this is an interesting tutorial on how to run OpenTelemetry at scale. When ingestion volume becomes super high, a queue like Kafka seems to be important to make otel collector handle spikes, etc.

I wish I'd included it in the piece, I'll maybe write a follow up: right now I don't think it makes sense to implement queueing just for your observability data. But if the queue is already there, it's worth considering handling OTLP data with a Kafka queue.