kubeovn / kube-ovn / 24622570643

25%

Build Type

push

github

Committed by web-flow

Commit Message

perf(controller): use informer indexers for hot-path lookups (#6646)

* perf(controller): use informer indexers for hot-path lookups

Replace full-store linear scans in two hot paths with O(1) indexer
lookups:

- checkAndUpdateNodePortGroup previously iterated every Pod for every
  Node (O(N_node × N_pod)). It now fetches pods by node via a
  byNodeName indexer. The cluster-wide NetworkPolicy existence check
  also switches from a full Lister.List to indexer ListKeys.
- findEndpointSlicesForServices previously listed every EndpointSlice
  in the namespace and filtered by label in a nested loop. It now
  queries a byServiceName indexer keyed on namespace/name.

Introduces pkg/controller/indexers.go as the first custom informer
indexer wiring in the controller package, registered before informer
startup, so future hot paths can reuse the same infrastructure.

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

* perf(controller): address review feedback on indexer wiring

- Use util.LogFatalAndExit for setupIndexers failure to match the
  rest of the Run() startup path.
- Reuse Controller.setupIndexers in the fake controller instead of
  manually calling AddIndexers, so tests exercise the same wiring as
  production and automatically pick up new indexers.

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

* test(controller): add benchmarks comparing indexer lookup vs full scan

TestIndexersResultParityWithFullScan confirms the indexer lookup returns
the same object set as the pre-indexer full scan + filter across every
key, including unscheduled pods and orphan endpointslices.

BenchmarkPodByNode and BenchmarkEPSByService quantify the improvement
at 10k objects:
- PodByNode: ~95x faster (2.5us vs 238us) / ~1... (continued)

Coverage Stats

28 of 50 new or added lines in 4 files covered. (56.0%)

3 existing lines in 2 files now uncovered.

13850 of 56393 relevant lines covered (24.56%)

0.29 hits per line