[ https://issues.jboss.org/browse/WFLY-10531?page=com.atlassian.jira.plugin... ] Jiri Ondrusek edited comment on WFLY-10531 at 10/4/18 8:41 AM: --------------------------------------------------------------- [~jwgmeligmeyling] I'm still trying to figure, what exactly is causing this error. Currently, I've created better test: https://github.com/JiriOndrusek/wildfly/commit/fd84ff61e074ade65d4b9f5a21... (which doesn't exhaust server) My current thoughts about this issue are following: * I have logs from two executions with small amount (the same numbers) of calls of injected contexts with leaking connections and without leaking connections -> error doesn't depend on precise number of calls from reproducer * if I set count of executions to higher values, there is an error with almost 100% certainty * I was able to count registered contexts in transaction synchronization on multiple executions, and then number of closed ones -> I see difference which I can not explain for now I have suspicion, that that problem is caused near registering of synchronization listener. I think that it is possible, that some contexts are not registered for closing at the end of transaction. It answers the questions: - why it is happening only with more injected contexts - and the fact that I see that not all contexts, which should be registered, are not closed by synchronization listener I'll investigate this scenario, to be sure. If my suspicion is correct, then proposed fix works only as workaround. was (Author: jondruse): [~jwgmeligmeyling] I'm still trying to figure, what exactly is causing this error. Currently, I've created better test: https://github.com/JiriOndrusek/wildfly/commit/fd84ff61e074ade65d4b9f5a21... (which doesn't exhaust server) My current thoughts about this issue are following: * I have logs from two executions with small amount (the same numbers) of calls of injected contexts with leaking connections and without leaking connections -> error doesn't depend on precise number of calls from reproducer * if I set count of executions to higher values, there is an error with almost 100% certainty * I was able to count registered contexts in transaction synchronization on multiple executions, and then number of closed ones -> I see difference which I can not explain for now I have suspicion, that that problem is caused near registering of synchronization listener. I think that it is possible, that some contexts are not registered for closing at the end of transaction. It answers the questions: - why it is happening only with more injected contexts - and the fact that I see that not all contexts, which should be registered, are not closed by synchronization listener I'll investigate this scenario, to be sure. -- This message was sent by Atlassian Jira (v7.12.1#712002)