How to avoid writing delay if cluster nodes are forcibly shutdown in Apache IoTDB?

I tried to generate high availability test in version 0.13.1 and 0.13.2 of Apache IoTDB that has 3 cluster nodes (3 replicas). I connect these 3 cluster nodes using SessionPool, then perform frequent and large-amount writing operation at regular intervals. This process can perform under normal conditions. I tried force shutdown on one of the nodes and found that basic writing performance became very slow or no writing operation generated for a long time. Writing can be recovered if you recover the node that was forcibly shut down before. I wonder if this is acceptable and if so, how should I avoid writing shutdown if I faced the above scenario in real production?