(grpc.MaxCallRecvMsgSize(4<<30),
		grpc.MaxCallSendMsgSize(4<<30)), grpc.WithInsecure())

Thank you!

(Daniel Mai) #5

The grpc.Dial address can be a DNS name.

(Haitao) #6

Hi dmai, thanks for your reply, do you have time to look at my another question?

(Haitao) #7

Thanks very much! I have solved the problem, the load balancer I did before is L7 load balancer, now I relize the load balancer in L4. Now it can work well with domain test.dgraph.xx.com in grpc.