Pattern 5 - Multi-cluster
In advanced use cases, you might prefer to have a central “control plane” cluster receiving the observability data from all your clusters. In general, we advise to send your telemetry data directly from the individual clusters to your telemetry backend instead, in order to avoid excessive network costs. However, you might have telemetry data that need to be processed together, such as when doing tail-sampling for traces spanning multiple clusters. The configuration for this setup is very similar for the sidecar -> collector use case, with the caveat that TLS is even more important here, given that data could potentially flow through the public internet.