Flink ML add isolationForest algorithm

I want to use flink solve some problems related to anomaly detection, but currently flink ml lacks algorithms related to anomaly detection, so I want to add the isolation forest algorithm to library/flink ml. During the implementation process, when IterationBody is used, I try to understand the implementation of the Kmeans algorithm, and use iterative behavior to calculate the center point of the isolation forest algorithm, but in the test, I found that when the parallelism > 1, the number of iterations > 1, and there will be sometimes succeed sometimes fail (fail to find the broadcast variable). Please teachers help me to review and point out my problem. Thank you 🙏