The document summarizes an incident where a Cassandra cluster experienced performance issues and nodes going down due to high load. It describes how the event loop threads became blocked waiting on the inbound request queue, causing high latency and client retries. The solution presented applies backpressure when the inbound queue becomes full by either returning an overloaded exception or applying throttling depending on a new startup option. Performance tests showed that with this backpressure patch applied, the cluster maintained higher throughput and lower latencies under heavy traffic loads compared to a cluster without the patch. Going forward, the backpressure mechanism could be improved to use more dynamic outlier detection rather than fixed thresholds. Applying the patch to production then prevented further blocked threads and out of memory errors under heavy loads