Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ecf-dev] connections/threads are not cleaned up after failing importService call

On 9/19/2017 6:00 AM, Peter Hermsdorf wrote:

Hi again,

the workaround described below does not work in all circumstances.

The problem still seems to be that remote services can only be successfully registered if the client uses the exact same hostname for the endpoint ID as used by the server itself. Otherwise the containerID's in org.eclipse.ecf.remoteservice.RemoteServiceID.namespaceEquals(BaseID) does not match and the service registration fails (I'm not sure if there are additional other places where the hostname matters while service import/export).

The problem for me is, that the "local" hostname where the service is provided is not always the network visibly name the clients need to use to connect to the server.

Is there a way to make the remote services work then the client (service consumer) and the server (service provider) use different hostnames for the server?


Hi Peter.   I'm sorry I haven't been able to respond to this whole thread as there are a number of things...e.g. the workaround you are using, other possible work-arounds, etc.   I've been very occupied recently.   But should we focus on this question specifically for the moment?

It is possible to customize remote services in a number of ways and so I'm pretty sure what you are asking for is possible.   It might mean creating a custom distribution provider (e.g. extending an existing one), and/or some additional customization of the remote services impl, but I think it's probably possible.

I will also go back to your original posting to understand your use case but because my memory is limited, it might also be helpful if you would review the behavior you would like to see for your environment...e.g. what network interface(s) the server is listening on, what externally-visible names (i.e. dns) names it has and how that relates to the interface(s) and associated ip addresses, and what you would like the clients to use in order to connect to that server and import services.   

Also maybe a short review on how you are doing export, discovery and import (e.g. via direct calls to RemoteServiceAdmin, via registerService, etc).

If you would prefer to not do this publicly I understand...please just contact me at slewis at composent.com and we'll take things offline.

Thanks,

Scott


I'm using ecf generic server.

Thanks for any hints!

Bye Peter


Am 18.09.2017 um 16:22 schrieb Peter Hermsdorf:

After some further debugging I was able to workaround the described problem by making the connection to the virtual machine work regardless of the used hostname in the endpoint description.

I used some constant value for ENDPOINT_ID and used RemoteConstants.ENDPOINT_CONNECTTARGET_ID to define the "real network" target/host to connect to:

props.put(RemoteConstants.ENDPOINT_CONNECTTARGET_ID, "ecftcp://" + host + ":8889/server");
props.put(ENDPOINT_ID, "p4://server");

So no more hostname matching to identify the correct service on the target machine.

But the original problem still remains unresolved ...

Bye, Peter

Am 18.09.2017 um 14:24 schrieb Peter Hermsdorf:

Hi Scott,

thanks for your suggestion, but the described  fix does not seem to work. The returned container is always null and so cannot be disconnected. This is caused by the importEndpoint of the importReference being null.

This is what I'm doing:

1) register my remote service and a reconnect listener for that service:

        final ImportRegistration registration = remoteServiceAdmin.importService(endpointDescription);

        final ReconnectServiceAdminListener reconnectListener = new ReconnectServiceAdminListener(remoteServiceAdmin,
                containerManager,
                registration, endpointDescription);
        context.registerService(RemoteServiceAdminListener.class.getName(), reconnectListener, null);

2) inside the ReconnectListener I basically try to re-import the service when the registration/connection fails:

public class ReconnectServiceAdminListener implements RemoteServiceAdminListener {

    private static final Duration RECONNECT_DELAY = Duration.ofSeconds(10);

    private final ReconnectJob reconnectJob;
    private final EndpointDescription endpointDescription;
    private final RemoteServiceAdmin remoteServiceAdmin;
    private ImportRegistration registration;

    private final IContainerManager containerManager;

    public ReconnectServiceAdminListener(final RemoteServiceAdmin remoteServiceAdmin,
            final IContainerManager containerManager, final ImportRegistration registration,
            final EndpointDescription endpointDescription) {
        this.remoteServiceAdmin = remoteServiceAdmin;
        this.containerManager = containerManager;
        this.registration = registration;
        this.endpointDescription = endpointDescription;
        reconnectJob = new ReconnectJob();
        if (registration.getException() != null) {
            reconnectJob.schedule(RECONNECT_DELAY.toMillis());
        }
    }

