Azure / kube-egress-gateway / 29442009881

81%

push

github

Periodically requeue GatewayVMConfiguration to self-heal in-place VMSS drift (#1381)

* Periodically requeue GatewayVMConfiguration to self-heal in-place VMSS drift

The gatewayvmconfiguration controller only reconciled on GatewayVMConfiguration
CR changes and Node Create/Delete events (the Node predicate's UpdateFunc returns
false) and never set a periodic requeue. Gateway VMSS pools upgrade in place (no
surge, no Node event), so when an out-of-band VMSS model rollout / ManualUpgrade
strips the secondary egress ipconfig from the instances, no event fires and the
drift is never repaired. In the originating incident this left static egress down
for ~12 hours until the customer scaled the gateway node pool (a Node event) to
force a reconcile.

Return ctrl.Result{RequeueAfter: 5m} from the successful reconcile path so each
GatewayVMConfiguration re-reconciles on a fixed cadence. That reconcile walks the
instances and force-updates any with a missing primary/secondary IP, so
instance-level NIC drift is detected and repaired within one interval regardless
of whether a Node or CR event was emitted.

- Fixed 5-minute interval (const); no flag/field configurability by design.
- Error paths already requeue via controller-runtime backoff; the delete path is
  unchanged (no requeue while terminating); the node-event path is unchanged.

* test: add unit test

4 of 4 new or added lines in 1 file covered. (100.0%)

3599 of 4438 relevant lines covered (81.1%)

0.95 hits per line