While testing an Accumulo release, I ran into ACCUMULO-3914[1]. In this scenario the dfs client failed to connect to the NN and this error propagated to the hdfs client user. In my experience, the hdfs client is usually really good about transparently retrying. In this case it does not seem to do that, why is that? What the rational underlying this behavior?
[1]: https://issues.apache.org/jira/browse/ACCUMULO-3914
