[HBASE-11541] Wrong result when scaning meta with startRow

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.99.0, 0.98.5
• Hadoop Flags: Reviewed
• Assignee: Shaohui Liu
• Reporter: Shaohui Liu
• Created: 18/Jul/14 05:15
• Updated: 06/Apr/18 17:51
• Resolved: 21/Jul/14 22:19

Description

When we scan the meta with STARTROW optiion, wrong result may be returned.
For exmaple: if there are two tables named "a" and "b" in hbase, when we scan the meta with startrow = 'b', the region location of table "a" will be returned but we expect to get the region location of table "b".

> create 'a', {NAME => 'f'}
> create 'b', {NAME => 'f'}
> scan '.META.', {STARTROW => 'b', LIMIT => 1}
a,,1405655897758.f8b547476b6dc80545e6413c31396, xxxx

The reason is the wrong assumption in MetaKeyComparator

See: KeyValue.java#2011

      int leftDelimiter = getDelimiter(left, loffset, llength,
          HRegionInfo.DELIMITER);
      int rightDelimiter = getDelimiter(right, roffset, rlength,
          HRegionInfo.DELIMITER);
      if (leftDelimiter < 0 && rightDelimiter >= 0) {
        // Nothing between .META. and regionid.  Its first key. 
        return -1;
      } else if (rightDelimiter < 0 && leftDelimiter >= 0) {
        return 1;
      } else if (leftDelimiter < 0 && rightDelimiter < 0) {
        return 0;
      }

It's a little troublesome to fix this problem for given a start row which contains more than two "," for meta, it's not easy to extract the startKey of region.
eg: STARTROW => 'aaa,bbb,ccc,xxx'.

Comments and suggestions are welcomed. Thanks