Sirius-LiXiaoQi opened a new issue, #16397:
URL: https://github.com/apache/dubbo/issues/16397

   ### Pre-check
   
   - [x] I am sure that all the content I provide is in English.
   
   
   ### Search before asking
   
   - [x] I had searched in the 
[issues](https://github.com/apache/dubbo/issues?q=is%3Aissue) and found no 
similar issues.
   
   
   ### Apache Dubbo Component
   
   Java SDK (apache/dubbo)
   
   ### Dubbo Version
   
   Dubbo 3.3.6, Triple protocol (`dubbo.protocol.name=tri`), instance mode , 
JDK 8. 
   
   ### Steps to reproduce this issue
   
   Graceful shutdown tears down the Triple `PathResolver` mapping *before* 
in-flight requests have drained.
   A request that has already been accepted by the server, but has not yet 
resolved its invoker, will fail with
   `UNIMPLEMENTED` once `afterUnExport` runs.
   
   The race can be reproduced deterministically with two breakpoints on the 
**provider** side — no timing luck
   required. Use a **thread-level** (suspend-thread, not suspend-all) 
breakpoint so the shutdown thread and the
   request thread can be released independently.
   
   Provider service (anything works, the body is irrelevant):
   
   ```java
   @DubboService
   public class DemoServiceImpl implements DemoService {
       @Override
       public String sayHello(String name) {
           return "hello " + name;
       }
   }
   
   Consumer: a single call is enough.
   
   @DubboReference(timeout = 500000)
   private DemoService demoService;
   
   demoService.sayHello("dubbo");
   
   Reproduction procedure (provider JVM, debugger attached):
   
   1. Set a thread-suspending breakpoint at
   
org.apache.dubbo.rpc.protocol.tri.h12.grpc.GrpcRequestHandlerMapping#getRequestHandler,
 on the line
   
   Invoker<?> invoker = pathResolver.resolve(path.getPath(), group, version);   
// GrpcRequestHandlerMapping.java:62
   
   1. Issue the consumer call. The request thread is now suspended before the 
path is resolved — i.e. the
   request has already been accepted by a fully healthy provider.
   2. Send TERM to the provider process to start graceful shutdown, with a 
second thread-suspending breakpoint at
   
   pathResolver.unregister(invoker);   // TripleProtocol.java:104, in 
TripleProtocol$1#afterUnExport
   
   2. Step over that line so the unregister actually executes, then leave the 
shutdown thread suspended.
   TriplePathResolver no longer contains the path mapping for the service.
   3. Release the request thread from step 1 and step over line 62. 
resolve(...) now returns null, so control
   falls into
   
   if (invoker == null) {
       throw notFound();               // GrpcRequestHandlerMapping.java:64
   }
   
   Result
   
   - Provider log:
   **Invoker for gRPC not found**
   - Consumer:
   **org.apache.dubbo.rpc.RpcException: ... UNIMPLEMENTED : Invoker for gRPC 
not found**
   
   Without the debugger, the same thing happens naturally on every shutdown 
that overlaps with concurrent
   traffic; the breakpoints only make the window deterministic.
   
   ### What you expected to happen
   
   The request accepted in step 1 was accepted while the service was still 
exported, and the configured graceful
   shutdown wait (dubbo.service.shutdown.wait) had not elapsed. It should have 
been served normally.
   
   Concretely:
   
   1. PathResolver.unregister(invoker) should not take effect until in-flight 
requests have drained — the path
   mapping must outlive the request-draining phase rather than being removed at 
the start of it.
   2. If the server has genuinely stopped accepting requests, the failure 
should be retriable: UNAVAILABLE,
   or an HTTP/2 GOAWAY that lets the consumer re-route to another provider 
instance.
   
   UNIMPLEMENTED is semantically wrong here — it means "this server does not 
implement the service", so
   consumer-side cluster fault tolerance treats it as a permanent definition 
error and does not fail over. Every
   rolling restart of a Triple provider therefore surfaces as business-visible 
errors on the consumer, which
   defeats the purpose of graceful shutdown.
   
   ### Anything else
   
   Code paths involved (3.3.6):
   
   - org.apache.dubbo.rpc.protocol.tri.TripleProtocol$1#afterUnExport — 
pathResolver.unregister(invoker)
   (TripleProtocol.java:111), registered by TripleProtocol#export
   - 
org.apache.dubbo.rpc.protocol.tri.h12.grpc.GrpcRequestHandlerMapping#getRequestHandler
   (GrpcRequestHandlerMapping.java:62 resolve, :64 throw notFound())
   - org.apache.dubbo.rpc.PathResolver#unregister / #resolve,
   org.apache.dubbo.rpc.protocol.tri.TriplePathResolver
   
   The same window applies to any RequestHandlerMapping that resolves through 
PathResolver during shutdown,
   not only the gRPC one.
   
   ### Do you have a (mini) reproduction demo?
   
   - [x] Yes, I have a minimal reproduction demo to help resolve this issue 
more effectively!
   
   ### Are you willing to submit a pull request to fix on your own?
   
   - [ ] Yes I am willing to submit a pull request on my own!
   
   ### Code of Conduct
   
   - [x] I agree to follow this project's [Code of 
Conduct](https://www.apache.org/foundation/policies/conduct)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to