-
Bug
-
Resolution: Unresolved
-
Major
-
None
-
21.0.1.Final
-
None
-
- run cluster of 2 nodes (running singleton deployment if it matters)
- start remote EJB client that makes some ejb calls in a loop
- kill one of the nodes (doesn't matter which one)
-
Undefined
-
---
-
---
When one of the initial cluster nodes becomes unavailable, every 5 seconds one of the remote invocations takes few seconds longer.
It seems that the delay is introduced by DiscoveryEJBClientInterceptor that tries to connect to node that's not available and blocks until connection attempt fails.
From what I've gathered, it looks like an unintended consequence of EJBCLIENT-309, because the slower call happens exactly when failed destination in RemotingEJBDiscoveryProvider expires, and increasing org.jboss.ejb.client.destination-recheck-interval makes slower invocations to happen less frequently, matching destination-recheck-interval value.
