Uploaded image for project: 'Cassandra'
  1. Cassandra
  2. CASSANDRA-12587

Log when there is a timestamp tie that is being broken

    Details

    • Type: Improvement
    • Status: Open
    • Priority: Major
    • Resolution: Unresolved
    • Fix Version/s: None
    • Component/s: Core
    • Labels:
      None

      Description

      When there is a timestamp tie, it can be very difficult to discern what has happened, since currently the columns will resolve individually by value. CASSANDRA-6123 would make this a bit more deterministic, but that would also make scenarios like this nearly impossible to troubleshoot. Since timestamp ties should be fairly rare, I propose we at least log the row key that had a tie so operators are aware that something that should almost never happen, is happening.

        Activity

        Hide
        jasobrown Jason Brown added a comment -

        +1

        Is it worthwhile to expose a metric, as well? I'm suspecting not (as, hopefully, it will always output zero), but just throwing out the idea for discussion.

        Show
        jasobrown Jason Brown added a comment - +1 Is it worthwhile to expose a metric, as well? I'm suspecting not (as, hopefully, it will always output zero), but just throwing out the idea for discussion.
        Hide
        brandon.williams Brandon Williams added a comment -

        Hmm, I'm not sure how useful that would be. Ok, there have been 3 ties, now what? Unless you lost the logs somehow that doesn't seem very useful to me.

        Show
        brandon.williams Brandon Williams added a comment - Hmm, I'm not sure how useful that would be. Ok, there have been 3 ties, now what? Unless you lost the logs somehow that doesn't seem very useful to me.
        Hide
        jasobrown Jason Brown added a comment -

        I was thinking that it might be easier to trigger a monitoring system alert based on a metric going above zero, rather than simply monitoring logs. The log would be more informational, for sure, with the CQL partition.

        With or without a metric is fine with me, but the log entry I'm definitely +1 on

        Show
        jasobrown Jason Brown added a comment - I was thinking that it might be easier to trigger a monitoring system alert based on a metric going above zero, rather than simply monitoring logs. The log would be more informational, for sure, with the CQL partition. With or without a metric is fine with me, but the log entry I'm definitely +1 on

          People

          • Assignee:
            Unassigned
            Reporter:
            brandon.williams Brandon Williams
          • Votes:
            3 Vote for this issue
            Watchers:
            12 Start watching this issue

            Dates

            • Created:
              Updated:

              Development