Bela Ban created JGRP-1815: ------------------------------ Summary: TP: sending a message to a non-existent physical address takes too much time Key: JGRP-1815 URL: https://issues.jboss.org/browse/JGRP-1815 Project: JGroups Issue Type: Enhancement Reporter: Bela Ban Assignee: Bela Ban Fix For: 3.5 When sending a message in the transport, e.g. a message batch, and one or more destinations have no physical addresses in {{logical_addr_cache}}, then we loop {{physical_addr_max_fetch_attempts}} (default: 3) times and also sleep a random number of ms (in range [1..500]). This delays the sending of other messages in the same batch, or of other messages in general. Also, if TransferQueueBundler is used, the transfer queue might get filled, so other sender threads are blocked. SOLUTION: * Remove the loop when sending a message: if the physical address for a message is not found, simply send a discovery request and drop the message * The discovery request needs to be sent on a separate thread, as calling {{up(Event)}} directly could also delay the sending of the message or message batch. * JGRP-1814 will also help, as connections to left members are closed, and therefore not finding a physical address for a destination should be rare -- 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