Bug #39611 ReplicationDriver not selecting last host specified in JDBC config

Description:
When configuring the Java Replication Driver we find that the last slave specified is never used.

We're using com.mysql.jdbc.ReplicationDriver in our datasource configuration and the following datasource URL:

jdbc:mysql://masterHost:3306,slaveHost1:3306,slaveHost2:3306/database?autoReconnect=true&useServerPrepStmts=false&cachePrepStmts=true&jdbcCompliantTruncation=false&useUnicode=true&roundRobinLoadBalance=true

How to repeat:
Set up a Master and 1 Slave.  Perform some reads.  The slave is never used.

Set up 1 Master and 2 Slaves. Perform some reads.  The second slave is never used.

Suggested fix:
The key issue was with the way Java will effectively Math.floor the random number when converting to the int value.  Therefore, index can never equal indexRange.

Here's the diff I applied to com.mysql.jdbc.ConnectionImpl.java within the getNextRoundRobinHostIndex() method.

315c315
< 		int index = (int)(Math.random() * indexRange);
---
> 		int index = (int)(Math.random() * (indexRange+1));

This maintains an even distribution selection across the slaves 0..n

patch

Attachment: patch (application/octet-stream, text), 142 bytes.

Hi and thanks for your report.

Verified as described by looking into code.

This was resolved in 5.1.7 by patch #739:

-		int indexRange = hostList.size() - 1;
+		int indexRange = hostList.size();

Comment from merge:

Fixed BUG#39611 - ReplicationConnection never sends queries to last host in slave list.

Workaround is to list the final slave twice in pre-5.1.7 versions.

An entry was added to the 5.1.7 changelog:

When configuring the Java Replication Driver the last slave specified was never used.

An entry was added to the 5.1.7 changelog:

When configuring the Java Replication Driver the last slave specified was never used.