For OOM failures, Ironclad will try to free internal storage used for cache and retry allocation until it is deemed the situation is not recoverable, in which case it will panic. To this end, Ironclad will kill high memory consuming processes, this behaviour is configurable in failure_policy.