Intermmitent freeze in tests while deleting environment

[rimolive]

The test was running as expected (see[1]), but maven didn't finish the test. After capturing Thread dump of surefire process, it seems that some threads started an infinite recursion (see [2]).

[1] https://gist.github.com/rimolive/56c0153cae0dc41120411a335cd6f2a8
[2] https://gist.github.com/rimolive/b888e2f8af87bb9b95fd87a2e7bfd587

[jwendell]

Could you post the logs here, or in a gist?
This pastebin is not available for everyone and it might expire.

[rimolive]

@jwendell, you are correct. I'll post in a different place

[rimolive]

I noticed that the test freezes after some errors in Remoting protocol. Details of the error in [1].

[1] https://gist.github.com/rimolive/85df12f2e4d2760a1ba99f10e3ec9dd6

[jwendell]

@rimolive any update on this? I was caught by this issue as well...

[rcernich]

It doesn't look like there's infinite recursion. If there were, you'd get a stack overflow exception at some point. Also, the top of the stack breaks you out of the "loop" and appears to be related to reading the error stream. When I initially wrote the port forwarder, I had issues with the termination of the error streams and this may be the same problem. I don't think this should have an effect any more, since we're invoking the port forwarder asynchronously. The only side effect would be leaking connections.

[jwendell]

After a big investigation, found out the issue belongs to undertow.

Filled https://issues.jboss.org/browse/UNDERTOW-977 which was quickly fixed, because this is a known issue that had already been fixed on master, but not on 1.3.x branch, which is the one we use.

I've applied the patch locally and indeed it seems to work.

So, I'll ask UT maintainers for a 1.3.27.Final release so that we can benefit from it.