[ https://issues.redhat.com/browse/JBTM-3325?page=com.atlassian.jira.plugin... ] Luis Barreiro commented on JBTM-3325: ------------------------------------- Thanks [~ochaloup] for bringing up this topic. Agroal takes a different approach than IronJacamar on recovery. It was debated internally, and the conclusion was that it's better to open a new connection every time than to have to manage the connection. To prevent leaks in this scenario, there must be a way to notify that recovery scan has ended. Ideally, this would be done in either 1) the {{XAResource}} itself, or 2) {{XAResourceRecovery}} as a counterpart of {{getXAResources()}}. The former is impossible to change, for obvious reasons, and I don't see any will to make changes to the latter. I asked for feedback on how to manage the connection liyfecycle and [~mmusgrov] suggested to close the connection on {{TMENDRSCAN}}. Agroal now checks for it and it fixed the connection leaks that were being reported. I then suggest to go beyond that and have the {{XAResource}} implementation to express it's intention to be notified that it's no longer needed by implementing a marker interface, in this case I suggested {{AutoClosable}}. I believe this would be the cleanest way to deal with this problem, given the circumstances. This would be an alternative to 1) above. If [~jhalliday] it's right then it's really concerning, because not only it means that the current implementation is wrong and there is no reliable way to get notified of the end of the recovery, but it also implies that Narayana does not have control on the recovery procedure which baffles me. Go back to keep a connection open for recovery is always an option, but it comes with it's own drawbacks and implementation challenges. I much rather see the effort put into having a reliable way to notify the end of recovery. -- This message was sent by Atlassian Jira (v7.13.8#713008)