Dear folks, I have a terribly weird issue with EAP 7.0.0.ER5 on platforms with high numbers of CPUs. I suspect Undertow web subsystem from doing something fishy with sessions, i.e. jsession ids.

• both with jsessionid in URL and in COOKIE, one isn't able to deterministically get the same session data
• sometimes client receives the propper session - i.e. no new session is created, sometimes a new session is created
• both Apache HTTP Server mod_cluster and Undertow mod_cluster proxy fails at keeping sticky sessions, despite the fact that the mod_cluster worker subsystem correctly sends an MCMP mesage stating that sticky sessions are true, sticky sessions force is false
• I was unable to reproduce it on any regular machines, only powerful 18+ cores machines show this kind of race condifiton
• alongside sticky sessions, as a side effect, fail-over is also broken – this was reproduced only with Undertow mod_cluster proxy though, i.e. it sent a request deliberately to a dead node, getting HTTP 503

I recorded one session on HP-UX with Undertow mod_cluster balacncer. The undermentioned files contain curl log from reproducing the issue, test app and configs:

• curl.log
• undertow_balancer_debug.log.zip
• undertow_balancer_standalone.xml
• worker-jboss-eap-7.0-1.log
• worker-jboss-eap-7.0-1-standalone-ha.xml
• worker-jboss-eap-7.0-2.log
• worker-jboss-eap-7.0-2-standalone-ha.xml
• clusterbench.war

Any ideas?