I have just migrated an application from JBoss 7.1.1 to Wildfly 8.1.0 Final. But now from time to time (like every 2 days) Wildfly freezes, and I have to restart it. My application is basically EJB/JPA/JAX-RS w/ Resteasy.
By freezing I mean that every HTTP access to Wildfly keeps waiting forever. I suspect that only Undertow is frozen, because I can see in my application log that tasks that were scheduled are executing, although none of these are network related.
I also noticed in my log the following error: "Response is committed, can't handle exception" caused by IOException "Channel was closed mid chunk, if you have attempted to write chunked data you cannot shutdown the channel until after it has all been written".
Server is an Ubuntu VM running on Azure. "Physical memory" and "hard disk" space are fine. Heap, non-heap and thread usages are fine according to JVM Runtime in Wildfly Management.
Only thing I noticed is that I have many connections established forever when I "netstat -aln". I have fewer than 100 clients per day but right now I count 2K established connections on my port 8080.
What can I do to better diagnose and resolve this problem?
Best Regards,
Renan
UPDATE: it was indeed a bug on 8.1.0 fixed for 8.2.0, see my answer with link to issue opened at JbossDeveloper.
Best Answer
I would like to share my experience as a developer/architect on Wildfly 8.1
I migrate from 7.1.1 to 8.1 on Windows/Eclipse a few day ago. I also noticed some instabilities; a member of the developement team did the migration, and complains about eclipse crashes / not the same behaviour as 7.1.1. An other team member is still working on Wildfly 8.0; it seems that previous version is more stable.
JBOss would tell you that Wildfly is not a production app server; it is for development purpose. But I think that you are pointing to some serious bug / converging with my sucpicions / without any real facts, and no answer to your question.
You should forward your post directly to the JBoss forums in order to be shure that the Wildfly team investigate the problem and hope that the the fix will be part of 8.2 releases.