openconfig / kne / 37549155772

48%

push

github

Wait for service ready before considering topology created (#780)

* Wait for service ready before considering topology created

Immediately after all pods are ready, KNE calls service.show() and expects all services to be ready; no retry on error.

For large topologies, it can take a while for metallb to assign IPs across all services. KNE should explicitly wait for service ready condition before treating bring up as finished.

1. **topo.go**:
    * Added checkServiceStatus(ctx, timeout):
        * Iterates over all topology nodes that have services defined.
        * Concurrently polls nod.Services(ctx) until all services for the node are ready.
        * Evaluates LoadBalancer services to ensure len(Status.LoadBalancer.Ingress) > 0 and that an external IP or hostname is populated.
        * Retries with a 500ms backoff until all services are ready or the timeout/context expires.
    * Updated Manager.Create():
        * Calls m.checkServiceStatus(ctx, serviceTimeout) immediately after m.checkNodeStatus(ctx, timeout) completes, preventing Create()
        from returning before LoadBalancer services have their external IPs assigned.

2. **topo_test.go**:
    * Added a fake reactor on create services in TestCreate to simulate load balancer ingress assignment.
    * Added a test case success with hanging service + timeout to verify timeout behavior when services fail to get an IP.

All unit tests in ~/src/kne/topo/... are passing.

* Address review comments for service ready check

- Add floor and default timeout for checkServiceStatus when timeout == 0 or elapsed >= timeout
- Log nodes pending service ready status on timeout
- Add nil check in isServiceReady to avoid panic on unpopulated services
- Add services reactor to TestCreateSonicNode
- Add unit tests for isServiceReady and calculateServiceTimeout

90 of 107 new or added lines in 1 file covered. (84.11%)

6983 of 14786 relevant lines covered (47.23%)

0.54 hits per line