Uploaded image for project: 'Infinispan'
  1. Infinispan
  2. ISPN-3426

L1 inconsistency in tx caches when backup owner replies to remote get

This issue belongs to an archived project. You can view it, but you can't modify it. Learn more

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Duplicate
    • Icon: Critical Critical
    • 6.0.0.Final
    • 6.0.0.Final
    • None

      Consider the following scenario

      node1: performs a remote get
node2 (backup owner): receives and replies to the remote get with v1
node1: receives the repli and store in L1_ key => v1
node2 (backup owner): commits a new value (v2) for the key (i.e. processes a commit command)
node3 (primary owner): sends the invalidation (but not for node1 because it hasn't received the remote get yet) and commits a new value (v2) for the key (i.e. processes a commit command) 
node3 (primary owner): replies to the remote get with v2
node1: ignores the reply because it used the node2 reply

      conclustion: node1 keeps the old value stored in L1.

      Possible solutions described here:
      L1 consistency for transactional caches

      Could also be interesting:
      Staggered get question
      and
      ISPN-825

              wburns@redhat.com Will Burns
              pruivo@redhat.com Pedro Ruivo
              Archiver:
              rhn-support-adongare Amol Dongare

                Created:
                Updated:
                Resolved:
                Archived: