While troubleshooting an issue with Marathon consuming massive CPU resources, we determined it was falling behind processing the event stream. While analyzing the event stream, we observed that the deployment_success message was 1.8MB in size and appeared to represent the root of the entire Marathon system. We logged 88 such messages in a 10 minute time period when no deployments were occurring, to our knowledge, and I suspect Marathon's high CPU utilization is due to trying to parse these messages.
For reference, here is the beginning of one such message, annotated:
edited to add: deployment_info events also seem to have the highest node as the payload.
"id": "/", <<< this right here just seems unnecessary and excessive.