Tomcat 7.0.55 and 8.0.9 on Mac OS X Mavericks latest I am doing server sent events with AsyncContext. Basic stuff is working.
I have set the timeout to be infinite so onTimeout() never happens. That’s fine — I don’t mind hangers on in this case. Testing with curl, I can get a constant stream of events pretty much as they happen. However, when I kill the curl, I see the socket that was being used go completely away according. However, the background message pump is still going (I didn’t tell it to stop) but I noticed that no event listener method other than onComplete is called and that’s only when I explicitly complete the context. I am planning on maintaining a large a large collection of AsyncContexts and had thought that the closing of the socket would have caused onError to be called — it does not get called under this circumstance. In a busy system, when I’m running through the queue, I call PrinterWriter.checkError() to see if the client has gone away and then just complete the AsyncContext. However, if there are no messages to send, then orphaned AsyncContext objects will build up. This is problematic since anyone authenticated can just connect/disconnect on SSE topics and run out of resources. I can periodically scan through all saved AsyncContext objects and see if the PrintWriter on the response has an error, but this seems wrong. Until I put the checkError into my code, a test message pump would happily call write on the PrinterWriter with no exceptions being thrown. I let the test run over night and still no exception on my log. Question: What errors cause onError to actually get called? It seems wrong that PrintWriter on the response is not throwing an exception when there is an error. Am I missing something? The socket has gone away (this is not the case of an long idle connection) and Tomcat must surely know that it did. Why does it not bubble this up to the AsyncListener? I have not had a chance to try other Servlet engines but the behaviour was consistent between 7 & 8 latest. I would really like to avoid having to do periodic scanning of AsyncContext to see if the the PrintWriter noticed there’s an error. This also would not seem to catch the case of a closed socket with no writes pending. thanks! --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
