HeartbeatInterceptor executed before IdleResourceInterceptor

[freongrr]

I'm using client heartbeats, but IdleResourceInterceptor keeps cancelling my resources.

It turns out that IdleResourceInterceptor#inspect() is never called (and the "MAX_INACTIVE" property is never updated), because HeartbeatInterceptor is invoked first, and it cancels the request from the client.

How about bumping up IdleResourceInterceptor in the list of default interceptors in AtmosphereFramework?

For reference, this is my handler definition:

    @ManagedService(
      path = "/push",
      interceptors = {
        AtmosphereResourceLifecycleInterceptor.class,
        TrackMessageSizeInterceptor.class,
        SuspendTrackerInterceptor.class,
        GwtRpcInterceptor.class,
        AtmosphereMessageInterceptor.class,
      },
      atmosphereConfig = {
        "org.atmosphere.interceptor.HeartbeatInterceptor.clientHeartbeatFrequencyInSeconds=60",
        "org.atmosphere.cpr.CometSupport.maxInactiveActivity=180000"
      }
    )
    public class TestPushManager {
      ...
    }

And this is the order of interceptors when I hit AsynchronousProcessor#action():

• UUID Tracking Interceptor
• Track Message Size Interceptor using |
• CORS Interceptor Support
• Default Response's Headers Interceptor
• Browser Padding Interceptor Support
• Android Interceptor Support
• Heartbeat Interceptor Support
• SSE Interceptor Support
• JSONP Interceptor Support
• Atmosphere JavaScript Protocol
• org.atmosphere.interceptor.WebSocketMessageSuspendInterceptor
• Browser disconnection detection
• org.atmosphere.interceptor.IdleResourceInterceptor

[jfarcand]

@freongrr It really depends on how you see it, IdleResourceInterceptor means there is no real byte sent on the wire. That exclude heartbeat bytes ... if I change the framework the way you are asking for, that means IdleResourceInterceptor will be useless in that case. So not sure...

[nicolasllk]

I am having the same issue. What jfarcand says is logical but is there any way to achieve this without tampering with the framework?

[freongrr]

In my case, I update the MAX_INACTIVE attribute on the resource in my onHeartbeat() method.

But then, shouldn't HeartbeatInterceptor do that?

[nicolasllk]

In either case, shouldn't heartbeat lunch an exception when sending data through a closed connection?

[jfarcand]

@nicolasllk Lunch an exception to where? If you turn the log to trace, you will see the exeption.

[jfarcand]

@freongrr If you make the change in the code by moving the interceptor, does it help? If yes, pull request welcomed :-)

[freongrr]

Yes, the change is easy. I'll have to keep it running for a while to see how it works.

[jfarcand]

@freongrr Any update?

[freongrr]

So, I could not test because the load balancer my company uses drops some websocket messages...

[jfarcand]

@freongrr Any feedback? I'm tempted to switch the order, but I first want feedback from you.

[freongrr]

As I said above, I had to stop using websockets. The change worked during my limited testing, but I could not see how it worked in a real deployment.

[jfarcand]

With #1877 , IdleResourceInterceptor will always be executed first. Closing as fixed, but in case you have time, please test :-)