mitrevf / airhead-research

Automatically exported from code.google.com/p/airhead-research
0 stars 0 forks source link

Open files in OnDiskSemanticSpace can't be closed #97

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago
The files opened by OnDiskSemanticSpace aren't closed until the JVM exits. On 
Windows this leads to locks on the files so they can't be removed while the 
java process is alive.
The SemanticSpace-Interface should be extended with a close- or destroy-method 
in which OnDiskSemanticSpace does close the files.

Original issue reported on code.google.com by keepal...@gmail.com on 28 Jun 2011 at 12:08

GoogleCodeExporter commented 9 years ago
Which files are you specifically referring to? In some cases, the backing 
sspace file has to be maintained throughout the lifetime of the process because 
we do random access on the file when the whole space can't fit into ram.  Is 
there a particular reason you want to delete the backing sspace while the 
sspace is running?

Original comment by FozzietheBeat@gmail.com on 20 Sep 2011 at 5:36

GoogleCodeExporter commented 9 years ago
Actually you are pointing to the problem. Here we have a long time running 
process (ApplicationServer) which creates multiple large temporary sspace 
(~1GB) during runtime.
The sspace doesn't fit into ram so we use SPARSE_BINARY format and 
OnDiskSemanticSpace. In loadOffsetsFromFormat() the RandomAccessFile is opened, 
stored in binarySSpace and never closed. Because a native file handle is 
attached to RandomAccessFile the sspace-file is locked on Windows and can't be 
deleted. So we have a lot of large dead temp files lying around which can't be 
deleted for a long time.

Original comment by keepal...@gmail.com on 21 Sep 2011 at 9:11

GoogleCodeExporter commented 9 years ago
Is it that the files are never unlocked when the program finishes, and 
therefore the file can't be deleted until Windows cleans up the file locks?

Is it possible to tell us more about your use case?  Typically, we've left the 
files in place assuming that the user might access them at any point.  We could 
potentially close those files and then re-open as necessary, but that incurs 
some overhead.

Original comment by David.Ju...@gmail.com on 21 Sep 2011 at 9:49

GoogleCodeExporter commented 9 years ago
The files are unlocked when the program is finished. Then they can be removed 
manually. They can't be removed by the application without a restart.

Our use case is to create one huge semantic space from time to time which is 
used only once.

Closing and reopening the files as necessary would slow the usage down. I 
thought of a way to close or destroy a SemanticSpace, so resources of the 
SemanticSpace (eg. open files) can be released. In OnDiskSemanticSpace it would 
look like this:

@Override
public void close() {
    if (binarySSpace != null) {
        binarySSpace.close();
    }
}

Original comment by keepal...@gmail.com on 22 Sep 2011 at 3:20

GoogleCodeExporter commented 9 years ago
I think this would be a pretty weird addition to the SemanticSpace interface, 
but we could definitely add the close method to specific models that need to 
close files.  If your progam is just using OnDiskSemanticSpace in this manner, 
then you could forfit the use of the interface and just keep these object 
declared as OnDiskSemanticSpace.  ALternatively, we could overload the 
finalize() method and code it carefully so that files are only ever closed 
once.  This would let you always do 

// Close out any files in the SemanticSpace
sspace.finalize();

The only issue with finalize is that it greatly complicates the garbage 
collection process.

I find it kind of frustrating that java doesn't just automatically close the 
file handler when a File object gets garbage collected.

Original comment by FozzietheBeat@gmail.com on 23 Sep 2011 at 5:29

GoogleCodeExporter commented 9 years ago
I wrote a small Test which shows that closing the RandomAccessFile in 
finalize() would work (see attachment). If you remove the "System.gc()" the 
file can't be deleted. I would prefer an additional close()-method in 
OnDiskSemanticSpace though, because then you don't need to wait for the garbage 
collection to be able to remove the file.

Original comment by keepal...@gmail.com on 26 Sep 2011 at 2:55

Attachments:

GoogleCodeExporter commented 9 years ago
The close method seems like the best call for cleaning up the resources when 
you're done with the OnDiskSemanticSpace.  It would actually be nice to use the 
new AutoCloseable interface from JSE7 which would let you do something like

try (OnDiskSemanticSpace sspace = new OnDiskSemanticSpace(...) {
    // do things there  
}

and have it close the backing RandomAccessFile in the event of an exception, or 
when you're finished with the sspace itself.

I'm not sure if we're ready to move to JSE7, but certainly putting the close() 
method in seems like the best choice here.

Original comment by David.Ju...@gmail.com on 26 Sep 2011 at 5:20