    @Override
    public void remoteAdminEvent(final RemoteServiceAdminEvent e) {
        final org.eclipse.ecf.osgi.services.remoteserviceadmin.RemoteServiceAdmin.RemoteServiceAdminEvent event = (org.eclipse.ecf.osgi.services.remoteserviceadmin.RemoteServiceAdmin.RemoteServiceAdminEvent) e;
        if (event.getEndpointDescription().isSameService(endpointDescription)) {
            switch (event.getType()) {
                case RemoteServiceAdminEvent.IMPORT_UNREGISTRATION:
                case RemoteServiceAdminEvent.IMPORT_ERROR:
                    reconnectJob.schedule(RECONNECT_DELAY.toMillis());
                    break;
            }
        }
    }

    class ReconnectJob extends Job {

        public ReconnectJob() {
            super("ReconnectJob");
        }

        @Override
        protected IStatus run(final IProgressMonitor monitor) {

            final org.eclipse.ecf.osgi.services.remoteserviceadmin.RemoteServiceAdmin.ImportRegistration remoteServiceImportRegistration = (org.eclipse.ecf.osgi.services.remoteserviceadmin.RemoteServiceAdmin.ImportRegistration) registration;
            final ID containerID = remoteServiceImportRegistration.getContainerID();
            final IContainer container = containerManager.getContainer(containerID);
            if (container != null) {
                container.disconnect();
            }

            registration.close();
            registration = remoteServiceAdmin.importService(endpointDescription);
            return Status.OK_STATUS;
        }
    }

}

The Thread count still increases with every run of the ReconnectJob.


Any suggestions on how to solve this issue?

Thanks!
Bye Peter

Am 16.09.2017 um 05:03 schrieb Scott Lewis:
Hi Peter,

I'm assuming that you are using the generic provider.

I think you can do what you need to do by:

1) You can cast the ImportRegistration to org.eclipse.ecf.osgi.services.remoteserviceadmin.RemoteServiceAdmin.ImportRegistration
2) Call org.eclipse.ecf.osgi.services.remoteserviceadmin.RemoteServiceAdmin.ImportRegistration.getContainerID() to get the container ID
3) Using the IContainerManager service (a singleton ecf service) and the ID from 2 call:   IContainer c = containerManager.getContainer(id) to get the associated container instance.
4) Call c.disconnect() to release the connection and associated threads.

Scott



On 9/15/2017 7:27 AM, Peter Hermsdorf wrote:
Hi,

we encountered a problem while trying to import services exported from a process running in a local VM (_not_ JVM). The VM forwards the necessary port to the host machine.

As for the problem: RemoteAdminService#importService causes a RemoteAdminServiceEvent.IMPORT_ERROR to be signalled and the connections/threads opened due to the importService call do not get closed, even though we call ImportRegistration#close() on failed attempts. The latter does not happen, if there is no one listening on the port at all. Since, we have code that regularly calls importService again for services, that should be, but are not currently imported (according to RemoteAdminService#getImportedEndpoints), we slowly accumulate more and more threads.

The tcp connection to the local VM seems to be successful, but due to issues regarding requested hostname and actual hostname ECF (correctly) does not "connect the services" (see https://dev.eclipse.org/mhonarc/lists//ecf-dev/msg07259.html for a discussion of the "hostname problem - localhost != getCanonicalHostname()" in EndPointDescriptions; in this case it is host::getCanonicalHostname() != VM::getCanonicalHostname())

So importService creates threads/connections, fails to import, and even with a call to ImportRegistration#close() does not clean up the threads/connections.

Any suggestions on how to solve this issue or more generally handle this reconnect use-case are more than welcome!

Best regards,

Bye Peter


PS: we use the ecf generic provider on the latest ecf release


_______________________________________________
ecf-dev mailing list
ecf-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/ecf-dev



Back to the top