Uploaded image for project: 'DC/OS'
  1. DC/OS
  2. DCOS_OSS-4398

dcos-net continously restarting systemd-networkd on a bare-metal server with bond interfaces

    Details

    • Sprint:
      Networking: RI-7 Sprint 33, Networking: RI-7 Sprint 32
    • Story Points:
      5

      Description

      On a bare-metal server with two bond interfaces (LAN and NFS), dcos-net service restarts systemd-network indefinitely, causing interfaces flapping (and possibly ports disabling on the switches).

      Steps to reproduce:

      • install CoreOS 1855.5.0 (latest stable) and configure bond0 and bond1
      • OS is perfectly stable
      • install DC/OS (i.e. as a slave)
      • reboot
      • network starts to flap indefintely, in rare cases it stops after MANY up/down cylcles
        As proof, setting "Restart=no" in dcos-net.service stops network flapping.

      Probably, bonding has timing issues different from normal interfaces and dcos-net keeps to restart systemd-network waiting for networking coming up.

      WORKAROUND
      Commenting out:

      ExecStartPre=/opt/mesosphere/active/dcos-net/dcos-net/bin/dcos-net-setup.py networkd add /opt/mesosphere/etc/dcos.network
      

      in /etc/systemd/system/dcos-net.service unit file solves the issue.

        Attachments

          Activity

            People

            • Assignee:
              sergeyurbanovich Sergey Urbanovich
              Reporter:
              mimmus mimmus
              Team:
              Networking Team
              Watchers:
              Deepak Goel, Lisa Gunn (Inactive), Mergebot, mimmus, Sergey Urbanovich
            • Watchers:
              5 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: