[ https://issues.jboss.org/browse/JGRP-1675?page=com.atlassian.jira.plugin.... ] Radim Vansa edited comment on JGRP-1675 at 9/25/13 4:34 AM: ------------------------------------------------------------ Ah, stupid me. I've finally realized why there may be so many requests that the OOB threadpool gets depleted. Inifinispan does GETs to both owners, waiting just for the first response. Therefore, when the request get stuck because of credits on one node, the other owner serves it and the requesting node may issue another, blocking more and more OOB threads in the node lacking credits. I should have got this way before. was (Author: rvansa): Ah, stupid me. I've finally realized why there may be so many requests that the OOB threadpool gets depleted. Inifinispan does GETs to both owners, waiting just for the first response. Therefore, when the request get stuck because of credits on one node, the other owner serves it and the requesting node may issue another, blocking more and more OOB threads in the node lacking credits. I should have got this way before. So, when we run out of credits, OOB TP gets depleted and no more messages get there. Neither the credit replenishments, because several OOB packets get lost and even when the request is INTERNAL, it's ordered in relation to the OOB messages and as noone can receive these (they're dropped when we try to pass them to OOB threads), it just waits in the UNICAST tables. The question is how we've run out of credits in the beginning, but I'd say that even situation when this happens should be handled. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira