kubeovn / kube-ovn / 25917468877

27%

push

github

fix(controller): make BgpConf/EvpnConf informers optional (#6726) (#6737)

* fix(controller): make BgpConf/EvpnConf informers optional (#6726)

The kube-ovn-controller used to unconditionally add the BgpConf and
EvpnConf informers to its WaitForCacheSync list. On clusters upgraded
from <v1.16 via Helm those CRDs are missing (Helm does not re-apply
the chart's crds/ directory on `helm upgrade`), so the controller
hangs forever in "Waiting for informer caches to sync" and breaks
IPAM cluster-wide.

Treat BgpConf/EvpnConf like the other optional CRDs (NetworkAttachment,
ServiceCIDR):

* Detect the CRDs at startup via the discovery API and only register
  the informers when present.
* When missing, start a 10s background poller that picks them up once
  they appear, with no controller restart required.
* Publish the lister/synced fields only after WaitForCacheSync returns
  true so concurrent VEG reconcilers cannot observe a non-nil lister
  backed by an empty cache.
* Guard the vpc-egress-gateway consumer paths with a nil-lister check
  that surfaces a clear error and lets the workqueue retry.

Add an e2e regression test (kube-ovn/optional_crd) that deletes both
CRDs, restarts the controller and asserts the rollout completes and
IPAM still works, then restores the CRDs and verifies the controller
self-heals.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: Mengxin Liu <liumengxinfly@gmail.com>

* fix(controller): address review feedback on optional BgpConf/EvpnConf (#6726)

Two follow-ups requested by Copilot on PR #6737:

* Race-free publication of the optional listers. The
  bgpConfLister/evpnConfLister fields are now atomic.Pointer values so
  the background poller's Store happens-before the VEG worker's Load,
  eliminating the data race flagged in bgp_evpn_conf.go and
  vpc_egress_gateway.go.

* Fail-fast e2e polling. The three wait loops in the new optional_crd
  spec previously masked every error other than NotFo... (continued)

0 of 107 new or added lines in 3 files covered. (0.0%)

14318 of 57287 relevant lines covered (24.99%)

0.29 hits per line