You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@activemq.apache.org by "Tom (JIRA)" <ji...@apache.org> on 2009/12/03 05:31:54 UTC
[jira] Updated: (AMQ-2520) Oracle 10g RAC resource usage VERY high
from the passive servers SQL requests to the Database.
[ https://issues.apache.org/activemq/browse/AMQ-2520?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Tom updated AMQ-2520:
---------------------
Description:
Two active MQ brokers are installed on RH EL 5 servers (one per server).
They're configured as a JDBC master / slave failover (as per examples). Failover is tested and working and messages delivered.
Oracle is used for synchronisation (ACTIVEMQ_ tables), persistence etc.
We run a durable subscriber, and the client connects via a failover operation.
The SELECT * FROM ACTIVEMQ_LOCK FOR UPDATE is causing spin lock on the Oracle database.
Basically the indefinite waiting from the passive mq instance is causing high resource usage on Oracle.
After a short period Oracle dashboard shows a high number of active sessions from Active MQ due to the continuous execution of
UPDATE ACTIVEMQ_LOCK SET TIME = ? WHERE ID = 1
in the keepAlive method in
https://svn.apache.org/repos/asf/activemq/trunk/activemq-core/src/main/java/org/apache/activemq/store/jdbc/DatabaseLocker.java
As a workaround we've had to push out the lockAcquireSleepInterval to 5 minutes in the configuration of ActiveMQ, but this didn't work.
<jdbcPersistenceAdapter dataSource="#oracle-ds" useDatabaseLock="true" lockAcquireSleepInterval="300000" createTablesOnStartup="true"/>
We're currently changing the broker to poll rather than block so in Statement.java we've added a WAIT 0 that throws an exception if the lock is not acquired.
public String getLockCreateStatement() {
if (lockCreateStatement == null) {
lockCreateStatement = "SELECT * FROM " + getFullLockTableName();
if (useLockCreateWhereClause) {
lockCreateStatement += " WHERE ID = 1";
}
lockCreateStatement += " FOR UPDATE WAIT 0";
}
return lockCreateStatement;
}
Any suggestions to this issue, this seems to be a quite fundamental issue?
was:
Two active MQ brokers are installed on RH EL 5 servers (one per server).
They're configured as a JDBC master / slave failover (as per examples). Failover is tested and working and messages delivered.
Oracle is used for synchronisation (ACTIVEMQ_ tables), persistence etc.
We run a durable subscriber, and the client connects via a failover operation.
The SELECT * FROM ACTIVEMQ_LOCK FOR UPDATE is causing spin lock on the Oracle database.
Basically the indefinite waiting from the passive mq instance is causing high resource usage on Oracle.
After a short period Oracle dashboard shows a high number of active sessions from Active MQ due to the continuous execution of
UPDATE ACTIVEMQ_LOCK SET TIME = ? WHERE ID = 1
in the keepAlive method in
https://svn.apache.org/repos/asf/activemq/trunk/activemq-core/src/main/java/org/apache/activemq/store/jdbc/DatabaseLocker.java
As a workaround we've had to push out the lockAcquireSleepInterval to 5 minutes in the configuration of ActiveMQ, but this didn't work.
<jdbcPersistenceAdapter dataSource="#oracle-ds" useDatabaseLock="true" lockAcquireSleepInterval="300000" createTablesOnStartup="true"/>
Any suggestions to this issue, this seems to be a quite fundamental issue?
Summary: Oracle 10g RAC resource usage VERY high from the passive servers SQL requests to the Database. (was: Oracle 10g RAC resource usage VERY high from the passive servers keepAlive requests to the Database.)
> Oracle 10g RAC resource usage VERY high from the passive servers SQL requests to the Database.
> ----------------------------------------------------------------------------------------------
>
> Key: AMQ-2520
> URL: https://issues.apache.org/activemq/browse/AMQ-2520
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.3.0
> Environment: Redhat Enterprise Linux 5, Oracle 10g RAC
> Reporter: Tom
> Priority: Blocker
>
> Two active MQ brokers are installed on RH EL 5 servers (one per server).
> They're configured as a JDBC master / slave failover (as per examples). Failover is tested and working and messages delivered.
> Oracle is used for synchronisation (ACTIVEMQ_ tables), persistence etc.
> We run a durable subscriber, and the client connects via a failover operation.
> The SELECT * FROM ACTIVEMQ_LOCK FOR UPDATE is causing spin lock on the Oracle database.
> Basically the indefinite waiting from the passive mq instance is causing high resource usage on Oracle.
> After a short period Oracle dashboard shows a high number of active sessions from Active MQ due to the continuous execution of
> UPDATE ACTIVEMQ_LOCK SET TIME = ? WHERE ID = 1
> in the keepAlive method in
> https://svn.apache.org/repos/asf/activemq/trunk/activemq-core/src/main/java/org/apache/activemq/store/jdbc/DatabaseLocker.java
> As a workaround we've had to push out the lockAcquireSleepInterval to 5 minutes in the configuration of ActiveMQ, but this didn't work.
> <jdbcPersistenceAdapter dataSource="#oracle-ds" useDatabaseLock="true" lockAcquireSleepInterval="300000" createTablesOnStartup="true"/>
> We're currently changing the broker to poll rather than block so in Statement.java we've added a WAIT 0 that throws an exception if the lock is not acquired.
> public String getLockCreateStatement() {
> if (lockCreateStatement == null) {
> lockCreateStatement = "SELECT * FROM " + getFullLockTableName();
> if (useLockCreateWhereClause) {
> lockCreateStatement += " WHERE ID = 1";
> }
> lockCreateStatement += " FOR UPDATE WAIT 0";
> }
> return lockCreateStatement;
> }
> Any suggestions to this issue, this seems to be a quite fundamental issue?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.