Description of problem: - The system keeps producing DeletingLoadBalancerFailed error, but the Message is "(events with common reason combined)" and cannot find the root cause. $ oc get events LASTSEEN FIRSTSEEN COUNT NAME KIND SUBOBJECT TYPE REASON SOURCE MESSAGE 1m 40d 3589450 xxx-sample-api-dev Service Normal DeletingLoadBalancer {service-controller } Deleting load balancer 1m 40d 3589441 xxx-sample-api-dev Service Warning DeletingLoadBalancerFailed {service-controller } (events with common reason combined) 1m 37d 3027567 yyy-sample-api-sit Service Normal DeletingLoadBalancer {service-controller } Deleting load balancer 1m 37d 3027558 yyy-sample-api-sit Service Warning DeletingLoadBalancerFailed {service-controller } (events with common reason combined) 1m 42d 4057605 zzz-request-api-v1 Service Normal DeletingLoadBalancer {service-controller } Deleting load balancer 1m 42d 4057596 zzz-request-api-v1 Service Warning DeletingLoadBalancerFailed {service-controller } (events with common reason combined) Version-Release number of selected component (if applicable): OCP 3.3 How reproducible: Steps to Reproduce: 1. We couldn't determine the steps, but it suddenly started and all application produced the DeletingLoadBalancerFailed error. Actual results: The message only says "(events with common reason combined)". Expected results: The error should say some message which allow us to debug this issue. Additional info: - After restarting the master-controllers services, the Warning message has stopped.
Reassigning because the bug seems to be about the event aggregator... and the code in vendor/k8s.io/kubernetes/pkg/client/record/events_cache.go at line 94 talks about prefixing the message... but just returns a constant string without the original message.
The event aggregator intentionall collapses all events with the same reason to the fixed message in order to prevent event spam. It is possible to lose the original source of the event message if the original event prior to aggregation has been purged from the data store. The "Could not determine public IP message" is spam that is being removed here: https://github.com/kubernetes/kubernetes/pull/40763 If the load balancer itself is not being deleted, please open a separate bug, but I am closing this bug as the event compaction is per the intended design.