[HBASE-12068] [Branch-1] Avoid need to always do KeyValueUtil#ensureKeyValue for Filter transformCell

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.99.0
• Fix Version/s: 0.99.1
• Component/s: Filters
• Hadoop Flags: Reviewed

Release Note

When custom Filters are used, make sure to override Cell based methods and not keyValue based methods. ie.
transformCell(Cell) instead of transform(KeyValue)
getNextCellHint(Cell) instead of getNextKeyHint(KeyValue)
Else there can be possible performance hit of a conversion of a non KV Cell to a KV which required key and value deep copy.

• Assignee: Anoop Sam John
• Reporter: Anoop Sam John
• Created: 23/Sep/14 18:01
• Updated: 21/Feb/15 23:41
• Resolved: 24/Sep/14 07:54

Description

During read with Filters added to Scan/Get, the core code calls transformCell(Cell) on the Filter. Most of the filters do not implement transform API so the method from FilterBase will get executed

  @Override
  public Cell transformCell(Cell v) throws IOException {
    // Old filters based off of this class will override KeyValue transform(KeyValue).
    // Thus to maintain compatibility we need to call the old version.
    return transform(KeyValueUtil.ensureKeyValue(v));
  }

Here always it do KeyValueUtil.ensureKeyValue. When a non KV cell comes in, we need recreate KV and do deep copy of key and value!
We have to stick with this model in branch-1 for BC.
So as a workaround to avoid possible KV convert, we can implement transformCell(Cell) method in all of our individual Filter classes which just return the incoming cell (So that method from FilterBase wont get executed)

Issue Links

• is related to: HBASE-12048 Remove deprecated APIs from Filter