Uploaded image for project: 'Kudu'
  1. Kudu
  2. KUDU-2486

Leader should back off heartbeating to failed followers

    XMLWordPrintableJSON

Details

    • Improvement
    • Status: Open
    • Major
    • Resolution: Unresolved
    • 1.7.1
    • None
    • consensus
    • None

    Description

      At the time of writing, the replica leader -> follower heartbeat mechanism does not have a backoff mechanism built in. Rather it simply sends a heartbeat every configured period (say, 500ms). If a server is offline this can cause log spam until that replica is evicted, and if a server is overloaded this lack of a backoff contributes to the problem.

      Since we now have pre-election support, having leaders slow down their heartbeat attempts when follower requests are returning errors should not cause unnecessary leader elections, so backing off is feasible.

      Attachments

        Activity

          People

            Unassigned Unassigned
            mpercy Mike Percy
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

            Dates

              Created:
              Updated: