Uploaded image for project: 'JGroups'
  1. JGroups
  2. JGRP-2286

VERIFY_SUSPECT can return false positives if member is killed and immediately restarted

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Won't Do
    • Icon: Major Major
    • None
    • 4.0.13
    • None

      When a member is killed, it is suspected and VERIFY_SUSPECT attempts to send a message to the member to verify that it has left. If, however, the member is restarted before the VERIFY_SUSPECT sends its message, the newly started member (with same physical address, but different UUID) can receive it and erroneously reply with an I_AM_NOT_DEAD response. This results in a zombie member in the view that will remain until the reincarnated member leaves.

      To fix this, we should add the address of the suspected member to the VerifyHeader and validate that the member is us before sending an I_AM_NOT_DEAD response.
      Additionally, we can respond with a new MBR_IS_DEAD response on behalf of the suspected member, in the case that the message was receive by the reincarnated member to expedite the removal of the suspected member from the view.

              pferraro@redhat.com Paul Ferraro
              pferraro@redhat.com Paul Ferraro
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated:
                Resolved: