how to use health check?

[xiaoyulei]

I see the health check document, but I am confused. Is there any example code to show how to enable health check?

I call EnableDefaultHealthCheckService(true) before server start, is it open health check? Do I need something else?

[Falco20019]

See health_service_end2end_test.cc for an example. If you have enabled it, you should be fine. health_service_end2end_test.cc / VerifyHealthCheckService shows an example verifier code you could try.

[xiaoyulei]

@Falco20019 thanks, I saw this file yesterday, but I am still confused. After EnableDefaultHealthCheckService, I saw server had enable health-checking from log, but client do nothing. I do not know client how to do about health-checking.

[Falco20019]

std::string server_address("0.0.0.0:50051");
const grpc::string kHealthyService("healthy_service");

// Server code:
EnableDefaultHealthCheckService(true);
ServerBuilder builder;
builder.AddListeningPort(server_address, grpc::InsecureServerCredentials());
std::unique_ptr<Server> server(builder.BuildAndStart());

HealthCheckServiceInterface* service = server->GetHealthCheckService();
service->SetServingStatus(kHealthyService, true);

// Client code:
HealthCheckRequest request;
request.set_service(kHealthyService);

HealthCheckResponse response;
ClientContext context;
std::shared_ptr<Channel> channel = CreateChannel(server_address, InsecureChannelCredentials());
std::unique_ptr<Health::Stub> hc_stub = grpc::health::v1::Health::NewStub(channel);
Status s = hc_stub->Check(&context, request, &response);
EXPECT_EQ(Status::OK.error_code(), s.error_code());
if (s.ok()) {
	EXPECT_EQ(HealthCheckResponse::SERVING, response.status());
}

server->Shutdown();

[xiaoyulei]

@Falco20019 Clients must write code to call Check manually? Not like server only EnableDefaultHealthCheckService? I thought it is only a option.

[Falco20019]

This service is meant mostly for load balancers. You can just call a client method even if the client is not there yet or if there are network problems. GRPC will block the call and retry until the client can execute the call or the deadline is defined and reached. For that you don’t have to enable this option at all.

[yang-g]

@YuleiXiao what do you want to achieve? Indeed the service makes more sense for load balancers and (centralized) health checkers.

[xiaoyulei]

@yang-g I thought health check can be used to check if network connection is OK

[yang-g]

@YuleiXiao
Yes. Probing the health check service can be a way to check the connection but it is a bit heavy weight, as it does not only check the connection is alive but also check whether the server is actively serving the rpcs and it also requires a grpc client.

Assuming you are using C++ API, there are other ways you can monitor the connectivity. You may want to use the C++ channel state API e.g. https://github.com/grpc/grpc/blob/master/include/grpc++/impl/codegen/channel_interface.h#L74

There is also built-in keep-alive style checks that can be tuned with channel arguments here:
https://github.com/grpc/grpc/blob/master/include/grpc/impl/codegen/grpc_types.h#L217