Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/5449
We cannot fully guard against incorrect network configurations.
Many systems (even HDFS,, AFAIK) simply use `InetAddress.getLocalHost()`
and rely on correct network configs. Our current strategy already tries to be
smarter by trying to fond a connecting address, and only fall back to
`getLocalHost()`.
My feeling is that the strategy is fine (I have not seen connectivity
issues in a while) and focus on stabilizing the tests.
- We could reducing the heuristic check to "returns something we can bind
to"
- We could add an assumption to the test that `getLocalHost()` is not a
loopback address and something we can bind to, and then run the tests.
What do you think?
---
