Uploaded image for project: 'JGroups'
  1. JGroups
  2. JGRP-505

UNICAST: sequence numbering after merge leads to messages never being delivered

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Done
    • Icon: Major Major
    • 2.5
    • 2.4.1
    • None

      During refactoring of TUNNEL to use TP I ran MultiplexerMergeTest very often. This test basically consists of two members A and B that are exchanging message through GossipRouter. When GossipRouter is killed members form their own subgroups that are subsequently merged once router starts again.

      I noticed that on very rare occasions (1 in 20 test runs) one of the members (say A) stops passing up unicast messages from B above UNICAST protocol. More detailed logging and inquiry at UNICAST revealed that upon merge A stops passing up messages from B and keeps adding messages from B seq>1 into B's window but does not pass up messages further since message seq B1 is missing. After further investigation I concluded that message B1 must have been delivered to A prior to MergeView. MergeView wiped clean all windows and added next sequence B2 to appropriate window.

      I am not sure if this is only applicable to Multiplexer setup. The issue is hard to reproduce. I will see if it is possible to make a unit test to reproduce it.

              vblagoje Vladimir Blagojevic (Inactive)
              vblagoje Vladimir Blagojevic (Inactive)
              Votes:
              0 Vote for this issue
              Watchers:
              0 Start watching this issue

                Created:
                Updated:
                Resolved: