GRPC java-Eureka client side Load balancing and Channel creation

[iis...@gmail.com]

Hi All,

I am trying to use Eureka as a discovery server and do roundrobin clientside load balancing in the grpc client level.I have 3 GRPC server nodes registered in Eureka.

Here is the way I create a ManagedChannel in  the client.

EurekaClientConfig eurekaClientConfig = new DefaultEurekaClientConfig();

ManagedChannel channel =  ManagedChannelBuilder

               .forTarget("eureka://" + "service-aeroline-passenger-messaging")                

                .nameResolverFactory(new EurekaNameResolverProvider(eurekaClientConfig, "9071"))

                .loadBalancerFactory(RoundRobinLoadBalancerFactory.getInstance())

                .usePlaintext(true)

                .build();

My question is is it ok to create one channel and share across the client application presuming that channel rotation across various nodes of GRPC server is taken care by

the ManagedChannel it self.  i.e if server1 goes down does the channel automatically diverted to server2 without creating a new channel object???

Thanks,

Isuru

[Carl Mastrangelo]

It is safe to share the channel.   The decision of which server is actually up to the load balancer, and it will correctly stop sending traffic to a server if the connection fails.  

[Isuru Samaraweera]

Hi Carl,

Thanks for the  reply.However when I do the eureka address lookup ,lookup seems fine.But at the time stub method is invoked asynchronously below error pops up.

Caused by: java.nio.channels.UnresolvedAddressException

at sun.nio.ch.Net.checkAddress(Net.java:101)

at sun.nio.ch.SocketChannelImpl.connect(SocketChannelImpl.java:622)

at io.netty.util.internal.SocketUtils$3.run(SocketUtils.java:83)

at io.netty.util.internal.SocketUtils$3.run(SocketUtils.java:80)

at java.security.AccessController.doPrivileged(Native Method)

at io.netty.util.internal.SocketUtils.connect(SocketUtils.java:80)

at io.netty.channel.socket.nio.NioSocketChannel.doConnect(NioSocketChannel.java:310)

at io.netty.channel.nio.AbstractNioChannel$AbstractNioUnsafe.connect(AbstractNioChannel.java:254)

at io.netty.channel.DefaultChannelPipeline$HeadContext.connect(DefaultChannelPipeline.java:1366)

at io.netty.channel.AbstractChannelHandlerContext.invokeConnect(AbstractChannelHandlerContext.java:545)

at io.netty.channel.AbstractChannelHandlerContext.connect(AbstractChannelHandlerContext.java:530)

at io.netty.handler.codec.http2.Http2ConnectionHandler.connect(Http2ConnectionHandler.java:461)

at io.netty.channel.AbstractChannelHandlerContext.invokeConnect(AbstractChannelHandlerContext.java:545)

at io.netty.channel.AbstractChannelHandlerContext.connect(AbstractChannelHandlerContext.java:530)

at io.netty.channel.ChannelDuplexHandler.connect(ChannelDuplexHandler.java:50)

at io.grpc.netty.ProtocolNegotiators$AbstractBufferingHandler.connect(ProtocolNegotiators.java:466)

at io.netty.channel.AbstractChannelHandlerContext.invokeConnect(AbstractChannelHandlerContext.java:545)

at io.netty.channel.AbstractChannelHandlerContext.access$1000(AbstractChannelHandlerContext.java:38)

at io.netty.channel.AbstractChannelHandlerContext$11.run(AbstractChannelHandlerContext.java:535)

at io.netty.util.concurrent.AbstractEventExecutor.safeExecute(AbstractEventExecutor.java:163)

at io.netty.util.concurrent.SingleThreadEventExecutor.runAllTasks(SingleThreadEventExecutor.java:404)

at io.netty.channel.nio.NioEventLoop.run(NioEventLoop.java:465)

at io.netty.util.concurrent.SingleThreadEventExecutor$5.run(SingleThreadEventExecutor.java:884)

at io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)

I do the lookup as below .

Application application = eurekaClient.getApplication(serviceName);

List<EquivalentAddressGroup> servers = new ArrayList<>();

for (InstanceInfo serviceInstance : application.getInstances()) {

servers.add(new EquivalentAddressGroup(

InetSocketAddress.createUnresolved(serviceInstance.getHostName(), serviceInstance.getPort())));

}

listener.onAddresses(servers, Attributes.EMPTY);

Lookup looks up addressed properly.

Do you have any clue on why above exception is thrown when stub method is invoked??

 

--
You received this message because you are subscribed to a topic in the Google Groups "grpc.io" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/grpc-io/7g4PAv_7Clo/unsubscribe.
To unsubscribe from this group and all its topics, send an email to grpc-io+unsubscribe@googlegroups.com.
To post to this group, send email to grp...@googlegroups.com.
Visit this group at https://groups.google.com/group/grpc-io.
To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/4460b577-ebd2-40d5-aa1f-182a1f4af411%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Isuru Samaraweera

[Carl Mastrangelo]

I believe you need to create the InetSocketAddress from an IP address, rather than a host name.  Typically host names are looked up via DNS (not sure what Eureka returns to you).   If you use the normal InetSocketAddress constructor, Java will do the DNS lookup for you.

(It is also possible to not use Java's DNS resolver, but that's more complex.  I'd avoid it until you know you need it)

To unsubscribe from this group and all its topics, send an email to grpc-io+u...@googlegroups.com.

To post to this group, send email to grp...@googlegroups.com.
Visit this group at https://groups.google.com/group/grpc-io.

