Hi,
I have implemented Golang grpc on a blockchain for message passing. My
client side and server side code looks as follows:
1. Client code
/////////////////////code that establishes the grpc connection - grpc
dial//////////////////////////////
func (net DbftNetwork) EstablishConnection() {
if connected == false {
for _, p := range net.Peers {
client, err := rpc.Dial("tcp", p.Address)
fmt.Println("Making a dial")
log.Printf("Dial Error : %v\n", err)
return
}
ClientsList = append(ClientsList, client)
}
connected = true
}
}
///////////////////////code that makes the grpc call to the
server/////////////////////////////////
for _, p := range net.Peers {
go p.Call(method, msg, new(interface{}), ClientsList[count])
count++
}
func (rpcClient *DbftRpcClient) Call(method string, args
messages.ConsensusMsg, reply *interface{}, Client *rpc.Client) {
Client.Go(method, args, reply, nil)
}
2. Server code
func (server *DbftRpcServer) Start() {
handler := rpc.NewServer()
handler.Register(server)
l, err := net.Listen("tcp", fmt.Sprintf("0.0.0.0:%d", server.port))
if err != nil {
log.Fatal("HTTP listen error: ", err)
}
go func() {
for {
cxn, err := l.Accept()
if err != nil {
log.Printf("Error Accept Request: %s\n", err)
return
}
go handler.ServeConn(cxn)
}
}()
}
Problem - I have made the client code non-blocking by adding a go routine
for each client request. I had to do this because, in AWS multi-region the
latency was high and the number of grpc connections blocking made the CPU
stall and performance bad. After adding go routines and increasing the GOGC
value, I was able to get it to work up to 50 peers across 10 AWS regions.
But when I increase the number of peers further, the non-blocking go
routines increases further and starts stalling the CPU again. I was
wondering if there is any solution considering the above code, to make my
gRPC scalable. Any optimizations that anyone can think of.
Machines used - AWS c5.4xlarge, 16vCPUs, 16GB RAM, Ubuntu 18, go version -
1.13.1(Cannot upgrade the version because the code base depends on it)
Thank you for your time.
Highly appreciated.
Deepal
--
You received this message because you are subscribed to the Google Groups
"grpc.io" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to grpc-io+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/grpc-io/6ddd74e9-054b-4a14-b4c2-ea355437c253n%40googlegroups.com.
