More reliable K8s Event logging

The Fluent Bit (FLB) component that watches Kubernetes events and ships them to Loki has been replaced by Grafana Alloy. The main reason for this move is that we were noticing quite some performance/resource issues with the FLB-based component across several clusters, especially larger ones that also have many scaling-related events. Alloy provides us a simpler and more reliable experience for this use-case.

If you’ve been following, you’ll also notice this is not the first time we’ve needed to change our pipeline for shipping K8s Events. So far we’ve used and migrated from Eventrouter (deprecated) –> grafana-agent (deprecated) –> fluent bit –> alloy.

We’re keeping changes backwards-compatible, so your existing queries keep working. The {job="events"} selector is preserved, and the namespace_name and reportingComponent labels are emitted with the same values as before. Other event fields (kind, name, reason, type, apiVersion) are now exposed as Loki structured metadata, which keeps them filterable without inflating the label index.

Pod-log shipping continues to run on a Fluent Bit DaemonSet — only the events pipeline moved to Alloy.