[ https://issues.redhat.com/browse/WFLY-13514?page=com.atlassian.jira.plugi... ] Matěj Novotný commented on WFLY-13514: -------------------------------------- First of all, huge thanks for this reproducer! It makes it so much easier to discuss things on a common ground. As for the issue, the scope here is definitely going to matter. You cannot get OOM when you use {{@ApplicationScoped}} as you will have only one instance active. Try adding {{@Named}} and {{@ApplicationScoped}} on that bean instead of {{@Model}} and it won't crash. Likewise, when you use {{@SessionScoped}} you will run OOM because for each {{curl{{ you create new session and you do that much faster then those session can timeout (or be anyhow invalidated) hence your post construct code allocating bytes will cause the OOM eventually. So I think we should specifically talk about request scope where the behaviour is really awkward. I can see that for each request, we correctly create a bean instance and I can also see that after each request we attempt to destroy it (if you add {{@PreDestroy}} callback, you can see that it will be invoked). I'm going to look at the heap dump and what keeps the references... -- This message was sent by Atlassian Jira (v7.13.8#713008)