Uploaded image for project: 'Lustre'
  1. Lustre
  2. LU-13648

Route status can be set incorrectly via lnet_notify()

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: Minor
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: Lustre 2.14.0
    • Labels:
      None
    • Severity:
      3
    • Rank (Obsolete):
      9223372036854775807

      Description

      Found an issue when testing my patch to skip health/resends for single rail config - https://review.whamcloud.com/#/c/38448

      Problem is in lnet_notify(). LND tells us some nid went down. But we calculate aliveness based on health values and NI status as reported by discovery:

              /* recalculate aliveness */
              alive = lnet_is_peer_ni_alive(lpni);
      
      static inline bool
      lnet_is_peer_ni_alive(struct lnet_peer_ni *lpni)
      {
              bool halive = false;
      
              halive = (atomic_read(&lpni->lpni_healthv) >=
                       (LNET_MAX_HEALTH_VALUE * router_sensitivity_percentage / 100));
      
              return halive && lpni->lpni_ns_status == LNET_NI_STATUS_UP;
      }
      

      And then use that to set routes up/down:

                      lp = lpni->lpni_peer_net->lpn_peer;
                      lp->lp_alive = alive;
                      list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
                              lnet_set_route_aliveness(route, alive);
      

      But, at least with gnilnd, we can get notification from LND before any tx sent to the router has failed. So health may not have been ding'd yet. This can lead to routes in the wrong up/down state.

        Attachments

          Activity

            People

            • Assignee:
              hornc Chris Horn
              Reporter:
              hornc Chris Horn
            • Votes:
              0 Vote for this issue
              Watchers:
              3 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: