kubeovn / kube-ovn / 23588768977

24%

push

github

fix(controller): defer subnet processing until vlan is fully ready (#6527)

* fix(controller): clear subnet IP status when vlan conflict is detected

When two VLANs with the same ID are created on the same provider network,
a race condition causes the conflict subnet to retain its IP range in
status. This happens because:

1. The subnet handler can process the subnet before the vlan handler
   marks the vlan as conflicting (informer cache propagation delay of
   ~3ms), allowing IPAM allocation and IP status to be set.

2. When the vlan handler detects a conflict, it returns early without
   re-enqueuing associated subnets, so the subnet is never re-validated.

3. Even when the subnet is re-processed and detects the vlan conflict,
   patchSubnetStatus serializes the full status (including the stale IP
   range) without clearing IP fields.

Fix all three issues:
- In handleAddVlan/handleUpdateVlan: re-enqueue subnets referencing the
  conflicting vlan so they can re-validate.
- In handleAddOrUpdateSubnet: when vlan validation fails, remove the
  subnet from IPAM and clear all IP status fields before patching.

Signed-off-by: Mengxin Liu <liumengxinfly@gmail.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Mengxin Liu <liumengxinfly@gmail.com>

* fix(controller): address review feedback for vlan conflict handling

Address Copilot review comments:

1. Prevent queue churn: only re-enqueue subnets when vlan conflict
   status transitions from false to true (not on every retry). Save
   wasConflict before checkVlanConflict and gate the re-enqueue.

2. Scope IPAM cleanup: introduce errVlanConflict sentinel error in
   validateSubnetVlan and only clear IPAM/IP status fields when
   errors.Is(err, errVlanConflict) is true. Transient lister errors
   no longer trigger IPAM deletion.

3. Add unit tests: Test_validateSubnetVlan_conflict verifies the
   sentinel error is correctly returned for conflict vs normal vs
   missing... (continued)

9 of 33 new or added lines in 2 files covered. (27.27%)

13360 of 55422 relevant lines covered (24.11%)

0.28 hits per line