Performance tuning of a Hibernate+Spring+MySQL project operation that stores images uploaded by user

Question

Hi

I am working on a web project that is Spring+Hibernate+MySQL based. I am stuck at a point where I have to store images uploaded by a user into the database. Although I have written some code that works well for now, but I believe that things will mess up when the project would go live.

Here's my domain class that carries the image bytes:

@Entity
public class Picture implements java.io.Serializable{
    long id;
    byte[] data;
    ... // getters and setters
}

And here's my controller that saves the file on submit:

public class PictureUploadFormController extends AbstractBaseFormController{
    ...
    protected ModelAndView onSubmit(HttpServletRequest request, HttpServletResponse response, Object command, BindException errors) throws Exception{
         MutlipartFile file; 
         // getting MultipartFile from the command object
         ...
         // beginning hibernate transaction
         ...
         Picture p=new Picture();
         p.setData(file.getBytes());
         pictureDAO.makePersistent(p); // this method simply calls getSession().saveOrUpdate(p)

         // committing hiernate transaction
         ...
    }
    ...
}

Obviously a bad piece of code. Is there anyway I could use InputStream or Blob to save the data, instead of first loading all the bytes from the user into the memory and then pushing them into the database?

I did some research on hibernate's support for Blob, and found this in Hibernate In Action book:

java.sql.Blob and java.sql.Clob are the most efficient way to handle large objects in Java. Unfortunately, an instance of Blob or Clob is only useable until the JDBC transaction completes. So if your persistent class defines a property of java.sql.Clob or java.sql.Blob (not a good idea anyway), you’ll be restricted in how instances of the class may be used. In particular, you won’t be able to use instances of that class as detached objects. Furthermore, many JDBC drivers don’t feature working support for java.sql.Blob and java.sql.Clob. Therefore, it makes more sense to map large objects using the binary or text mapping type, assuming retrieval of the entire large object into memory isn’t a performance killer.

Note you can find up-to-date design patterns and tips for large object usage on the Hibernate website, with tricks for particular platforms.

Now apparently the Blob cannot be used, as it is not a good idea anyway, what else could be used to improve the performance? I couldn't find any up-to-date design pattern or any useful information on Hibernate website. So any help/recommendations from stackoverflowers will be much appreciated.

Thanks

Answer 1

The revised edition of the book (Java Persistence with Hibernate) says:

Table 5.3 lists Hibernate types for handling binary data and large values. Note that only binary is supported as the type of an identifier property.

Mapping type   Java type   Standard SQL built-in type
binary         byte[]               VARBINARY
text           java.lang.String     CLOB
clob           java.sql.Clob        CLOB
blob           java.sql.Blob        BLOB
serializable   Any Java class that  VARBINARY
               implements
               java.io.Serializable

If a property in your persistent Java class is of type byte[], Hibernate can map it to a VARBINARY column with the binary mapping type. (Note that the real SQL type depends on the dialect; for example, in PostgreSQL, the SQL type is BYTEA, and in Oracle it’s RAW.) If a property in your persistent Java class is of type java.lang.String, Hibernate can map it to an SQL CLOB column, with the text mapping type.

Note that in both cases, Hibernate initializes the property value right away, when the entity instance that holds the property variable is loaded. This is inconvenient when you have to deal with potentially large values.

One solution is lazy loading through interception of field access, on demand. However, this approach requires bytecode instrumentation of your persistent classes for the injection of extra code. We’ll discuss lazy loading through bytecode instrumentation and interception in chapter 13, section 13.1.6, “Lazy loading with interception.”

A second solution is a different kind of property in your Java class. JDBC supports locator objects (LOBs) directly.1 If your Java property is of type java.sql.Clob or java.sql.Blob, you can map it with the clob or blob mapping type to get lazy loading of large values without bytecode instrumentation. When the owner of the property is loaded, the property value is a locator object—effectively, a pointer to the real value that isn’t yet materialized. Once you access the property, the value is materialized. This on-demand loading works only as long as the database transaction is open, so you need to access any property of such a type when the owning entity instance is in a persistent and transactional state, not in detached state. Your domain model is now also bound to JDBC, because the import of the java.sql package is required. Although domain model classes are executable in isolated unit tests, you can’t access LOB properties without a database connection.

Hope this helps.