cuijianwei created HBASE-12797:
----------------------------------
Summary: ReplicationSyncUp will return before recovered queue
transfered with default configuration
Key: HBASE-12797
URL: https://issues.apache.org/jira/browse/HBASE-12797
Project: HBase
Issue Type: Improvement
Components: Replication
Affects Versions: 0.99.2
Reporter: cuijianwei
Priority: Minor
ReplicationSyncUp will check all old sources has been removed every 10 seconds
(ReplicationSyncUp.SLEEP_TIME):
{code}
int numberOfOldSource = 1; // default wait once
while (numberOfOldSource > 0) {
Thread.sleep(SLEEP_TIME);
numberOfOldSource = manager.getOldSources().size();
}
{code}
However, the default "replication.sleep.before.failover" is 30 seconds, which
will make NodeFailoverWorker sleep at least 30 seconds before transferring
recover queue. Therefore, ReplicationSyncUp will do nothing and return directly
with default configuration.
When using ReplicationSyncUp, the goal tends to be syncing-up the
not-replicated data to peer cluster as soon as possible, is it reasonable to
set "replication.sleep.before.failover" to 0 in this situation, and also make
sure ReplicationQueues.getListOfReplicators().size() = 0 before returned?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