To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/4460b577-ebd2-40d5-aa1f-182a1f4af411%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Isuru Samaraweera

[Isuru Samaraweera]

Hi Carl,

Thanks for the reply.Due to timeline constraints I switch back to round robin default dns solver as below.It is serving the purpose for the current work load.In future surely I ll do further R n D on Eureka considering your inputs as well.Here is the code I fallabacked to.

 ManagedChannelBuilder.forTarget(grpcHostTmp.getHost() + ":" + grpcHostTmp.getPort())

.nameResolverFactory(new DnsNameResolverProvider()).executor(messagingThreadPoolExecutor)

.loadBalancerFactory(RoundRobinLoadBalancerFactory.getInstance()).usePlaintext(true).build());

Is there anyway to load balance requests evenly across nodes with DnsNameResolverProvider .Rightnow whats happening is till a node is going down same node takes the load after that node goes down next node takes the load.

Is there a way to rotate requests as in active active mode rather than above active passive mode.

I am looking for a grpc built in solution without 3rd party eureka ,zoo keeper.Or Do you have plans to incorporate better load balancing apis to the grpc core libs??

Thanks,

Isuru

To unsubscribe from this group and all its topics, send an email to grpc-io+unsubscribe@googlegroups.com.

To post to this group, send email to grp...@googlegroups.com.
Visit this group at https://groups.google.com/group/grpc-io.

To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/4460b577-ebd2-40d5-aa1f-182a1f4af411%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Isuru Samaraweera

--

Isuru Samaraweera

[Carl Mastrangelo]

Responses inline

On Friday, August 10, 2018 at 7:40:25 PM UTC-7, Isuru Samaraweera wrote:

Hi Carl,

Thanks for the reply.Due to timeline constraints I switch back to round robin default dns solver as below.It is serving the purpose for the current work load.In future surely I ll do further R n D on Eureka considering your inputs as well.Here is the code I fallabacked to.

 ManagedChannelBuilder.forTarget(grpcHostTmp.getHost() + ":" + grpcHostTmp.getPort())

.nameResolverFactory(new DnsNameResolverProvider()).executor(messagingThreadPoolExecutor)

I would avoid using DnsNameResolverProvider.  It's in our io.grpc.internal package, which means we can make breaking changes to that API.   Also, DnsNameResolver is the default name resolver, so there is no need to specify it.

 

.loadBalancerFactory(RoundRobinLoadBalancerFactory.getInstance()).usePlaintext(true).build());

Is there anyway to load balance requests evenly across nodes with DnsNameResolverProvider .Rightnow whats happening is till a node is going down same node takes the load after that node goes down next node takes the load.  You can see the implementation in RoundRobinLoadBalancerFactory.Picker#nextSubchannel().

Are you using streaming RPCs, or unary?   Streaming RPCs can't change their backend once started, but every new RPC can.  RoundRobinLoadBalancer should pick a new backend for every single RPC.

 

Is there a way to rotate requests as in active active mode rather than above active passive mode.

I think you may be using PickFirstLoadBalancer which has that behavior.

 

I am looking for a grpc built in solution without 3rd party eureka ,zoo keeper.Or Do you have plans to incorporate better load balancing apis to the grpc core libs??

We keep most things out of the core library because not everyone needs them.   We do promote other implementations of gRPC components here: https://github.com/grpc-ecosystem  

 

Thanks,

Isuru

To unsubscribe from this group and all its topics, send an email to grpc-io+u...@googlegroups.com.

To post to this group, send email to grp...@googlegroups.com.
Visit this group at https://groups.google.com/group/grpc-io.

To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/4460b577-ebd2-40d5-aa1f-182a1f4af411%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Isuru Samaraweera

--

Isuru Samaraweera

[Isuru Samaraweera]

Hi Carl,

I got the above example from a sample code from google author it self where they use DNS factory.

https://github.com/saturnism/grpc-java-by-example/blob/master/kubernetes-lb-example/echo-client-lb-dns/src/main/java/com/example/grpc/client/ClientSideLoadBalancedEchoClient.java

Do you have a sample code for picker#subchannel because we are running out of time for delivery.Surely I 'll do Rn D in this picker/subchannel once our imminent delivery is completed.

Thanks

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+unsubscribe@googlegroups.com.

To post to this group, send email to grp...@googlegroups.com.
Visit this group at https://groups.google.com/group/grpc-io.

To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/65c00c3a-b675-4944-9583-8225f3d0cbe8%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Isuru Samaraweera

[Carl Mastrangelo]

Hmm, that should not be used, I might have to send a PR to the author.  

I don't know why you always get the same backend.  I'm afraid I would have to see a reproducing case before I could say with any authority why it's happening.

To unsubscribe from this group and stop receiving emails from it, send an email to grpc-io+u...@googlegroups.com.

To post to this group, send email to grp...@googlegroups.com.
Visit this group at https://groups.google.com/group/grpc-io.

To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/65c00c3a-b675-4944-9583-8225f3d0cbe8%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

Isuru Samaraweera

[Isuru Samaraweera]

Hi carl,Thanks for the invaluable replies.

Regards,

Isuru

To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/d35e5be6-e6f5-4251-958a-929e98a51751%40googlegroups.com.

[pooj...@gmail.com]

Hi 

Am here to ask how to registed  GRPC server to eureka server.

Thanks in advance.

[sanjay...@google.com]

This pointer might help https://github.com/LogNet/grpc-spring-boot-starter/blob/master/README.adoc#6-eureka-integration