[ 
https://issues.apache.org/jira/browse/TRAFODION-2046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15368057#comment-15368057
 ] 

Sean Broeder commented on TRAFODION-2046:
-----------------------------------------

This change has added a new MemoryUsageException that is not handled correctly 
and is causing test 130 in the daily builds to fail.  As a short term fix I am 
submitting a pull request to back out these changes until we can better and 
more predictably handle these exceptions, otherwise transactional DDL can leave 
a sustem in an unuseable state.

>>drop table tlob130ts1;
 
*** ERROR[8448] Unable to access Hbase interface. Call to 
ExpHbaseInterface::getRowOpen returned error HBASE_OPEN_ERROR(-704). Cause:
java.io.IOException: 
org.apache.hadoop.hbase.client.transactional.MemoryUsageException: get memory 
usage exceeds 90 percent, trxId is 101365
org.apache.hadoop.hbase.client.transactional.TransactionalTable.get(TransactionalTable.java:211)
org.apache.hadoop.hbase.client.transactional.RMInterface.get(RMInterface.java:278)
org.trafodion.sql.HTableClient.startGet(HTableClient.java:1049)
org.trafodion.sql.HBaseClient.startGet(HBaseClient.java:1596)
.
 
--- SQL operation failed with errors.
 

> change DTM's memoryUsageChore default settings to enable memory threthold
> -------------------------------------------------------------------------
>
>                 Key: TRAFODION-2046
>                 URL: https://issues.apache.org/jira/browse/TRAFODION-2046
>             Project: Apache Trafodion
>          Issue Type: Improvement
>            Reporter: liu ming
>            Assignee: liu ming
>
> DTM provide MemoryUsageChore to monitor the memory pressure in a 
> RegionServer, when memory is beyond a defined threshold, it will give a 
> warning or prevent new operations based on a flag.
> By default, the threshold is now 100%, and the flag is just warning. So that 
> means DTM coprocessor can use all RS memory and when it find memory pressure 
> , just print warning, instead of control new transaction.
> This will lead to un-predictable behavior for certain fat transaction.
> It is better to change the default setting to enable protection.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to