[jira] [Commented] (HBASE-15233) Bytes.toBytes() methods should allow arrays to be re-used

[Jean-Marc Spaggiari (JIRA)]

    [ 
https://issues.apache.org/jira/browse/HBASE-15233?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15281486#comment-15281486
 ] 

Jean-Marc Spaggiari commented on HBASE-15233:
---------------------------------------------

So. I search a bit for some places where this can be used and found many. I can 
not list them all here. I recommend to open another JIRA for that.

Here is one good example. In MetaTableAccessor.

{code}
  public static Put addLocation(final Put p, final ServerName sn, long 
openSeqNum, int replicaId){
    // using regionserver's local time as the timestamp of Put.
    // See: HBASE-11536
    long now = EnvironmentEdgeManager.currentTime();
    p.addImmutable(HConstants.CATALOG_FAMILY, getServerColumn(replicaId), now,
      Bytes.toBytes(sn.getHostAndPort()));
    p.addImmutable(HConstants.CATALOG_FAMILY, getStartCodeColumn(replicaId), 
now,
      Bytes.toBytes(sn.getStartcode()));
    p.addImmutable(HConstants.CATALOG_FAMILY, getSeqNumColumn(replicaId), now,
      Bytes.toBytes(openSeqNum));
    return p;
  }
{code}

Can become
{code}
  public static Put addLocation(final Put p, final ServerName sn, long 
openSeqNum, int replicaId){
  byte longBytes = new byte[Bytes.SIZEOF_LONG];
    // using regionserver's local time as the timestamp of Put.
    // See: HBASE-11536
    long now = EnvironmentEdgeManager.currentTime();
    p.addImmutable(HConstants.CATALOG_FAMILY, getServerColumn(replicaId), now,
      Bytes.toBytes(sn.getHostAndPort());
    p.addImmutable(HConstants.CATALOG_FAMILY, getStartCodeColumn(replicaId), 
now,
      Bytes.toBytes(sn.getStartcode(), longBytes));
    p.addImmutable(HConstants.CATALOG_FAMILY, getSeqNumColumn(replicaId), now,
      Bytes.toBytes(openSeqNum, longBytes));
    return p;
  }
{code}

It's not a HUGE saving, it just divides the number of arrays by 2.


Another example in CellCodec
{code}
    public void write(Cell cell) throws IOException {
      checkFlushed();
      // Row
      write(cell.getRowArray(), cell.getRowOffset(), cell.getRowLength());
      // Column family
      write(cell.getFamilyArray(), cell.getFamilyOffset(), 
cell.getFamilyLength());
      // Qualifier
      write(cell.getQualifierArray(), cell.getQualifierOffset(), 
cell.getQualifierLength());
      // Version
      this.out.write(Bytes.toBytes(cell.getTimestamp()));
      // Type
      this.out.write(cell.getTypeByte());
      // Value
      write(cell.getValueArray(), cell.getValueOffset(), cell.getValueLength());
      // MvccVersion
      this.out.write(Bytes.toBytes(cell.getSequenceId()));
    }
{code}

can be changed to 
{code}
    public void write(Cell cell) throws IOException {
      byte longBytes = new byte[Bytes.SIZEOF_LONG];
      checkFlushed();
      // Row
      write(cell.getRowArray(), cell.getRowOffset(), cell.getRowLength());
      // Column family
      write(cell.getFamilyArray(), cell.getFamilyOffset(), 
cell.getFamilyLength());
      // Qualifier
      write(cell.getQualifierArray(), cell.getQualifierOffset(), 
cell.getQualifierLength());
      // Version
      this.out.write(Bytes.toBytes(cell.getTimestamp(), longBytes));
      // Type
      this.out.write(cell.getTypeByte());
      // Value
      write(cell.getValueArray(), cell.getValueOffset(), cell.getValueLength());
      // MvccVersion
      this.out.write(Bytes.toBytes(cell.getSequenceId(), longBytes));
    }
{code}

Here again, not a huge benefit, but still divides the number of created arrays 
by 2...

So if we add them all, at the end, might still help.

> Bytes.toBytes() methods should allow arrays to be re-used 
> ----------------------------------------------------------
>
>                 Key: HBASE-15233
>                 URL: https://issues.apache.org/jira/browse/HBASE-15233
>             Project: HBase
>          Issue Type: Improvement
>          Components: API
>    Affects Versions: 1.1.3
>            Reporter: Jean-Marc Spaggiari
>            Assignee: Michael Ernest
>            Priority: Minor
>              Labels: beginner
>
> Today we have this:
> {code}
>   public static byte[] toBytes(long val) {
>     byte [] b = new byte[8];
>     for (int i = 7; i > 0; i--) {
>       b[i] = (byte) val;
>       val >>>= 8;
>     }
>     b[0] = (byte) val;
>     return b;
>   }
> {code}
> might be nice to also have this:
> {code}
>   public static byte[] toBytes(long val, byte[] reuse) {
>     for (int i = 7; i > 0; i--) {
>       reuse[i] = (byte) val;
>       val >>>= 8;
>     }
>     reuse[0] = (byte) val;
>     return reuse;
>   }
> {code}
> Same for all the other Bytes.toBytes() methods.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)