1 Reply Latest reply: Apr 6, 2012 11:26 AM by 743514 RSS

    Missing entry in replicated cache on new cluster member

      Coherence 3.5.2.

      I have 4 cluster members and a replicated cache with one key.

      I start a new cluster member that accesses the cache.

      All coherence log entries look fine, no issues or weirdness accessing the replicated cache service.

      The new cluster member doesn't see the one key in the replicated cache, and the cache size is reported as zero.

      I start the coherence command line utility in another client, it CAN see the cache key.
      If I put another key in the cache, the new and pre-existing key both become visible
      to the above cluster member that previously couldn't see it.

      Known bug? All I could find was this bug in the 3.7.1 release notes:

      Fixed an issue with ReplicatedCache that could cause new service
      members to hold outdated lease information.

      This is very reproducible, but not in a simple case I can give you.
      Since I can't upgrade from 3.5.2 right now, is there a workaround?
      I need to see that replicated cache entry in newly joined cluster members.

        • 1. Re: Missing entry in replicated cache on new cluster member
          FYI, with no change in code a machine reboot has cleared the problem, which just makes me worry more since
          this is (will be) all about rolling upgrade changes in production when we deploy the new members, and machine
          restarts are not an option there.

          If the new nodes come up and can't see what's in the replicated cache, that's a big problem.
          If it's going to be a problem, I need to know so I can try to work around it.
          It was consistently a problem for me for two days, and went away when I rebooted this morning.

          Also FYI, each time I would reproduce the problem the last two days, I'd be starting clusters
          from scratch for testing. So I'd start the old software, get things running, data populated, etc,
          then start the new member that needed to see the replicated cache entries of the existing cluster,
          and they would be missing.

          Can't reproduce it today after the reboot though.

          Yes, guilty of losing clarity by trying to add clarity by being redundant, apologies.