Why do NIOSequentialFile and NIOSequentialFileFactory use ByteBuffer.allocate() instead of ByteBuffer.allocateDirect()?

[rdelfino]

Hi;

I've been using HornetQ embedded for a while in a software solution we developed.

This week we were bitten hardly by an OutOfMemory error that occured on NIOSequentialFile

when HornetQ tried to create a new journal file. After that error the application could

no more send messages to the queues and all hell broke loose.

Browsing HornetQ source code to understand the error I would like to know why

NIOSequentialFile and NIOSequentialFileFactory use ByteBuffer.allocate() instead

of ByteBuffer.allocateDirect()?

If ByteBuffer.allocateDirect() was used instead the journal would not compete with

the rest of the application code for JVM heap space and this error could be avoided.

I know we should be monitoring memory usage and also that we should have tunned

Xmx parameters, and so on, however I couldn't help ask you guys.

Thanks in advance

Reginaldo

[jbertram]

I can't speak for Clebert, but I believe allocate() is used because allocateDirect() is slower in general and it is assumed that users will have their memory configured in such a way as to avoid OutOfMemoryErrors.

[clebert.suconic]

You should have used the user's forum instead.. I will move this message.

allocateDirect is only used when required. There are users with allocateDirect exactly for what you said. GC won't take care of these buffers as they are not holding any reference, and you may get worse OMEs if we were doing this way. Another reason is performance.. allocateDirect is really slow.