Uploaded image for project: 'Kudu'
  1. Kudu
  2. KUDU-951

Concurrent remote bootstraps corrupts replicas

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Resolved
    • Blocker
    • Resolution: Fixed
    • Public beta
    • Public beta
    • tablet
    • None

    Description

      I deleted all the tablets metadata on a TS and after restarting it everything goes to hell. I found that we use the fs_manager's UUID as the session id in remote bootstrap, effectively limiting us to 1 remote bootstrap per peer, but we still try to do multiple tablets at the same time. The first one that starts remote bootstrapping "wins", the rest get that first tablet's data. Eventually those unlucky tablets fail to bootstrap when they replace their superblock and re-open it, as they now see that it's not theirs.

      Attachments

        1. throughput.png
          26 kB
          Todd Lipcon

        Activity

          People

            mpercy Mike Percy
            jdcryans Jean-Daniel Cryans
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: