mina-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jonathan Valliere (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (DIRMINA-1076) Leaking NioProcessors/NioSocketConnectors hanging in call to dispose
Date Sun, 11 Feb 2018 15:22:00 GMT

    [ https://issues.apache.org/jira/browse/DIRMINA-1076?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16359966#comment-16359966

Jonathan Valliere commented on DIRMINA-1076:

Here is the patch to fix your issue; it also passes maven:test.  It passes the tests  but
is not the final patch because I have some concerns about it long term.
diff --git a/mina-core/src/main/java/org/apache/mina/core/polling/AbstractPollingIoProcessor.java
index 79885fa..503df41 100644
--- a/mina-core/src/main/java/org/apache/mina/core/polling/AbstractPollingIoProcessor.java
+++ b/mina-core/src/main/java/org/apache/mina/core/polling/AbstractPollingIoProcessor.java
@@ -661,7 +661,9 @@
                     // And manage removed sessions
                     nSessions -= removeSessions();
+                    assert nSessions > -1 : "Number of Sessions has
gone negative";
                     // Last, not least, send Idle events to the idle
@@ -695,9 +697,8 @@
                         for (Iterator<S> i = allSessions();
i.hasNext();) {
                             IoSession session = i.next();
-                            scheduleRemove((S) session);
                             if (session.isActive()) {
+                                scheduleRemove((S) session);
                                 hasKeys = true;

Emmanual,  while this patch fixes the problem Christoph is seeing.  I am concerned that
relying on this change could cause a problem where the SelectionKey.isValid() i.e. `session.isActive()`
is false but the Session was not previously removed correctly.  The AbstractIoProcessor is
relying on the validity of the SelectionKey and the Set<SelectionKey> to internally
manage active sessions which is a really bad move because the SelectionKeys can become invalid
totally independent of the AbstractIoProcessor which could leave pending writing data and
other things hanging forever.  Fundamentally, the problem is that, in the 2.0 master the
Sessions are being removed TWICE from AbstractIoProcessor leaving the Session Counter in the
Negative.  One way to fix the problem safely is to store all the Sessions for the IoProcessor
as a List and use that as the master list of 'active' Sessions but I don't know if there is
some lazy Session stealing between IoProcessors in Mina which could conflict with that idea.


> Leaking NioProcessors/NioSocketConnectors hanging in call to dispose
> --------------------------------------------------------------------
>                 Key: DIRMINA-1076
>                 URL: https://issues.apache.org/jira/browse/DIRMINA-1076
>             Project: MINA
>          Issue Type: Bug
>    Affects Versions: 2.0.16
>            Reporter: Christoph John
>            Priority: Major
>         Attachments: mina-dispose-hang.txt, mina-test-log.txt, mina-test-patch.txt
> Follow-up to mailing list discussion.
> I was now able to reproduce the problem with a MINA test. Or let's say I did the brute-force
approach by re-running one test in an endless loop.
> I have attached a patch of AbstractIoServiceTest (against [https://github.com/apache/mina/tree/2.0])
and a stack trace. After a few loops the test is stuck. You can see a lot of threads hanging
in dispose() and the test is stuck when it tries to dispose the acceptor.
> What is a little strange is that the javadoc says that connector.dispose(TRUE) should
not be called from an IoFutureListener, but in the test it is done anyway. However, changing
the parameter to FALSE does not help either.
>  Is there anything that can be done to prevent this hang?

This message was sent by Atlassian JIRA

View raw message