Hyunsik Choi created TAJO-95:
--------------------------------

             Summary: Eliminate the laze copy approach from the classes 
wrapping protobuf-generated classes.
                 Key: TAJO-95
                 URL: https://issues.apache.org/jira/browse/TAJO-95
             Project: Tajo
          Issue Type: Improvement
            Reporter: Hyunsik Choi
            Assignee: Hyunsik Choi


We have used two (de)serialization methods, such as JSON and Protocol Buffer. 
In a (de) serializable object, each attribute can be placed on the  
corresponding member variable or the member variable of the protocol buffer 
object.

For reducing the frequency of data copies, we have used the laze copy approach 
that delays copying values from the contents of PB object to member variables 
or vice versa. It is similar to copy-on-wrote (COW).

In many cases, however, this mechanism is actually burden rather than the 
performance benefit. Especially, this problem becomes more complicated if the 
wrapper class can be (de)serialized via both JSON and protocol buffer.

{code:title=A wrapper class}
  protected TableDescProto proto = TableDescProto.getDefaultInstance();
  protected TableDescProto.Builder builder = null;
  protected boolean viaProto = false;
  
  @Expose protected String tableId;
  @Expose protected Path uri;
  @Expose protected TableMeta meta;
   
  .
  .

  public String getId() { // <- even a simple getter becomes very complicated.
    TableDescProtoOrBuilder p = viaProto ? proto : builder;
    
    if (tableId != null) {
      return this.tableId;
    }
    if (!p.hasId()) {
      return null;
    }
    this.tableId = p.getId();
    
    return this.tableId;
  }
{code}

Therefore, I would like to propose the simplification of wrapper classes by 
eliminating the laze copy approach. My proposal is as follows:
 * A wrapper class only keeps values in member variables when it is created.
 * A wrapper class only builds the corresponding protobuf object when 
getProto() is called.

I think that the performance benefit has been very low. We should remove the 
lazy copy approach for more robust system.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to