There are minimum two differences which I suspect.
1) In the testcase the g_service is global, but in my case it is not a
global object.
It's a class object member. And when the class object dies, it also
dies.
In the testcase, multiple tests are sharing this global object.
Another scenario can be two
tests don't share any state at all. (complete reset).
2) In my case the RegisterService part is being done in a dynamically
loaded library. (.so).
In my testcase, I create a Builder, then do a DL_OPEN a plugin
lib.
I have a helper abstrct class, for which the plugin provides the
concrete implementation.
Then I pass the Builder object to helper object--gets dispatched to
the concrete implementor.
There the concrete implemenator in that shraed library adds some
services to the Builder.
The concrete implementor also stores the server (unique_ptr) object.
In the TearDown step, I ask the implmenator to ShutDown.
It's exactly similar, but I see two differences.
One nothing is global and two the different libraries. (Parent has
builder, and DL-OPENed has the service).
On Wednesday, March 23, 2022 at 5:11:55 PM UTC [email protected] wrote:
> Well, what issue are you having that's not currently exposed in our test
> suite?
>
> On Wednesday, March 23, 2022 at 8:25:30 AM UTC-7 [email protected]
> wrote:
>
>> Hi Guys,
>> I am new here ( as well as new to google groups).
>> We are trying to do something similar to what "christo" is doing.
>> In our case - we are trying to create automated C++ tests ( using
>> gtest).
>> In our Setup() -- we want to create server/ register services
>> In our TearDown() -- we want to shutdown server/ delete every thing.
>> -------------
>> So that all the tests are exactly same precondition in terms of servers.
>> We are running the server and it's Wait function ( event loop) in a
>> separate thread.
>> One thread is client, the testcases and another thread is server --
>> wjhich waits on the port.
>> All this effort is also aimed to create a debugable server side testcases.
>> As Chrsto has mentiond there are tickets which are closed without
>> resolution.
>> https://github.com/grpc/grpc-go/issues/928
>> But in case if anyone has any ideas/suggestions, it will be helpful.
>>
>> On Wednesday, July 21, 2021 at 6:32:04 PM UTC+1 [email protected] wrote:
>>
>>> What assert are you seeing exactly? We have tests explicitly that are
>>> creating the same service multiple times within the same process, so this
>>> is supposed to be a supported use case:
>>>
>>>
>>> https://github.com/grpc/grpc/blob/master/test/cpp/server/server_builder_test.cc
>>>
>>> On Wednesday, July 14, 2021 at 11:57:18 AM UTC-7 [email protected]
>>> wrote:
>>>
>>>> I've got a server class that has a `start` method and a `stop` method.
>>>> The listening port is unknown until a user calls start.
>>>>
>>>> As such, I build the server there, something like:
>>>>
>>>> void GrpcStreamingServerImpl::start(QUrl const& endpoint)
>>>> {
>>>> stop();
>>>>
>>>> ::grpc::ServerBuilder builder;
>>>> builder.AddListeningPort(endpoint.toString().toStdString(),
>>>> grpc::InsecureServerCredentials());
>>>> builder.RegisterService(&m_service);
>>>> m_completionQueueCalls = builder.AddCompletionQueue();
>>>> m_completionQueueNotifications = builder.AddCompletionQueue();
>>>> m_server = builder.BuildAndStart();
>>>>
>>>> m_running = true;
>>>> ....
>>>>
>>>> and then I tear everything down when the server stops. Something like:
>>>>
>>>> if(!m_running)
>>>> {
>>>> return;
>>>> }
>>>>
>>>> m_running = false;
>>>>
>>>> {
>>>> std::lock_guard<std::mutex> local_lock_guard {m_mutexSessions};
>>>> for(const auto& it : m_sessions)
>>>> {
>>>> it.second->finish();
>>>> }
>>>> }
>>>>
>>>> if(m_server)
>>>> {
>>>> m_server->Shutdown();
>>>> m_completionQueueCalls->Shutdown();
>>>> m_completionQueueNotifications->Shutdown();
>>>>
>>>> // Wait for the threads to exit before allowing the destruction
>>>> of the completion queues they use
>>>> if(m_callQueueThread && m_callQueueThread->isRunning() )
>>>> {
>>>> m_callQueueThread->wait();
>>>> }
>>>>
>>>> if(m_notificationQueueThread &&
>>>> m_notificationQueueThread->isRunning())
>>>> {
>>>> m_notificationQueueThread->wait();
>>>> }
>>>>
>>>> delete m_completionQueueNotifications;
>>>> delete m_completionQueueCalls;
>>>> delete m_server;
>>>> ....
>>>>
>>>>
>>>> Problem is, if the user wants to start, stop, and start again, the
>>>> RegisterService call results in abort being called and complains a service
>>>> is already registered.
>>>>
>>>> I tried to move the registration part to the constructor and only add a
>>>> listening port when start() is called, but that results in a seg fault.
>>>> Builder isn't a singleton, although I dunno how or why it retains the
>>>> registration after the server object it returned was destroyed.
>>>>
>>>> A little googling and I found issues were submitted regarding this, but
>>>> there are no plans to implement an "unregisterService" method. So, what
>>>> does one to work around this problem?
>>>>
>>>>
--
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 [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/grpc-io/0930228e-3b8b-4b8d-a652-30532ce3b425n%40googlegroups.com.
