We are using flink1.4.2 in our company. When the flink job is submitted to run on yarn with yarn-cluster mode，we actually find that the flink-dist_2.11-1.4.2.jar is uploaded to HDFS. The jars lies in different directories e.g.,
Through reviewing source code of flink and having some tests, we suppose that the code below may have a bug in the Linux environment.
!(file.getFileName().startsWith("flink-dist") &&file.getFileName().endsWith("jar")) is Always be true