[
https://issues.apache.org/jira/browse/HBASE-423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12610595#action_12610595
]
Jean-Daniel Cryans commented on HBASE-423:
------------------------------------------
Isn't the call on HConnectionManager.deleteConnectionInfo() in the tearDown of
HBaseClusterTestCase enough? I tried adding a boolean called acceptingRequests
in HCM that is checked like this :
{code}
public static HConnection getConnection(HBaseConfiguration conf) {
TableServers connection = null;
if(acceptingRequests) {
synchronized (HBASE_INSTANCES) {
String instanceName = conf.get(HBASE_DIR);
connection = HBASE_INSTANCES.get(instanceName);
if (connection == null) {
connection = new TableServers(conf);
HBASE_INSTANCES.put(instanceName, connection);
}
}
}
return connection;
}
{code}
and I think I saw that each tests took 1 second less.
> Test cases need a mechanism to tell HConnectionManager to shutdown
> ------------------------------------------------------------------
>
> Key: HBASE-423
> URL: https://issues.apache.org/jira/browse/HBASE-423
> Project: Hadoop HBase
> Issue Type: Test
> Components: client
> Reporter: Jim Kellerman
> Assignee: Jean-Daniel Cryans
>
> One of the reason many test cases take so long is that HConnectionManager
> continues trying to refind tables, meta and root region even when the test is
> done but has not completely shut down yet.
> The test case should be able to call the connection manager and tell it to
> shutdown, meaning:
> - release all references (clear out all its maps)
> - do not accept any new requests
> - if any requests are in progress, abort them.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
