Third and last attempt, Apache spam filter seems to hate me! Hi,
I've remember a reported issue on the mailing list mentioning the funky interval you describe but it had no replies. I've done several set ups with replication of which one is a very high load service with a pollInterval of 2 seconds. The other set ups have a much higher interval. I've never seen this behaviour before in any set up. Is there something else going on? Can you reproduce this weird behaviour with the same index, software versions etc in a development environment? About the replication.properties file's number of failed replication; i might not remember correctly but this value, i think, is incremented when a replication fails. A replication can fail when the slave is trying to download a (large) list of large files when, in the meantime, the master merges some segments. This specific issue can be remedied using the commitReserveDuration replication property. However, if this occurs there should be an exception in your log.