Consider a scenario where a Spark streaming job is failing to meet its latency SLA. Upon inspection, you find that the Spark UI shows a high number of late batches. What steps would you take to address this issue?